diff --git a/docs/action-aware-constraint-pilot-v0-protocol-20260714.md b/docs/action-aware-constraint-pilot-v0-protocol-20260714.md new file mode 100644 index 0000000..062685d --- /dev/null +++ b/docs/action-aware-constraint-pilot-v0-protocol-20260714.md @@ -0,0 +1,179 @@ +# Action-aware constraint pilot v0 protocol + +Status: **FROZEN BEFORE NEW GPU RUNS**. + +Date: 2026-07-14 (Asia/Singapore). + +## Headline question + +Can telemetry from one complete initial-config benchmark identify which of two +competing knob families should be changed, before either target configuration +is evaluated? + +This pilot tests a narrow prerequisite, not an end-to-end tuner claim. It +uses fields already present in the per-step OpProf stream to reconstruct exact +zero-slack conditions for `max_num_seqs` (MNS) and +`max_num_batched_tokens` (MBBT). No new vLLM instrumentation is justified +unless those action-conditioned conditions predict crossed real-system +intervention responses. + +## Hypothesis + +I believe config-normalized scheduler constraints provide a stronger tuning +signal than an aggregate queue symptom because the same waiting queue can be +blocked by different admission limits. + +I will verify it by holding model, hardware, TP, request bands, arrival times, +and offered load fixed while constructing two source configurations with +different binding constraints. From each source run alone, the larger +exclusive binding fraction predicts the action family. Both candidate +actions are then measured on the same requests for the full 300-second replay. + +## Frozen platform and workload + +- Host: `dash0`, solo placement on GPUs 0-3, four NVIDIA H20 GPUs. +- Model: Qwen3-30B-A3B BF16. +- Engine: patched vLLM `0.24.1.dev3+opprof`, TP=4. +- Workload: the three disjoint `mid` bands from + `chat_w20260312_1000`, 2.125 requests/s/GPU, 300-second arrival window, + exactly 128 output tokens. +- SLO: the unchanged study TTFT/TPOT thresholds and 0.95 pass-rate target. +- Every config starts one fresh server, performs the accepted 16-request + warm-up and the existing burn-in, then runs all three disjoint measured + bands in its frozen order. +- SLO early stopping is disabled. A measured run must drain all selected + requests and finish within the 450-second client deadline. + +## Frozen configuration and action matrix + +| ID | MNS | MBBT | Role | +|---|---:|---:|---| +| `b_base` | 64 | 256 | token-budget-bound source; operational gate runs first | +| `a_base` | 16 | 8192 | MNS-bound source | +| `shared` | 64 | 8192 | MNS action from A; MBBT action from B | +| `b_mns` | 128 | 256 | competing MNS action from B | +| `a_mbbt` | 16 | 16384 | competing MBBT action from A | + +The two decisions are therefore: + +```text +Regime A: a_base -> {shared (increase MNS), a_mbbt (increase MBBT)} +Regime B: b_base -> {b_mns (increase MNS), shared (increase MBBT)} +``` + +The candidate magnitudes are intentionally large in this feasibility pilot so +that a missing crossed response is not explained by an imperceptibly small +intervention. This does not establish that these are production step sizes. + +Frozen config order is `b_base`, `a_base`, `shared`, `b_mns`, `a_mbbt`. +Frozen repetition orders are respectively `123`, `231`, `312`, `132`, and +`213`, reducing band/time alignment without reusing a server across configs. + +## Pre-action signal + +For each source run, let `waiting` include the normal and deferred waiting +queues, and let `scheduled_tokens = prefill_tokens + decode_tokens`. + +```text +mns_exclusive = waiting > 0 + and running == configured MNS + and scheduled_tokens < configured MBBT + +mbbt_exclusive = waiting > 0 + and scheduled_tokens == configured MBBT + and running < configured MNS + +both = waiting > 0 + and running == configured MNS + and scheduled_tokens == configured MBBT +``` + +Each score is the fraction of all scheduler records in the measured interval +that satisfies the condition. The predicted action is the family with the +larger exclusive fraction. This uses no target telemetry or target outcome. +KV usage and preemptions are reported as possible alternative constraints but +are not silently reassigned to either score. + +These conditions reproduce two scheduler loop boundaries, but they are still +a Level-0 proxy: they do not expose the exact request rejected at the boundary +or run a shadow schedule. The pilot explicitly tests whether that additional +engine patch is warranted. + +## Outcomes and baselines + +Primary intervention outcome: + +```text +SLO-goodput = full-run SLO pass count / 300-second arrival window +``` + +Also report pass rate, TTFT p50/p95/p99, TPOT p50/p95/p99, drain elapsed time, +KV usage, preemptions, queue area, and CUDA-graph padding. + +Required decision baselines: + +1. always choose the MNS family; +2. always choose the MBBT family; +3. queue-pressure-only, which has no candidate-specific score and therefore + must use one frozen family for both regimes; +4. the pre-action exclusive-binding prediction. + +This is a mechanism ablation. It does not compare against a trained black-box +tuner because two regimes are not a valid training surface. + +## Gates and failure meanings + +Data validity requires 15 uncensored measured runs, exact request/arrival/input +hashes across each repetition, full request accounting, one continuous OpProf +stream per config, zero dropped records, monotonic timestamps and step indices, +nonnegative counters, bounded ratios, clean GPU placement, and config values in +the result matching the server command. + +The crossed-response gate passes only if, in all three repetitions: + +- the MNS target has higher SLO-goodput than the MBBT target in Regime A; +- the MBBT target has higher SLO-goodput than the MNS target in Regime B; +- each winning target exceeds its competing target by at least 10% of the + source SLO-goodput. A source with zero goodput makes the run invalid for + this relative gate rather than changing the denominator. + +The binding gate passes only if, in both regimes: + +- the predicted family matches the measured winning family in all three + repetitions; +- the median winning-family exclusive fraction is at least 0.10; +- it is at least 5x the median competing-family exclusive fraction; +- the direction is unchanged under cumulative 25%, 50%, 75%, and 100% + checkpoints after the 25% checkpoint. + +Decision meanings: + +- `STOP_WORKLOAD_NOT_CROSSED`: candidate outcomes do not have different + winners; the experiment cannot test action selection. +- `STOP_BINDING_NOT_PREDICTIVE`: outcomes cross but source-only constraint + scores do not select them; do not implement shadow scheduling from this + hypothesis. +- `STOP_NO_NEW_INSTRUMENTATION_NEEDED`: the signal works but every required + field was already present; keep it as an analysis/tuner feature and do not + claim a new engine-instrumentation contribution. +- `OPEN_EXACT_ATTRIBUTION_ABLATION`: the signal works but unresolved/both/KV + cases are material enough that exact rejection reasons could change a + decision. Only this result authorizes a minimal vLLM attribution patch. + +Ambiguity is material only when, in either regime, the median +`both + waiting_unresolved` fraction is at least the median absolute gap +between the two exclusive fractions, or when any source run records a +preemption or median source KV maximum is at least 0.90. Otherwise all fields +needed for the observed decision were already present and the result is +`STOP_NO_NEW_INSTRUMENTATION_NEEDED`. + +No result from this development pilot is a paper-level E2E tuning claim. + +## Cost and stopping discipline + +- Hard cap: 8.0 H20-hours, including failed sessions. +- Expected: 6.0-7.2 H20-hours and 90-110 minutes wall time. +- `b_base` runs first. If its first measured band cannot drain by 450 seconds, + the controller stops before any comparative analysis; MBBT=256 is then an + operationally invalid source, not negative evidence. +- Any data red flag stops analysis before computing a tuning conclusion. diff --git a/docs/action-aware-constraint-pilot-v1-protocol-20260714.md b/docs/action-aware-constraint-pilot-v1-protocol-20260714.md new file mode 100644 index 0000000..4937903 --- /dev/null +++ b/docs/action-aware-constraint-pilot-v1-protocol-20260714.md @@ -0,0 +1,44 @@ +# Action-aware constraint pilot v1 amendment + +Status: **FROZEN AFTER V0 OPERATIONAL STOP AND BEFORE V1 GPU RUNS**. + +Date: 2026-07-14 (Asia/Singapore). + +The complete claim, workload, baselines, metrics, action matrix, analysis gates, +and data-validity requirements remain those in the v0 protocol. This amendment +changes only the token-bound source severity and adds an operational burn-in +gate. + +## Why v0 produced no comparative evidence + +The first v0 session used MNS=64 and MBBT=256. During the 510-request, +60-second burn-in, the client had run for 197 seconds and the engine still held +64 requests: 13 running and 51 waiting. The last step scheduled exactly 256 +tokens, KV usage was 0.01151, and there were zero preemptions. No measured run +or target configuration had started. + +The session was stopped and cleanly released all GPUs after consuming +0.3859868995 H20-hours. This is evidence that MBBT=256 is a real token-budget +bottleneck, but it is not an admissible source for the 2.125 requests/s/GPU +comparison because it cannot sustain the offered load. V0 contributes no +tuning label and none of its runtime data is reused by V1. + +Authoritative failure artifact: +`/home/admin/cpfs/wjh/action-aware-constraint-v0-20260714/operational-stop-v0.json`. + +## V1 changes + +- `b_base`: MNS=64, MBBT **2048** instead of 256. +- `b_mns`: MNS=128, MBBT **2048** instead of 256. +- The B-family MBBT action is therefore 2048 -> 8192. +- All five configurations and all three repetitions run fresh under a new V1 + run root. +- Before any measured run, every config's 510-request/60-second burn-in must + drain in at most **90 seconds**. A slower config is an operational failure; + the controller stops before comparative analysis. +- V1 incremental hard cap is 7.6140131005 H20-hours so that V0 plus V1 remains + within the original global 8.0 H20-hour cap. + +The V0 protocol's crossed-response and source-only binding gates are unchanged. +In particular, V1 still requires three-of-three action-family predictions in +both regimes and a different real winner in Regime A versus Regime B. diff --git a/docs/action-aware-constraint-pilot-v2-protocol-20260714.md b/docs/action-aware-constraint-pilot-v2-protocol-20260714.md new file mode 100644 index 0000000..d51ca29 --- /dev/null +++ b/docs/action-aware-constraint-pilot-v2-protocol-20260714.md @@ -0,0 +1,35 @@ +# Action-aware constraint pilot v2 amendment + +Status: **FROZEN AFTER V1 CONTROLLER STOP AND BEFORE V2 GPU RUNS**. + +Date: 2026-07-14 (Asia/Singapore). + +The V1 configuration matrix and every scientific gate remain unchanged. V2 +fixes one controller bug and reruns every configuration and repetition fresh. + +## V1 controller failure + +The MNS64/MBBT2048 burn-in completed all 510 requests in 61.259 seconds, below +the frozen 90-second operational limit. However, the controller accidentally +assigned the preceding 16-request warm-up result to `burnin_result`; its state +therefore recorded 4.376 seconds and evaluated the wrong object. + +The first measured replay was terminated after 75 seconds, before it produced +a result. No target configuration had started. V1 consumed +0.3184109431 H20-hours and contributes no action label or telemetry to V2. + +## V2 correction and regression gate + +- The completed burn-in `run_client()` return value is assigned to + `burnin_result`. +- A dedicated `burnin_gate()` rejects any non-anchor object, any request count + other than 510, and elapsed time above 90 seconds. +- Unit tests explicitly pass a warm-up object and require rejection, then test + accepted and over-limit burn-ins. +- All five configs and 15 measured runs use a new run root; no V0/V1 runtime + artifact is reused. + +V0 and V1 together consumed 0.7043978426 H20-hours. V2's incremental hard cap +is therefore 7.2956021574 H20-hours, preserving the original global 8.0 +H20-hour cap. The authoritative accounting file is +`runs/action-aware-v0/prior-attempts-v2.json`. diff --git a/docs/action-aware-constraint-pilot-v2-results-20260714.md b/docs/action-aware-constraint-pilot-v2-results-20260714.md new file mode 100644 index 0000000..b0f1ed0 --- /dev/null +++ b/docs/action-aware-constraint-pilot-v2-results-20260714.md @@ -0,0 +1,260 @@ +# Action-aware constraint pilot v2 results + +Date: 2026-07-14 (Asia/Singapore). + +Decision: **`STOP_WORKLOAD_NOT_CROSSED`**. + +The pilot produced one valid positive regime and one invalid-for-effect-size +regime. It supports continuing a narrower action-response investigation, but +it does not justify an end-to-end telemetry-guided tuner claim or new engine +instrumentation yet. + +## Question tested + +Given only a completed source run, can existing engine telemetry distinguish +which of two one-knob interventions will improve SLO-goodput more? + +The frozen score counted scheduler steps with backlog where either MNS or MBBT +was exclusively at its configured limit. It made two pre-intervention +predictions on the same workload and offered load: + +- Regime A: source `(MNS=16, MBBT=8192)` predicts increasing MNS to 64 over + increasing MBBT to 16384. +- Regime B: source `(MNS=64, MBBT=2048)` predicts increasing MBBT to 8192 over + increasing MNS to 128. + +The primary outcome was 300-second SLO-goodput. A predicted action had to beat +the alternative on every paired request band by at least 10% of that band's +source goodput. Telemetry direction also had to remain stable at 25%, 50%, +75%, and 100% of the replay. + +## Setup + +- Host: `dash0`, GPU 0-3 used exclusively; four NVIDIA H20 GPUs; TP=4. GPU + 4-7 remained idle to avoid co-location effects. +- Model: Qwen3-30B-A3B BF16 at + `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`. +- Runtime: patched vLLM `0.24.1.dev3+g668cfb7e2`, source commit `4b253fd`, with + OpProf Layer-1 telemetry. +- Workload: `chat_w20260312_1000`, 2.125 requests/s/GPU, 300-second arrival + window, 128 output tokens, three disjoint request bands. +- Five fresh-server configurations, three measured runs each, counter-rotated + repetition order, 16-request warm-up, and a 510-request/60-second burn-in. + SLO early stopping was disabled. +- Exact request-id, arrival-order, and input-length hashes matched for every + paired comparison. + +The authoritative run root is +`/home/admin/cpfs/wjh/action-aware-constraint-v2-20260714`. The final audit is +`pilot-audit-final.json`, SHA256 +`7ebe080fcc4970bef423bc587253d157e75aed1ea8b410bd37770c17708135ab`. + +## End-to-end result + +### Regime A: strong MNS constraint + +| Rep | Source 16/8192 | MBBT action 16/16384 | MNS action 64/8192 | MNS-only source steps | MBBT-only source steps | `(MNS action - MBBT action) / source` | +|---:|---:|---:|---:|---:|---:|---:| +| 1 | 4.710 | 7.687 | 8.500 | 79.85% | 0.038% | 17.27% | +| 2 | 2.803 | 4.150 | 8.500 | 94.61% | 0.005% | 155.17% | +| 3 | 2.227 | 3.943 | 8.500 | 79.87% | 0.005% | 204.64% | + +Units are SLO-goodput requests/s except the percentage columns. The source +prediction was stable at every phase checkpoint and correct in all three +paired bands. The predicted MNS action cleared the frozen 10% material-margin +gate in all three bands. + +This is a real but limited positive result. The source was an extreme case: +MNS was full on nearly every scheduler step that retained backlog, TTFT p50 was +1.64-5.61 seconds, KV usage remained below 2.45%, and no preemption occurred. +An expert or a simple rule could identify this case without a learned tuner. + +### Regime B: MBBT direction at an outcome ceiling + +| Rep | Source 64/2048 | MNS action 128/2048 | MBBT action 64/8192 | MBBT-only source steps | `(MBBT action - MNS action) / source` | +|---:|---:|---:|---:|---:| +| 1 | 8.423 | 8.420 | 8.500 | 10.90% | 0.950% | +| 2 | 8.447 | 8.500 | 8.500 | 12.03% | 0% | +| 3 | 8.500 | 8.497 | 8.500 | 8.79% | 0.039% | + +The telemetry direction was stable in all phase checkpoints. The predicted +action won twice and tied once, while the wrong MNS action left the MBBT-only +state intact. However, the source already delivered 99.10-100% of the offered +8.5 requests/s. Even a perfect action could not reach the preregistered 10% +margin. Regime B therefore does not test material weak-signal value; it is a +workload-selection failure, not evidence that telemetry does or does not help +near a decision boundary. + +The missing preflight condition is mathematical. For an effect threshold +`delta` and offered goodput ceiling `G`, the source must satisfy +`source <= G / (1 + delta)`. Here `G=8.5` and `delta=0.10`, so any source above +7.727 requests/s cannot possibly pass before either target is measured. + +## Why the exclusive-limit rule is incomplete + +The alternative MBBT action improved Regime A by 48.0%, 63.2%, and 77.1% over +the source even though MBBT was almost never the exclusive backlog constraint. +This rules out the binary interpretation "a knob that is not exclusively at +its cap cannot help." + +Existing richer telemetry provides a plausible mechanism: + +| Rep | Split-prefill requests, source -> MBBT action | Prefill steps, source -> MBBT action | Prefill requests/step, source -> MBBT action | Prefix-hit rate, source -> MBBT action | +|---:|---:|---:|---:|---:| +| 1 | 41 -> 2 | 2324 -> 2022 | 1.071 -> 1.249 | 13.851% -> 13.747% | +| 2 | 7 -> 0 | 2410 -> 2291 | 1.012 -> 1.076 | 13.078% -> 12.988% | +| 3 | 12 -> 1 | 2377 -> 2270 | 1.018 -> 1.096 | 13.613% -> 13.604% | + +Increasing MBBT allows more prefill work to be packed into one iteration and +nearly eliminates split prefills. Under MNS=16, this can reduce the number of +iterations for which long prompts occupy scarce running slots. Prefix-cache +hit rates differ by at most 0.104 percentage points, and exact workload hashes +match, so neither explains the gain. Step-duration p99 also remains similar; +one 1.127-second decode-step outlier appears in `a_mbbt/rep2`, but the same +action direction occurs in all three bands. + +This is a mechanism-consistent explanation, not a completed causal +decomposition. MBBT simultaneously changes total per-iteration token budget, +per-request chunk size, and multi-request packing. Instrumentation observes +their joint response but cannot separate those effects without another +intervention. + +## Instrumentation decision + +Do **not** add a new engine patch for this mechanism yet. The existing OpProf +stream already records submit/complete timestamps, prefill/decode composition, +chunked-prefill categories, prefix hits, queues, KV usage, and CUDA graph mode. +Those fields are sufficient to identify the interaction missed by the initial +exclusive-limit rule. + +The next narrow mechanism ablation is available in the current vLLM runtime: + +1. `(MNS=16, MBBT=8192, long-prefill-threshold=0)` is the current source. +2. `(MNS=16, MBBT=16384, long-prefill-threshold=8192)` keeps individual long + chunks at 8192 while increasing total packing budget. +3. `(MNS=16, MBBT=16384, long-prefill-threshold=0)` is the current MBBT action. + +The runtime exposes `--long-prefill-token-threshold`; with a threshold of 8192, +the second arm separates total packing headroom from the larger per-request +chunk allowed by the third arm. A formal test must rerun all three arms fresh +with counter-rotated order rather than reuse today's endpoints. + +## Correct tuning-research route + +The pilot does not support turning the frozen equality checks into a larger +rule tree. The supported route is **intervention-calibrated, action-conditioned +system identification**: + +```text +source event sequence + normalized config delta + -> predicted distribution of Delta SLO-goodput and evaluation cost + -> uncertainty-aware next-config selection +``` + +The policy input should retain continuous distributions and phase evolution: +queue/running residency, MNS and token slack, prefill/decode composition, +partial-prefill occupancy, step time, KV state, and graph behavior. Human +bottleneck labels and hand-authored `if queue then increase MNS` mappings are +not policy inputs. Mechanism summaries remain audit and interpretation tools; +the action response is learned from paired real interventions. + +The harness has a narrower, non-heuristic role: + +- define legal configurations and exact paired workloads; +- reject source points without outcome headroom before a full sweep; +- randomize/counter-rotate execution order and preserve failures/cost; +- validate stream coverage, hashes, request accounting, and censoring; +- expose target outcomes only after a source-only prediction is frozen; +- evaluate fixed-budget regret and H20-hours, not explanation quality alone. + +The next tuning experiment should use non-extreme, non-ceiling source points +and a local two-dimensional MNS/MBBT neighborhood. A short run may screen load +only; every inferential telemetry and outcome result remains a 300-second run. +At least one held-out workload must be reserved before choosing features or +thresholds. + +Primary evaluation is H20-hours/trials to reach 95% of the real local oracle +and cost-normalized regret AUC. Required baselines are random search, +config/outcome-only sequential search, the current rule heuristic, and the +same action-response model with telemetry removed. Action-ranking accuracy is +supporting evidence only. + +## What this pilot establishes and does not establish + +Established: + +- Long-window engine state can make a correct, phase-stable action-family + prediction in an extreme MNS-constrained regime. +- A naive `queue > 0 -> increase MNS` rule would choose an ineffective action + in Regime B; action-conditioned state distinguishes the mechanism direction, + although the measured effect is immaterial at the selected load. +- Binary exclusive-cap attribution misses a substantial MNS/MBBT interaction; + existing chunk/step telemetry reveals a plausible explanation. +- Source outcome headroom must be an explicit experiment admission gate. + +Not established: + +- telemetry improves an end-to-end tuner over an outcome-only baseline; +- weak or mixed constraints can be ranked with material gain; +- the response transfers across workloads, models, TP, or hardware; +- new engine instrumentation is necessary; +- the chunking/packing breakdown is causal rather than mechanism-consistent. + +## Change and verification + +Reproduction: + +```bash +python3 runs/action-aware-v0/test_pilot.py + +python3 runs/action-aware-v0/analyze_pilot.py \ + --run-root /home/admin/cpfs/wjh/action-aware-constraint-v2-20260714/runs/pilot \ + --manifest runs/action-aware-v0/pilot-manifest-v2.json \ + --output /home/admin/cpfs/wjh/action-aware-constraint-v2-20260714/pilot-audit-final.json +``` + +The fresh GPU run used AITuner commit `c5ab073`; asynchronous coverage was +corrected in `3facb18`; reproducible mechanism summaries were added in +`2af22db`. The raw run is unchanged across those analyzer-only commits. + +Change: added a crossed real-intervention controller and audit, fixed the +burn-in result gate, corrected asynchronous per-step coverage accounting, and +added reproducible step/chunk/prefix mechanism summaries. + +Expected effect: distinguish descriptive telemetry from source-only action +predictions that survive paired real interventions. + +Verification: local and remote action-aware test suites pass; all five sessions +completed; all stream/footer and request-accounting invariants pass; the final +analyzer was run twice and produced byte-identical output. + +Result: Regime A passes; Regime B is invalid for the frozen effect-size test; +the global decision is `STOP_WORKLOAD_NOT_CROSSED`. + +Remaining risk: one model, one TP, one trace family, three bands, two action +families, and deliberately constructed endpoints are development evidence +only. The strong positive regime is too obvious to support a paper claim. + +## Data sanity + +- Measured runs: n=15; elapsed 300.610-317.350 seconds; 15 distinct. Pass + rate min/max 0.2620/1.0 with 11 distinct values; SLO-goodput min/max + 2.2267/8.5 requests/s with 11 distinct values. +- Telemetry intervals: n=15; records min/max 13,621/23,711; 15 distinct. + Start gaps min/max 0.0412/0.1227 seconds; end gaps 0.00053/0.0649; uncovered + internal gaps 0/0.3231. One submit gap reached 1.1190 seconds but was fully + covered by a 1.1269-second recorded execution; contiguous indices and zero + drops were preserved. +- Sessions: n=5; cost min/max 1.1691/1.2730 H20-hours; 5 distinct. V2 cost + was 6.0862 H20-hours; V0/V1/V2 total was 6.7906, below the 8.0 cap. +- Regime-A split-prefill observations: n=6; min/max 0/41 requests; 6 distinct. + Prefix-hit rates: n=6; min/max 0.12988/0.13851; 6 distinct. +- Checked invariants: non-negative counters and durations; ratios in `[0,1]`; + exact request, arrival, and length hashes; 2550/2550 request accounting per + measured run; uncensored outcomes; outcomes across configurations not all + identical; + five complete streams; monotonic timestamps; contiguous step indices; zero + drops; footer/sidecar agreement; chunk-token accounting; bounded prefix hits; + no OOM, controller error, or residual GPU allocation. No unresolved red + flag remains. The three identical shared goodputs are reported as the + offered-load ceiling, not treated as independent performance variation. diff --git a/docs/active-intervention-v0-protocol-20260715.md b/docs/active-intervention-v0-protocol-20260715.md new file mode 100644 index 0000000..6d68231 --- /dev/null +++ b/docs/active-intervention-v0-protocol-20260715.md @@ -0,0 +1,136 @@ +# Active intervention + measurement v0 protocol + +Date: 2026-07-15 (Asia/Singapore) + +Status: **FROZEN BEFORE THE `chat_w20260313_1000` GPU RUN**. + +## Research question + +This experiment asks whether a tuner conditioned on direct engine-state +trajectories can choose both a measurement horizon and a coupled configuration +intervention with lower real-GPU cost than the same tuner using only external +prefix outcomes. + +The contribution is not the controller, legality checks, telemetry collection, +or the ridge model. The route remains open only if engine state changes an +actual decision and reduces cost-to-near-oracle on unseen workloads. + +## Development result that motivates, but does not pass, the route + +The frozen trace-12 dataset contains 72 examples: six source decisions, four +measurement checkpoints, and `noop/MNS/MBBT` actions. Features are direct +continuous Layer-1 state summaries; cap-exclusive and bottleneck labels are +excluded. Leave-one-repetition-out sequential replay uses the same model, +candidate set, confidence rule, and checkpoint set for both modes. + +The external-outcome policy and telemetry policy both put all six decisions +within 2% regret. Outcome-only selected a mean 262.5-second source measurement +and cost 3.750 replay H20-hours across the six replayed decisions; telemetry +selected 275 seconds and cost 3.833 H20-hours. Telemetry therefore increased +the replay lower-bound cost by 2.22%, with no regret reduction. This is a +negative result. It does not settle the question because the dataset has only +two source regimes, one source is at the offered ceiling, and there is no joint +MNS+MBBT action. + +Sanity: n=6 decisions; regret min=0, max=0.009412, distinct=3; source cutoff +min=150s, max=300s, distinct=3 across the two policies; all costs are +non-negative, regrets are in `[0,1]`, target results are not all identical, and +the six decisions are complete exact-workload pairs. + +## Frozen prospective setup + +- Host: `dash0`, 8 NVIDIA H20 GPUs available; each TP4 server runs alone on + GPUs 0-3. Co-location is prohibited for SLO verdicts. +- Engine: patched vLLM `0.24.1.dev3+g668cfb7e2` from clean source commit + `4b253fd8619764b6971a7f2e3a3aa7545f6ace05` at + `/home/admin/cpfs/wjh/opprof-phase2-dash0-20260711/vllm-v0.24.0`, using + `/tmp/wjh-opprof-phase2-dash0-20260711/.venv`. +- Model: `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`, BF16. +- Workload: unseen `chat_w20260313_1000`; input 0-8192; output exactly 128; + replay scale 0.5; 300-second arrival window. +- Three disjoint repetitions: source rows are assigned by a deterministic + SHA-256 modulo-3 partition before input filtering. Each repetition selects + approximately 3300 requests, or 2.75 requests/s/GPU at TP4. +- SLO: at least 95% pass; stepped TTFT 2/4/6 seconds; TPOT at most 50 ms. +- Checkpoints: 75, 150, 225, and 300 seconds. +- Full 2x2 surface: + - source: `MNS=32, MBBT=4096`; + - MNS action: `64,4096`; + - MBBT action: `32,8192`; + - joint action: `64,8192`; + - `noop` retains the source. +- Four config sessions are serialized. Each session uses a fresh server, + warm-up, burn-in, and counter-rotated repetition order. +- Expected campaign cost: 4.6-5.5 H20-hours; hard cap: 6.0 H20-hours; + expected wall time: 75-100 minutes. + +The source is executed first. The frozen telemetry policy selects the next +real config session; all remaining cells are then measured only to construct +the exact finite-surface oracle. Oracle annotation after the selected action +is reported separately from tuner cost. + +## Frozen policies + +Both policies fit the paired treatment effect + +```text +target normalized SLO-goodput - source normalized SLO-goodput +``` + +from source config, full config delta, offered load, and external prefix +outcomes. The telemetry policy additionally receives fixed direct Layer-1 +summaries and their interactions with `delta_log2(MNS)` and +`delta_log2(MBBT)`. It does not receive a bottleneck label or a +diagnosis-to-knob rule. + +At each checkpoint, jackknife models produce an effect distribution for +`noop`, MNS, MBBT, and joint actions. Measurement stops at the earliest second +consecutive checkpoint with the same confident best action; otherwise it uses +the full 300 seconds. Confidence requires a predicted margin of at least 0.02 +and the best lower bound to exceed the second-best upper bound. If the final +choice is not confident, the next run is the positive-UCB action, explicitly +marked as a diagnostic intervention. The exact same rule is used for the +outcome-only baseline. + +## Hypotheses and gates + +### H1: action value + +Engine state must change the selected intervention or its ranking and reduce +real action regret. Prediction error or bottleneck-label accuracy is not a +success metric. + +### H2: measurement value + +Engine state must select a shorter stable source measurement without increasing +action regret. A shorter reconstructed prefix is only a trigger; it is not an +actual GPU-cost claim until an early-terminated confirmation run measures +startup, warm-up, drain, and cleanup. + +### H3: end-to-end cost + +Primary development metric is H20-hours to first reach a configuration within +2% of the exact median-goodput oracle. The outcome-only and telemetry policies +use the same measured config costs and differ only in source information. + +- At least 10% prospective replay cost reduction, telemetry regret at most 2%, + and no outcome-only-to-telemetry harm triggers an actual early-stop + confirmation. +- At least 20% measured all-in H20-hour reduction is required for a contribution + claim. This one task can only establish development feasibility; a paper + claim additionally requires task-held-out replication. +- Source median normalized goodput at or above 0.98 stops the surface before + target runs because the workload has no material improvement headroom. +- Any hash mismatch, missing/censored result, telemetry drop, non-monotonic + phase, negative cost, ratio outside `[0,1]`, or all-identical config outcomes + is a red flag and stops analysis. + +If the 10% trigger fails, this route is closed for the current engine-state +representation. The experimental control plane is not retained as a fallback +research contribution. + +Pre-run provenance amendment: the first controller dry-run on 2026-07-15 +rejected two stale engine paths before starting a server. The paths and exact +runtime version above were recovered from the accepted trace-12 campaign and +corrected before any trace-13 GPU work. No scientific treatment or gate was +changed. diff --git a/docs/active-intervention-v0-results-20260715.md b/docs/active-intervention-v0-results-20260715.md new file mode 100644 index 0000000..d4fc98c --- /dev/null +++ b/docs/active-intervention-v0-results-20260715.md @@ -0,0 +1,106 @@ +# Active intervention v0: held-out trace-13 result + +Date: 2026-07-15 (Asia/Singapore) + +Decision: **close the passive-telemetry treatment-effect route**. The held-out +campaign produced no telemetry-induced action change, measurement reduction, or +GPU-cost reduction. It did show that the engine state contained the correct +action-specific mechanism; the current feature model failed to use it. + +## Headline result + +The outcome-only and telemetry policies both measured the source for 300 +seconds, selected `joint=(MNS64,MBBT8192)`, and produced the same complete +acquisition order. Both reached the exact finite-surface oracle after the +first intervention at a reconstructed all-in lower-bound cost of 2.4284 +H20-hours. Telemetry GPU-cost reduction was therefore exactly 0%, below the +10% confirmation trigger and 20% contribution gate. No actual early-stop +confirmation was launched. + +The complete annotation campaign cost 5.0379 H20-hours, below the 6.0 H20-hour +hard cap. It ran 12 uncensored real-GPU outcomes: four configs, three disjoint +request partitions, and a fresh server per config. + +## Exact response surface + +Median normalized SLO-goodput was: + +| Config | Rep values | Median | +|---|---|---:| +| `MNS32,MBBT4096` source | 0.40091 / 0.39788 / 0.42061 | 0.40091 | +| `MNS64,MBBT4096` | 1.00000 / 0.99970 / 1.00000 | 1.00000 | +| `MNS32,MBBT8192` | 0.44394 / 0.41515 / 0.42606 | 0.42606 | +| `MNS64,MBBT8192` joint | 1.00000 / 1.00000 / 1.00000 | 1.00000 | + +Increasing MNS alone was sufficient and joint was redundant. Increasing MBBT +alone improved the median by only 0.02515, versus 0.59909 for MNS. This is a +strong non-additive action response, not a setting where independently tuning +the knobs and merging their improvements is valid. + +## What the telemetry actually said + +Across 41,086 source scheduler records, 93.12% of steps had waiting work, +85.36% were MNS-exclusive binding, 1.11% were MBBT-exclusive, mean running-slot +utilization was 97.39%, mean token-budget utilization was 15.69%, mean KV usage +was 2.75%, and there were no preemptions. + +The intervention transition agreed with that state: + +| Config | Waiting | MNS-exclusive | MBBT-exclusive | Median goodput | +|---|---:|---:|---:|---:| +| source | 93.12% | 85.36% | 1.11% | 0.40091 | +| MNS only | 5.38% | 0% | 5.38% | 1.00000 | +| MBBT only | 91.19% | 91.09% | 0.04% | 0.42606 | +| joint | 0.89% | 0% | 0.89% | 1.00000 | + +Thus this experiment does **not** support the claim that engine telemetry lacks +tuning information. It rejects the narrower claim that adding passive state +summaries to the current small-data ridge policy converts that information into +lower tuning cost. + +## Why the learned policy failed + +At 300 seconds, the telemetry model predicted joint, MNS, and MBBT effects of +0.35190, 0.26118, and 0.09686. The actual median effects were 0.59909, +0.59909, and 0.02515. Telemetry therefore made the nonexistent joint-over-MNS +gap larger: 0.09072 predicted versus 0 actual; the outcome-only model predicted +0.03188. + +The failure has three concrete causes: + +1. The six training decisions contain no joint intervention. The + `delta_product` feature has no support, so joint ranking is extrapolation. +2. Passive raw summaries do not represent the counterfactual scheduler work + unlocked by each action. Capacity-normalized MNS pressure was visible, but + the model was not structurally required to map it to MNS marginal value. +3. The policy maximizes predicted effect. It does not identify the smallest + epsilon-optimal intervention or price unsupported action complexity. + +## Research implication + +Do not retain the harness or the passive telemetry model as a contribution. +The next defensible route is engine-native, action-conditional counterfactual +instrumentation: at a real scheduling state, shadow-replay the exact scheduler +decision under an MNS relaxation, MBBT relaxation, and their joint relaxation, +then expose the incremental queued work admitted by each action. Real paired +interventions calibrate how those one-step shadow effects map to E2E SLO +goodput. This is distinct from a hand-written cap-to-knob rule and from a +full-system simulator: it reuses the exact live queue, scheduler, and cache +state while simulating only the local decision boundary. + +That route should be evaluated against outcome-only search, the present passive +telemetry model, a cap-hit expert rule, and a full simulator. The paper-level +gate remains at least 20% measured H20-hour reduction to a 2%-oracle config on +task-held-out workloads with at most 2% regret. + +## Sanity + +Surface outcomes: n=12, min=0.39788, max=1.0, distinct=8. Session costs: n=4, +min=1.1702, max=1.3566 H20-hours, distinct=4. Scheduler-record counts: n=4, +min=37,001, max=41,348, distinct=4. All counters and costs were non-negative; +all ratios were in `[0,1]`; request hashes matched; all 12 runs were uncensored; +the controller and four sessions completed; and config outcomes were not all +identical. No red flags were found. + +Machine-readable summary: `runs/active-intervention-v0/trace13-results.json`. +Raw immutable root: `/home/admin/cpfs/wjh/active-intervention-prospective-20260715`. diff --git a/docs/assets/simulator-fidelity/data.json b/docs/assets/simulator-fidelity/data.json new file mode 100644 index 0000000..6df95a9 --- /dev/null +++ b/docs/assets/simulator-fidelity/data.json @@ -0,0 +1,60 @@ +{ + "schema": "simulator-fidelity-figure-data-v1", + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "qwen30_mixed": { + "sources": { + "real": "recovered-stores/aituner-interaction-runs-dash1-20260710/interaction-mixed-qwen30b-tp-mns-surface-high1-dash1-d8899c5-20260701T095858Z", + "comparison": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json" + }, + "configs": [ + {"name": "tp1_mns8", "tp": 1, "mns": 8, "real": 2.1, "frontier_profile_only": 1.1, "frontier_calibrated": 1.7166666666666666}, + {"name": "tp1_mns16", "tp": 1, "mns": 16, "real": 2.35, "frontier_profile_only": 1.1, "frontier_calibrated": 2.3833333333333333}, + {"name": "tp1_mns32", "tp": 1, "mns": 32, "real": 2.283333333333333, "frontier_profile_only": 1.1, "frontier_calibrated": 2.3833333333333333}, + {"name": "tp1_mns64", "tp": 1, "mns": 64, "real": 2.283333333333333, "frontier_profile_only": 1.1, "frontier_calibrated": 2.3833333333333333}, + {"name": "tp2_mns8", "tp": 2, "mns": 8, "real": 2.275, "frontier_profile_only": 0.0, "frontier_calibrated": 1.7416666666666667}, + {"name": "tp2_mns16", "tp": 2, "mns": 16, "real": 2.275, "frontier_profile_only": 1.1916666666666667, "frontier_calibrated": 2.3}, + {"name": "tp2_mns32", "tp": 2, "mns": 32, "real": 3.283333333333333, "frontier_profile_only": 0.0, "frontier_calibrated": 3.75}, + {"name": "tp2_mns64", "tp": 2, "mns": 64, "real": 3.2583333333333333, "frontier_profile_only": 0.0, "frontier_calibrated": 3.75}, + {"name": "tp4_mns8", "tp": 4, "mns": 8, "real": 1.2833333333333334, "frontier_profile_only": 0.0, "frontier_calibrated": 1.3208333333333333}, + {"name": "tp4_mns16", "tp": 4, "mns": 16, "real": 2.441666666666667, "frontier_profile_only": 0.0, "frontier_calibrated": 2.5}, + {"name": "tp4_mns32", "tp": 4, "mns": 32, "real": 2.441666666666667, "frontier_profile_only": 1.3208333333333333, "frontier_calibrated": 2.5}, + {"name": "tp4_mns64", "tp": 4, "mns": 64, "real": 2.441666666666667, "frontier_profile_only": 1.3208333333333333, "frontier_calibrated": 2.5} + ], + "profile_only_metrics": { + "kendall_tau_b": 0.0, + "pairwise_exact_sign_accuracy": 0.3787878787878788, + "simulator_top_set": ["tp4_mns32", "tp4_mns64"], + "real_top_set": ["tp2_mns32"], + "top1_regret_worst": 0.25634517766497456 + }, + "calibrated_metrics": { + "kendall_tau_b": 0.9668009539030813, + "pairwise_exact_sign_accuracy": 0.9393939393939394, + "simulator_top_set": ["tp2_mns32", "tp2_mns64"], + "real_top_set": ["tp2_mns32"], + "top1_regret_best": 0.0, + "top1_regret_worst": 0.0076142131979695165 + } + }, + "qwen235_prefill": { + "source": "runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json", + "configs": [ + {"name": "tp4_mns64_mbt8192", "tp": 4, "mns": 64, "mbt": 8192, "expert_parallel": false, "real": 0.05, "frontier": 0.0375}, + {"name": "tp4_mns128_mbt8192", "tp": 4, "mns": 128, "mbt": 8192, "expert_parallel": false, "real": 0.05, "frontier": 0.0375}, + {"name": "tp4_mns64_mbt16384", "tp": 4, "mns": 64, "mbt": 16384, "expert_parallel": false, "real": 0.075, "frontier": 0.0625}, + {"name": "tp4_mns128_mbt16384", "tp": 4, "mns": 128, "mbt": 16384, "expert_parallel": false, "real": 0.075, "frontier": 0.0625}, + {"name": "tp8_mns64_mbt8192", "tp": 8, "mns": 64, "mbt": 8192, "expert_parallel": true, "real": 0.05625, "frontier": 0.05}, + {"name": "tp8_mns128_mbt8192", "tp": 8, "mns": 128, "mbt": 8192, "expert_parallel": true, "real": 0.05625, "frontier": 0.05}, + {"name": "tp8_mns64_mbt16384", "tp": 8, "mns": 64, "mbt": 16384, "expert_parallel": true, "real": 0.05625, "frontier": 0.05625}, + {"name": "tp8_mns128_mbt16384", "tp": 8, "mns": 128, "mbt": 16384, "expert_parallel": true, "real": 0.05625, "frontier": 0.05625} + ], + "metrics": { + "spearman_rank_correlation": 0.9486832980505138, + "pairwise_non_tied_accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "simulator_top_set": ["tp4_mns64_mbt16384", "tp4_mns128_mbt16384"], + "real_top_set": ["tp4_mns64_mbt16384", "tp4_mns128_mbt16384"], + "top1_regret_worst": 0.0 + } + } +} diff --git a/docs/assets/simulator-fidelity/qwen235-prefill-config-ranking.png b/docs/assets/simulator-fidelity/qwen235-prefill-config-ranking.png new file mode 100644 index 0000000..34b4045 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen235-prefill-config-ranking.png differ diff --git a/docs/assets/simulator-fidelity/qwen235-t0-fixed-shape-ranking.png b/docs/assets/simulator-fidelity/qwen235-t0-fixed-shape-ranking.png new file mode 100644 index 0000000..317b418 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen235-t0-fixed-shape-ranking.png differ diff --git a/docs/assets/simulator-fidelity/qwen30-mixed-config-ranking.png b/docs/assets/simulator-fidelity/qwen30-mixed-config-ranking.png new file mode 100644 index 0000000..df77685 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen30-mixed-config-ranking.png differ diff --git a/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png b/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png new file mode 100644 index 0000000..afe8d37 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png differ diff --git a/docs/assets/simulator-fidelity/qwen30-vllm020-profile-ablation.png b/docs/assets/simulator-fidelity/qwen30-vllm020-profile-ablation.png new file mode 100644 index 0000000..598a1a3 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen30-vllm020-profile-ablation.png differ diff --git a/docs/fidelity-aware-harness-headroom-20260714.md b/docs/fidelity-aware-harness-headroom-20260714.md new file mode 100644 index 0000000..6de9328 --- /dev/null +++ b/docs/fidelity-aware-harness-headroom-20260714.md @@ -0,0 +1,135 @@ +# Fidelity-aware harness headroom audit + +Status: **HISTORICAL PREMISE DID NOT PASS PROSPECTIVE P1; NO CONTRIBUTION CLAIM**. + +The audit answers whether engine instrumentation has enough incremental signal +to justify a prospective experiment. It does not establish generalization. + +Post-run update: the exact-timestamp held-out P1 completed and failed the +registered gate. Under the stronger simulator-aware `k=2` end-to-end replay, +telemetry preserved zero regret but saved only 1.426% online H20-hours versus +sim top-k + real final. The current route is closed; see +`docs/fidelity-aware-harness-p1-report-20260714.md`. + +## Simulator shortlist lower bound + +On the frozen 12-cell SimFid task, the strongest calibrated SLO simulator +reading places TP2/MNS32 and TP2/MNS64 in the same first tie bucket. Real-final +evaluation of that two-cell bucket selects TP2/MNS32 and has zero real regret. +A method requiring a real calibration probe plus final verification cannot beat +two real cell evaluations on this task. Therefore “better initial selection” +is not a viable claim here; the remaining headroom is shorter real verification +inside the same shortlist. + +## Five-second prefix result + +The retrospective Phase-6 dataset contains 37 primary anchors across 12 cells. +Stable labels use the frozen same-placement 2-of-3 adjudication: 28 feasible and +9 infeasible. Three TP4 primary measurements disagree with their repeated +labels, so single-run feasibility is not treated as ground truth. + +Using leave-one-cell-out folds, identical L2 logistic models, and a 5-second +prefix: + +| Metric | Outcome-only | Instrumentation-aware | Delta | +|---|---:|---:|---:| +| Accuracy | 78.38% | 89.19% | +10.81 pp | +| Balanced accuracy | 70.63% | 81.55% | +10.92 pp | +| Brier score | 0.1297 | 0.0901 | -0.0396 | +| Correct only in this model | 0 | 4 | +4 | +| McNemar exact two-sided p | — | 0.125 | not significant | + +At the frozen conservative threshold 0.95, both policies make zero false +accepts and zero false rejects on this retrospective set. Outcome-only safely +cuts 36.35% of measured primary-trial cost; instrumentation-aware safely cuts +61.10%, an additional 24.75 percentage points. Regularization sensitivity for +accuracy delta is `[0.00, +10.81]` percentage points, so the sign is +non-negative but the magnitude is not stable. + +Longer prefixes do not strengthen the case monotonically. At 10 seconds, +headline accuracy is 91.89% outcome-only versus 89.19% instrumentation-aware; +at 15 seconds it is 88.89% versus 91.67%; at 20 seconds it is 86.11% versus +91.67%, but both 0.95 policies make one false reject. Five seconds is therefore +a training-selected operating point, not a test result. + +## Strong simulator-aware calibration baseline + +The original nested comparison used the same simulator shortlist but did not +put Frontier's per-anchor prediction in either model. A stronger retrospective +audit now gives both models frozen-calibrated simulated throughput, simulated +SLO pass rate, and simulated feasibility. Under the same leave-one-cell-out +folds, 5-second cutoff, L2 logistic family, regularization 1.0, and threshold +0.95: + +| Metric | Sim + outcome | Sim + outcome + instrumentation | Delta | +|---|---:|---:|---:| +| Accuracy | 81.08% | 89.19% | +8.11 pp | +| Balanced accuracy | 72.42% | 81.55% | +9.13 pp | +| Brier score | 0.1058 | 0.0957 | -0.0101 | +| Safe early decisions | 20/37 | 25/37 | +5 | +| Valid full-trial cost reduction | 50.89% | 68.98% | +18.09 pp | +| Residual verification H20-hours | 0.5240 | 0.3310 | -36.84% | + +Both 0.95 policies have zero false accept and zero false reject on this +retrospective task. Only three 0.5-threshold classifications differ in favor +of instrumentation and none in favor of the strong baseline; McNemar's exact +two-sided p-value is 0.25. The cell-bootstrap accuracy-delta interval is +`[0.00,+18.18]` percentage points. The result is not robust to regularization: +at 0.1 the strong baseline is more accurate and the instrumentation policy +makes two unsafe decisions; at 10.0 the strong baseline is also more accurate. +Thus the stronger comparison still has enough point-estimate headroom for a +held-out test, but it materially weakens the evidence and makes a prospective +task-level result mandatory. + +## Interpretation + +There is enough headroom to run a held-out pilot, but not enough evidence to +claim the harness contribution: + +- the 5-second cost gap is operationally large; +- only four paired classifications differ, so significance is absent; +- all examples share one workload/SLO/engine task; +- completion timestamps are reconstructed from arrival + TTFT + TPOT rather + than recorded directly; +- three adjudication disagreements are concentrated in transient TP4 runs; +- outcome-only already recovers the simulator shortlist oracle with very few + real cells. + +The next experiment must therefore freeze the 5-second model and threshold, +record exact monotonic completions, use a held-out trace, and label each anchor +with three full repetitions. The registered protocol is +`docs/fidelity-aware-harness-protocol-20260714.md`. + +## Artifacts + +- `runs/fidelity-headroom/analyze_existing.py` +- `runs/fidelity-headroom/metrics.json` +- `runs/fidelity-headroom/analyze_prefixes.py` +- `runs/fidelity-headroom/prefix-metrics.json` +- `runs/fidelity-headroom/test_analysis.py` +- `runs/fidelity-headroom/test_prefix_analysis.py` +- `runs/fidelity-headroom/analyze_strong_baseline.py` +- `runs/fidelity-headroom/strong-baseline-metrics.json` +- `runs/fidelity-headroom/test_strong_baseline.py` + +## Sanity block + +| Family | n | Min | Max | Distinct | Invariant/result | +|---|---:|---:|---:|---:|---| +| Real SimFid cell scores | 12 | 1.2833 | 3.2833 | 7 | Non-negative; not identical | +| Prefix examples at 5 s | 37 | 5 s | 5 s | 1 expected | All 12 cells represented | +| Adjudicated labels | 37 | 0 | 1 | 2 | 28 positive / 9 negative | +| Primary/adjudicated disagreement | 37 | 0 | 1 | 2 | 3 TP4 disagreements retained | +| Full primary elapsed time | 37 | 14.566 s | 62.064 s | 37 | Every 5 s prefix is in range | +| Outcome probability | 37 | in `[0,1]` | in `[0,1]` | >1 | Checked before metrics | +| Instrumentation probability | 37 | in `[0,1]` | in `[0,1]` | >1 | Checked before metrics | +| Layer-1 streams | 12 | 14,174 records | 58,725 records | 12 | Contiguous, zero drops | +| Matched frozen simulator anchors | 37 | pass rate 0.0688 | pass rate 1.0 | 12 pass-rate values | Every prefix matched exactly once | +| Frozen simulator anchor corpus | 92 | positive throughput | positive throughput | >1 | No duplicate cell/anchor run | + +Checked invariants: same folds/model family and cutoff; no full verdict in a +feature; prefix-only Layer-1 slicing; non-negative costs/counters; bounded +ratios/probabilities; both labels present; per-config results not identical; +tie expansion before top-k; no imputation of non-monotonic frontiers. The main +limitation is reconstructed request completion time, explicitly marked on all +37 five-second examples. diff --git a/docs/fidelity-aware-harness-p1-report-20260714.md b/docs/fidelity-aware-harness-p1-report-20260714.md new file mode 100644 index 0000000..b996dca --- /dev/null +++ b/docs/fidelity-aware-harness-p1-report-20260714.md @@ -0,0 +1,242 @@ +# Fidelity-aware harness P1 result + +Status: **REGISTERED ROUTE REJECTED; DO NOT OPEN P2/P3 FOR THE CURRENT METHOD**. + +Date: 2026-07-14 (Asia/Singapore). + +## Outcome + +The registered five-second instrumentation-aware verifier did not pass P1. +The stronger simulator-aware comparison also failed the independent +contribution bar. On the frozen `k=2` end-to-end replay: + +- `sim top-k + real final` selected the real oracle with zero regret; +- instrumentation-aware also selected the oracle, but reduced online H20-hours + by only **1.426%** (1.329% when the prior failed attempt is added to both); +- the required reduction was 30% versus full real final and 20% versus a safe + outcome-only calibrator; +- the outcome-only calibrator was not safe: it rejected the true best cell, so + its apparent cost saving is not a deployable comparison. + +This rejects the claim that the **current joint logistic verifier**, trained on +one historical workload, gives the harness an independent tuning contribution. +It does not prove that engine telemetry contains no useful signal. Telemetry +improved held-out classification and removed unsafe decisions, but did not turn +that signal into meaningful end-to-end tuning-cost reduction. + +## Frozen setup + +- Host: `dash0`, 8 NVIDIA H20 GPUs; cells were serialized and used TP1, TP2, + or TP4 without co-resident serving jobs. +- Engine/model: patched vLLM 0.24.1.dev3, Qwen3-30B-A3B BF16. +- Workload: held-out `chat_w20260312_1000`, seven disjoint repeat bands, + 60-second replay after 0.1 time scaling, input `[0,8192]`, exactly 128 output + tokens. +- SLO: stepped TTFT 2/4/6 seconds, TPOT 50 ms, request pass rate at least 0.95. +- Cells: TP1/MNS8, TP1/MNS64, TP2/MNS8, TP2/MNS64, TP4/MNS16, TP4/MNS64. +- Per cell: burn-in, three low-rate repeats, and three high-rate repeats. The + first repeat supplied the five-second prefix; 2-of-3 supplied its label. +- Models: the registered pair used config/workload/outcome versus the same + vector plus Layer-1 engine telemetry. The strengthened pair additionally + gave both models identical frozen Frontier throughput, SLO pass-rate, and + feasibility predictions. +- Policy: accept at `p>=0.95`, reject at `p<=0.05`, otherwise continue the same + trial. Model, cutoff, threshold, role order, request hashes, and cap were + frozen before their applicable evaluation. + +The first launch failed its warm-up input-count validation before a measured +anchor. It cost 0.020552 H20-hours. The corrected primary attempt cost +1.722112 H20-hours, so aggregate campaign cost was **1.742664 H20-hours**, below +the 3.5 cap. The fix changed only warm-up validation; formal request counts and +hash checks were unchanged. + +## P1 labels are not an artificial easy split + +The 12 adjudicated anchor labels contain 7 feasible and 5 infeasible examples. +They are not simply “low feasible, high infeasible”: + +- TP2/MNS64 high was feasible in all three repeats; +- TP4/MNS64 low and high were feasible in all six repeats; +- TP4/MNS16 low and high were infeasible in all six repeats. + +That last pair creates a large real MNS interaction under an otherwise matched +TP4 configuration. Frontier correctly predicted TP4/MNS64 high as feasible, +but incorrectly predicted TP4/MNS16 low as feasible. It also incorrectly +predicted TP1/MNS64 high as feasible. Overall simulator-only feasibility was +10/12 correct: 83.33% accuracy, with two false-feasible predictions and no +false-infeasible prediction. + +The two false-feasible cases expose the intended latent-state problem. At five +seconds, all 26 completed TP4/MNS16-low requests and all 9 completed +TP1/MNS64-high requests still passed their SLO, although both full anchors were +infeasible. External outcomes had not yet exposed the future failure; queue, +running-batch, and scheduler state existed before the tail outcome. This is +mechanistic evidence that instrumentation can be useful, not evidence that the +current learned policy uses it well enough. + +## Registered and strengthened prefix results + +At the frozen 0.95 policy threshold: + +| Comparison | Accuracy | Balanced acc. | Early decisions | False accept | False reject | Valid primary-trial saving | +|---|---:|---:|---:|---:|---:|---:| +| Registered outcome-only | 41.67% | 50.00% | 6/12 | 0 | 2 | invalid | +| Registered + telemetry | 66.67% | 71.43% | 4/12 | 0 | 0 | 11.44% | +| Strong sim + outcome | 66.67% | 68.57% | 5/12 | 0 | 1 | invalid | +| Strong sim + outcome + telemetry | 83.33% | 85.71% | 4/12 | 0 | 0 | 11.44% | + +For the strong pair, telemetry was correct on two examples where the baseline +was wrong and lost none; McNemar's exact two-sided p-value is 0.5 at `n=12`. +This is a safety/classification improvement, not a cost contribution. The +registered instrumentation policy made two fewer early decisions than its +baseline, so it failed the registered `+3 decisions or +15 percentage points` +incremental gate. + +The result is not robust to the frozen regularization sensitivity: + +| L2 lambda | Sim+outcome acc. | +telemetry acc. | Base policy errors | Telemetry policy errors | Base saving | Telemetry saving | +|---:|---:|---:|---:|---:|---:|---:| +| 0.1 | 41.67% | 75.00% | 4 | 2 | invalid | invalid | +| 1.0 | 66.67% | 83.33% | 1 | 0 | invalid | 11.44% | +| 10.0 | 83.33% | 83.33% | 0 | 0 | 0.00% | 5.98% | + +Consequently the positive classification delta is neither statistically nor +hyperparameter robust. + +## End-to-end shortlist result + +Frontier's simulator-feasible ranking on the tested P1 surface was: + +| Rank | Cell / anchor | Sim throughput/GPU | Real feasible | Real offered goodput/GPU | +|---:|---|---:|---:|---:| +| 1 | TP4/MNS64 high | 3.0718 | yes | 3.1250 | +| 2 | TP1/MNS64 high | 2.8823 | no | 2.9833 | +| 3 | TP2/MNS64 high | 2.8096 | yes | 2.8750 | +| 4 | TP4/MNS16 low | 2.0866 | no | 2.1250 | +| 5 | TP1/MNS8 low | 1.9806 | yes | 2.0333 | +| 6 | TP2/MNS8 low | 1.8637 | yes | 1.9083 | + +The real oracle was TP4/MNS64 high at 3.125 req/s/GPU. Cost includes an +inferred per-cell startup/warm-up/burn-in/cleanup component and the selected +anchor; benchmark-only 2-of-3 annotation intervals are removed. Gaps around +annotation runs remain in the shared setup term, making this a conservative +method-cost estimate. + +| Frozen policy (`k=2`) | Online H20-hours | + prior failure | Real regret | Safety | Saving vs full | +|---|---:|---:|---:|---|---:| +| Sim top-2 + real final | 0.281383 | 0.301935 | 0.00% | valid | — | +| Sim + outcome prefix | 0.214664 | 0.235216 | no selected cell | 1 false reject | invalid | +| Sim + outcome + telemetry | 0.277370 | 0.297922 | 0.00% | valid | **1.426%** | + +For `k=1`, the simulator already picked the oracle and telemetry abstained, so +its saving was exactly 0%. For `k=3`, telemetry remained correct but saved +1.007%. The result therefore does not depend on an unfavorable choice of `k`. + +Across all six cells, a generous replay that removes benchmark-only repeat +intervals estimates 0.944244 H20-hours for the full online workflow. The +instrumentation policy saved 0.043467 H20-hours: 11.44% of primary trial time, +but only **4.603%** of end-to-end online cost. Even a post-hoc oracle symmetric +threshold cannot make the current telemetry model reach the contribution bar; +its best zero-error envelope saves at most 5.69% of inferred online cost. A +strong outcome-only model at another post-hoc regularization/threshold can save +16.13%, which further prevents attributing a unique advantage to telemetry. +These oracle-threshold numbers are diagnostics only and are not test evidence. + +## Why the learned verifier did not generalize + +The training corpus has only 37 anchors from one workload/SLO task. P1 shows +large covariate shift: + +- sim+outcome: 12/192 feature values exceed 3 training standard deviations and + 4 exceed 5; maximum absolute z-score is 10.36; +- sim+outcome+telemetry: 19/396 exceed 3 and 9 exceed 5; +- the largest shifts include admitted input-length mean (10.36), waiting state + (7.77), running maximum (6.38), and decode-batch maximum (6.08). + +Coefficient attribution shows that the input-length feature dominates several +wrong feasible-anchor logits. Because all training examples share one task, +the joint classifier can learn incidental within-task correlation and override +a correct simulator prior on TP2/MNS64-high and TP4/MNS64-high. This is a +supported diagnosis of model/data insufficiency; it is not a causal proof that +one feature alone caused the P1 failure. + +More importantly, retuning lambda, threshold, features, or cutoff on P1 and +then calling P1 a held-out result would violate calibration/evaluation +separation. P1 may now be used only as development data. + +## Decision and the only defensible reopening condition + +Do not run registered P2/P3 with the current model. It failed the predeclared +gate on the favorable primary-trial denominator and is even farther from the +bar under end-to-end cost. Spending six-task headline GPU budget on the same +method would be metric shopping, not replication. + +A new route may be opened only as a new hypothesis: + +1. Replace the joint classifier with a **simulator-residual verifier**. The + simulator prediction remains an explicit prior; nested outcome-only and + telemetry models learn when that prior is wrong, rather than freely + relearning feasibility and overriding it under workload shift. +2. Train on multiple complete workload/SLO tasks. SLO thresholds and target + pass rate must be explicit inputs; splits are by complete task. +3. Calibrate abstention with task-level risk control. No threshold is selected + on a headline task, and “never early decide” is included as the safe + outcome-only baseline. +4. Treat Phase 6 and P1 as development only, freeze the residual architecture, + features, cutoff, threshold, simulator reading, and `k`, then use entirely + new trace windows for a new gate. + +This reopening is justified only if development data show both (a) the +simulator's errors are predictable from pre-outcome engine state and (b) a +simulator-preserving residual model does not corrupt correct simulator +predictions. It is a new project decision, not a continuation automatically +authorized by P1. + +## Benchmark audit + +| Audit item | Verdict | Severity | Evidence / disposition | +|---|---|---|---| +| Calibration set separate from P1 | PASS | — | Phase 6/0311 trained; P1/0312 tested | +| Strong simulator-aware baseline | PASS | — | Identical Frontier features in both nested models | +| Sim top-k + real-final E2E baseline | PASS | — | Frozen `k=2`, tie expansion, measured setup/continuation cost | +| Multiple independent headline tasks | NEEDS EVIDENCE | Blocking for a positive claim | P1 gate failed; P2 correctly not opened | +| Statistical significance | NEEDS EVIDENCE | Blocking for a positive claim | n=12 anchors from one task; McNemar p=0.5 | +| Hyperparameter robustness | FAIL | Blocking | Lambda sensitivity changes safety and relative result | +| Full resource accounting | PASS for P1 | — | Failures, startup/warm-up/burn-in, continuation and annotation separated | +| Avoid post-test retuning | PASS only if route stops | Blocking if violated | P1 is now development-only | +| Selective winning-workload reporting | PASS | — | Negative P1 and TP/MNS losing cases retained | + +Overall recommendation: **Block the current independent harness contribution +claim.** + +## Artifacts + +- Registered protocol: `docs/fidelity-aware-harness-protocol-20260714.md` +- Historical headroom: `docs/fidelity-aware-harness-headroom-20260714.md` +- Registered P1 analysis: `runs/fidelity-headroom/analyze_pilot.py` +- Strong P1 analysis: `runs/fidelity-headroom/analyze_strong_pilot.py` +- E2E shortlist replay: `runs/fidelity-headroom/analyze_pilot_e2e.py` +- External immutable result root: + `/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714` + +## Data sanity block + +| Data | n | Min | Max | Distinct | Invariant | +|---|---:|---:|---:|---:|---| +| P1 labels | 12 | 0 | 1 | 2 | 7 feasible / 5 infeasible | +| Primary elapsed seconds | 12 | 19.448 | 61.435 | 12 | Every five-second prefix is in range | +| Prefix Layer-1 records | 12 | 332 | 557 | 12 | Contiguous; zero drops | +| Exact timestamped outcomes | 12 anchors | 54 | 750 | 11 | Monotonic completion timestamps | +| Simulator pass rate | 12 | 0.1548 | 1.0 | 7 | Ratios in `[0,1]` | +| Strong nested probabilities | 24 | 0.000208 | 0.809422 | 24 | Ratios in `[0,1]` | +| E2E cost components | 36 | 0.001389 | 0.169653 H20-h | 21 | Non-negative | +| GPU attempts | 2 | 0.020552 | 1.722112 H20-h | 2 | Aggregate 1.742664 < 3.5 | +| Copied raw files | 191 | — | 153,093,348 bytes total | — | Remote/local aggregate SHA identical | + +Checked invariants: six cells and twelve anchors; exact request count and +request-ID/arrival/length hashes; all cell validation flags true; both labels +present; probabilities bounded; costs and counters non-negative; simulator +results not all identical; committed simulator rerun 12/12 numerically +identical to the exploratory run; no prompt text in public simulator fixtures; +no co-resident serving process; final eight GPUs at 0 MiB and 0% utilization. +No red flag remains. diff --git a/docs/fidelity-aware-harness-protocol-20260714.md b/docs/fidelity-aware-harness-protocol-20260714.md new file mode 100644 index 0000000..b6176f5 --- /dev/null +++ b/docs/fidelity-aware-harness-protocol-20260714.md @@ -0,0 +1,231 @@ +# Fidelity-aware real-verification harness protocol + +Status: **P1 FAILED; P2/P3 CLOSED FOR THE REGISTERED METHOD; CONTRIBUTION NOT ESTABLISHED**. + +Date frozen: 2026-07-14 (Asia/Singapore). + +Post-run disposition (2026-07-14): P1 completed with valid data but failed its +registered incremental gate. The strengthened simulator-aware comparison and +end-to-end `k=2` replay also failed: instrumentation was safe and retained zero +regret, but reduced online H20-hours by only 1.426% versus sim top-k + real +final, against the 30% bar. Outcome-only was unsafe. P2/P3 are therefore not +opened for this model. Full results and the permitted reopening condition are +in `docs/fidelity-aware-harness-p1-report-20260714.md`; the protocol below is +retained unchanged as the pre-run record. + +## Research question and contribution bar + +The harness has an independent systems contribution only if engine-internal +instrumentation improves a tuning decision beyond what is already achievable +with a simulator shortlist and external benchmark outcomes. The intended +claim is therefore deliberately stronger than “telemetry explains a run”: + +> Given the same simulator ranking, the same candidate order, and the same +> short real-GPU probe, a learned instrumentation-aware verifier reaches a +> configuration with at most 5% real SLO-goodput regret using materially fewer +> H20-hours than both (a) simulator top-k followed by full real evaluation and +> (b) an outcome-only verifier given exactly the same probe. + +The paper-facing gate is: + +- at least 20% lower real-verification H20-hours than outcome-only calibration; +- at least 30% lower real-verification H20-hours than simulator top-k plus full + real final evaluation; +- paired 95% task-bootstrap confidence interval for the outcome-only cost + reduction strictly above zero; +- selected-configuration SLO-goodput regret at most 5% on every headline task; +- no false-safe early accept in the pilot and at most 1% in the expanded suite; +- profiling, warm-up, confirmation, instrumentation, and failed-run costs are + included rather than amortized away. An amortized profile-cost view may be + reported only as a secondary result. + +If these conditions fail, instrumentation remains a debugging facility. It is +not an independent tuning-harness contribution. + +## What is learned, and what is not a rule + +The decision target is a stable, repeated real verdict, not a hand-authored +diagnosis such as “queue length above N means reject.” Each anchor receives +three full real repetitions and a frozen 2-of-3 feasibility label. A nested +pair of regularized models predicts that label from a fixed prefix: + +- **Outcome-only input X:** configuration, offered rate, admitted/completed + progress, observed TTFT/TPOT margins, failures, and known workload lengths. +- **Instrumentation input Z:** the same X plus generic engine state: running and + waiting queues, decode-batch shape, KV usage, graph mode and padding, prefill + share, preemptions, and model-step rate. + +Both models use the same L2 logistic family, train split, standardization, +regularization, cutoff, and probability threshold. The only experimental +difference is Z. The initial family is intentionally simple: a positive result +then demonstrates value in the engine signal rather than capacity in a larger +learner. A sequence model is admissible only as a later, paired ablation. + +### Amendment A1: strengthen the calibration baseline before P2 + +Frozen 2026-07-14 13:08 Asia/Singapore, after P1 launch but before P1 +completion or analysis. A baseline audit found that the first frozen P1 +models use the simulator only to define candidate order; their feature vectors +do not contain the simulator's per-anchor prediction. This is insufficient +for the stronger term **outcome-only calibration**. P1 therefore remains a +prospective test of the originally frozen cross-workload predictor, but cannot +by itself open a contribution claim. + +For P2/P3, both nested models must additionally receive the identical frozen +simulator outputs available at that decision: predicted completed throughput +per GPU, predicted SLO pass rate, and predicted feasibility. The comparison +is consequently `sim + config + workload + real outcome prefix` versus that +exact vector plus real engine state. Simulator features, regularization, +cutoff, and thresholds are frozen before any P2 task. If telemetry does not +improve this stronger baseline, the harness has no independent contribution. + +The same audit also separates algorithm cost from benchmark-oracle cost. +Headline method cost includes every action the method would execute online: +simulator profiling/calibration, model onboarding, server startup, warm-up, +real prefix, continuation after abstention, method-requested confirmation, +logging overhead, failures, and cleanup. Exhaustive real-oracle runs and the +extra repetitions used only to construct 2-of-3 evaluation labels are common +benchmark annotation cost; they are reported separately and charged to no +method. A second, deliberately conservative table adds that common cost to +all methods. This prevents both hiding real method cost and making the +percentage gate mathematically depend on offline ground-truth annotation. + +The frozen first policy uses a 5-second prefix, L2 regularization 1.0, and a +two-sided abstaining threshold of 0.95: accept at `p(feasible)>=0.95`, reject at +`p(feasible)<=0.05`, otherwise continue the exact same trial to completion. +Threshold and cutoff were selected on the historical training task and are +therefore not evidence; all claims come from subsequent held-out tasks. + +## Fair baselines + +| Method | Simulator | 5-second real prefix | External outcomes | Engine state | Full real continuation | +|---|---:|---:|---:|---:|---:| +| Real-only oracle | no | no | full | optional diagnostic | every candidate/anchor | +| Sim top-k + real final | yes | included in full run | full | no decision use | every shortlisted candidate/anchor | +| Outcome-only calibration | yes, including its prediction features | yes | yes | no | only on abstention | +| Instrumentation-aware | same prediction features | yes | yes | yes | only on abstention | + +Tie buckets are expanded before top-k. `k` is selected on training tasks and +is fixed on held-out tasks; an oracle per-task k is forbidden. Outcome-only +receives all information available outside the engine, including config, +workload, and frozen simulator-prediction features. Instrumentation cannot use +any record submitted after the cutoff. The full label, confirmation votes, +realized simulator error, and later requests are never model features. + +## Staged experiment + +### R0: historical premise and headroom audit + +The frozen SimFid surface has 12 cells. The strongest calibrated SLO simulator +reading has a top tie bucket `{TP2/MNS32, TP2/MNS64}`; full real evaluation of +those two cells already finds the oracle with zero regret. Consequently this +single task cannot demonstrate a selection-count advantage: any method needing +one real calibration probe and one real final verification has a lower bound of +two real cells. + +The viable estimand is instead the duration and number of full real frontier +evaluations inside a fixed shortlist. Historical Phase-6 prefixes are analyzed +only as training/premise data. Their request completion times are reconstructed +from arrival, TTFT, TPOT, and token count, so they cannot support a final claim. + +### P1: exact-timestamp prospective pilot + +- Engine/model/hardware: patched vLLM 0.24.1.dev3, Qwen3-30B-A3B, one solo + server/client on dash0, NVIDIA H20, `TP in {1,2,4}`. +- Held-out workload: `chat_w20260312_1000`, 60-second replay after the frozen + 0.1 time scale, raw input `[0,8192]`, exactly 128 output tokens. +- SLO: stepped TTFT 2/4/6 seconds, TPOT 50 ms, 95% request pass rate. +- Cells: TP1/MNS8, TP1/MNS64, TP2/MNS8, TP2/MNS64, TP4/MNS16, TP4/MNS64. +- Per cell: one attainable low offered rate near 0.85x the historical v0.24 + frontier and one high rate near 1.25x. The exact threshold and selected + request hashes are frozen by a CPU preflight before launch. +- Each cell uses a fresh server, the accepted long-request warm-up, one + unmeasured full-window burn-in, then three repetitions per rate. Rate order + alternates and reverses across cells to prevent a fixed warm-state/order + confound. +- The first repetition supplies the exact prefix. All three repetitions supply + the 2-of-3 label. Every request records a monotonic completion timestamp; + Layer-1 records are cut at the same monotonic boundary. +- Placement is serialized. Co-location is forbidden because Phase 6 observed + up to 92.86 percentage-point pass-rate shifts under co-location. +- Hard cap: 3.5 H20-hours, including startup, warm-up, burn-in, all repetitions, + failures, and cleanup. Projected cap violation stops before the next cell. + +P1 opens P2 only if all data invariants pass and instrumentation-aware has zero +false accept/reject, is no worse than outcome-only, and either makes at least +three additional correct early decisions or improves total valid trial-cost +reduction by at least 15 absolute percentage points. The pilot is a gate, not +paper evidence. + +### P2: held-out task replication + +If P1 passes, freeze the model and run at least six independent task groups: +three trace windows spanning distinct date/slot combinations and two SLO +regimes. No task used for threshold/model selection enters the headline test. +The candidate surface is the full 12-cell `TP={1,2,4} x MNS={8,16,32,64}` +surface. Splits are by complete task, never by anchor or request. A task-level +paired bootstrap (10,000 repetitions, fixed seed) estimates cost and regret +intervals. Non-monotonic or split 2-of-3 anchors remain explicit; no frontier +is imputed. + +### P3: end-to-end shortlist and search replay + +For each P2 task, run the same frozen simulator and tie-expanded top-k policy. +Replay the real binary/frontier search under all three verification policies: +full real, outcome-only, and instrumentation-aware. The policy consumes only +prefixes that would have been available at that decision point. Report: + +- selected cell and real SLO-goodput regret; +- number of real cells, anchors, and confirmations; +- measured H20-hours and wall time; +- false accept, false reject, and abstention counts; +- profile, startup/warm-up, probe, full-continuation, confirmation, logging, and + failure cost breakdowns. + +### P4: simulator-rank-error attribution + +This phase distinguishes an outdated implementation/profile from a structural +simulator limitation. For each held-out task compare: + +1. the original simulator/profile; +2. a version-matched re-profiled simulator; +3. a trajectory-conditioned run supplied with the realized arrival and request + length sequence; +4. outcome-only residual calibration; +5. instrumentation-aware residual calibration. + +The engine trace is extended only as needed with a worker-level step UID and +CUDA-event duration, because current async submit-to-complete spans overlap and +are not GPU step time. Residuals are decomposed into operator-profile error, +scheduler/state error, and run-to-run noise. If re-profiling alone restores the +ranking, the old 30% loss was an implementation/profile defect. If exact +profiles and realized trajectories still mis-rank cells, and the residual is +systematically explained by queue/KV/graph/batch state unavailable to the +simulator, that is evidence of a structural state-abstraction gap. Correlation +alone is not called causal. + +## Failure modes that reject the route + +- Outcome-only matches or beats instrumentation-aware under the same cutoff. +- Instrumentation gains average accuracy but introduces false-safe decisions. +- Gains disappear under task-level rather than request/anchor-level splitting. +- Savings come only from excluding startup, warm-up, profiling, confirmations, + or failed trials. +- A different cutoff/threshold must be selected after seeing each test task. +- The simulator top-k baseline already reaches the target with equal or lower + total H20-hours. +- Exact instrumentation overhead exceeds 1% throughput or materially changes + p95/p99 latency. +- Results depend on TP4 transient/non-monotonic trials and do not replicate on + held-out tasks. + +## Data sanity contract + +Every analysis ends with n, min/max, distinct count, label balance, and these +invariants: non-negative counters/costs; probabilities and ratios in `[0,1]`; +per-config results not all identical; timestamps monotonic; every prefix record +at or before its cutoff; selected request ID/arrival/length hashes stable across +repetitions; exact 128-token completion or counted failure; no dropped Layer-1 +records; 2-of-3 labels reproducible; no co-resident GPU process; total H20-hours +below the hard cap; final GPUs idle. A red flag is reported first and blocks +the contribution claim. diff --git a/docs/intervention-response-v0-protocol-20260714.md b/docs/intervention-response-v0-protocol-20260714.md new file mode 100644 index 0000000..47742f6 --- /dev/null +++ b/docs/intervention-response-v0-protocol-20260714.md @@ -0,0 +1,71 @@ +# Telemetry intervention-response v0 protocol + +Status: **FROZEN BEFORE V0 ANALYSIS**. + +Date: 2026-07-14 (Asia/Singapore). + +## Claim boundary + +The closed residual route asked whether one absolute engine-state snapshot can +predict unmeasured configurations. V0 asks a different, narrower question: + +> Does an adjacent, controlled MNS intervention produce an early engine-state +> response that is distinguishable from same-config repeat noise? + +Passing this gate only authorizes a matched real-GPU pilot. It does not prove +that telemetry improves tuning, that any metric is a causal mediator, or that +the response transfers to a new workload, topology, or knob family. + +## Data and estimand + +- Source: Phase 6 solo-authoritative Qwen3-30B-A3B/vLLM 0.24 Layer-1 streams. +- Action pairs: primary runs at identical study hash, TP, sampling anchor, and + request-order hash, with adjacent `MNS={8,16,32,64}` values. +- Noise pairs: primary versus confirmation at the same complete config, + anchor, and request-order hash. Only primary-to-confirmation pairs are used; + confirmations are not combined into pseudo-independent all-pairs. +- Fixed early windows: 5 seconds and 10 seconds from the measured interval + start. All runs exceed 10 seconds, so early-stop censoring cannot change the + telemetry window. +- Full-run pass rate and feasibility are descriptive only because an early + stop can make full elapsed durations differ. + +The statistical unit is a run pair. Scheduler steps are summarized within a +run and are never counted as independent trials. + +## Frozen response gate + +The directly measured gate features are scheduler-step rate, decode-batch +mean, prefill-token fraction, waiting/running queue mean, KV-usage mean, and +CUDA-graph padding fraction. + +A feature qualifies at one horizon only if: + +1. at least 75% of nonzero action deltas have the same sign; +2. median absolute action delta is at least 2x the median absolute repeat + delta; and +3. at least 50% of action deltas exceed the repeat-noise absolute p95. + +V0 opens a GPU pilot only if: + +- there are exactly 17 frozen adjacent-MNS action pairs; +- there are at least 20 primary/confirmation repeat pairs; +- all identity, finite-value, counter, and ratio invariants pass; and +- at least two gate features qualify at both 5 and 10 seconds. + +Any data red flag stops the analysis before interpreting the response. + +## If V0 passes + +Register a dash0 pilot around a known scaling knee. The pilot must use the +same request sequence and arrival times, one serving job at a time, one changed +knob, randomized `A/B` versus `B/A` order, common non-censored measurement +windows, and trial-level repetitions. It must compare a response-aware next +action against an outcome-only policy under complete startup, warm-up, and +H20-hour accounting. + +## If V0 fails + +Do not add telemetry fields or train a larger model. The current Layer-1 state +does not identify even an MNS intervention above repeat noise on this task, so +the telemetry-guided tuning route remains diagnostic only. diff --git a/docs/intervention-response-v0-results-20260714.md b/docs/intervention-response-v0-results-20260714.md new file mode 100644 index 0000000..7bc2996 --- /dev/null +++ b/docs/intervention-response-v0-results-20260714.md @@ -0,0 +1,157 @@ +# Telemetry intervention-response v0/v1 results + +Date: 2026-07-14 (Asia/Singapore). + +## Decision + +**STOP before a new H20 pilot.** The current Layer-1 aggregate telemetry does +not identify a sufficiently general early response to an MNS intervention, +and it does not improve action-efficacy prediction over exact external prefix +outcomes on the available development tasks. + +This is a negative result about the present state representation and +experiment design. It does not establish that engine telemetry is useless for +tuning, and it is not held-out evidence. + +## Hypothesis and frozen test + +The tested hypothesis was: + +> With the workload and all non-MNS settings held fixed, increasing MNS causes +> a 5--10 second engine-state response that is larger than same-config repeat +> noise and that predicts whether the action makes the full run feasible. + +A response feature had to satisfy all three frozen conditions at both 5 and +10 seconds: at least 0.75 sign consistency, median absolute action effect at +least 2x the repeat median, and at least 0.50 of action deltas above the repeat +absolute p95. At least two features had to pass. A telemetry feature was +decision-relevant only if its leave-one-repeat-out balanced accuracy was at +least 0.75 and at least 0.15 above the best exact external prefix outcome. + +## What was implemented + +- A common-window analyzer over the existing per-scheduler-step Layer-1 stream. +- Exact action pairing with request-order hash, offered load, TP, load role, + and repetition held fixed. +- Same-config repeat-noise estimation without treating scheduler steps as + independent samples. +- Exact 5/10-second request-prefix outcomes using monotonic completion times. +- A one-feature leave-one-repeat-out efficacy audit; no multivariate model was + fitted to the 12 examples. +- Input hashes, stream hashes, frozen thresholds, pair-level deltas, and sanity + invariants in machine-readable audit artifacts. +- Trial-by-trial validation against the P1 manifest, plus content hashes for + every result, request file, and Layer-1 stream. + +## Experiment A: Phase-6 retrospective audit + +Phase 6 supplied 17 adjacent-MNS actions and 29 same-config +primary/confirmation pairs. No feature passed at either horizon, producing +`STOP_NO_IDENTIFIABLE_RESPONSE`. + +The confirmation sample is not a clean replication distribution: confirmations +were selectively run after disputed primary outcomes. Several same-config +pairs consequently followed radically different trajectories. This result +therefore remains a valid failure of the frozen v0 gate, but it cannot by itself +separate normal run variance from confirmation-selection bias. + +## Experiment B: prospective-repeat confirmation + +P1 supplied three pre-arranged, disjoint request bands for every cell/load. +Exact matched actions exist for TP1 `MNS 8 -> 64` and TP4 `MNS 16 -> 64`, at +low/high load and repetitions 1/2/3. This yields 12 action pairs and 24 +same-config consecutive-repeat pairs. + +The 24 adjacent repeat differences share their middle repetition within each +three-run group. They define a conservative empirical noise reference; they +are not used as 24 independent samples in an inferential test. + +The result is `STOP_NO_PROSPECTIVE_RESPONSE`: zero features passed the response +gate at either horizon. + +The strongest response was mean waiting-queue occupancy: + +| Horizon | Sign consistency | Action/repeat median | Action above repeat p95 | Gate | +|---|---:|---:|---:|---| +| 5 s | 1.000 | 1.292x | 0.167 | fail | +| 10 s | 1.000 | 2.611x | 0.250 | fail | + +The direction is real enough to merit diagnosis, but the effect is not broad +enough to guide a general action. It is large for TP4/high-load trials and +small or absent in other regimes. + +Full-run transitions contain six beneficial actions (`false -> true`) and six +non-beneficial actions (three `false -> false`, three `true -> true`). The +beneficial label is also perfectly confounded with TP4 in this small dataset, +so it cannot support a topology-general claim. + +| Horizon | Best telemetry delta | Balanced accuracy | Best external prefix delta | Balanced accuracy | Telemetry advantage | +|---|---|---:|---|---:|---:| +| 5 s | waiting queue | 0.750 | max TPOT / SLO | 0.833 | -0.083 | +| 10 s | waiting queue | 0.750 | outstanding / admitted | 0.750 | 0.000 | + +No telemetry feature reaches the preregistered `+0.15` incremental threshold. + +## What this rules out + +It rules out using the current vector of 5/10-second global means as a solid +mechanism for choosing the next config. In particular, adding these aggregates +to an LLM prompt or fitting a larger predictor would currently hide, rather +than solve, the identifiability problem. + +It does not rule out an instrumentation-aware tuner built around a deliberately +excited local system. The existing runs were designed for endpoint/fidelity +evaluation, not system identification: the MNS action is large, efficacy is +confounded with TP, repeat bands contain different requests, and global means +erase when queue buildup or service-rate changes occur. + +## Required redesign before spending H20-hours + +The next admissible experiment is a randomized, local A/B system-identification +pilot around one fixed TP and one load knee: + +1. Replay the exact same request sequence and arrival times for both endpoints. +2. Use small adjacent actions and randomized `A/B` versus `B/A` order. +3. Record event-aligned response curves, including queue growth/drain rate, + prefill/decode service rate, and per-step service time, rather than only one + global mean. +4. Separate a mechanism gate (repeatable response) from the end-to-end gate: + fewer trials or H20-hours to select a feasible near-optimal config than an + outcome-only tuner. +5. Hold out a second load/workload for the final policy comparison. + +Until that design is frozen, a wider sweep would only generate more correlated +observations and is not justified by the evidence above. + +## Reproduction + +```bash +python3 runs/intervention-response-v0/test_analysis.py +python3 runs/intervention-response-v0/test_p1_analysis.py + +python3 runs/intervention-response-v0/analyze_phase6.py \ + --metrics runs/opprof-phase6/phase6/metrics.json \ + --raw-root runs/opprof-phase6/phase6/solo-authoritative/cells \ + --output runs/intervention-response-v0/phase6-audit.json + +python3 runs/intervention-response-v0/analyze_p1.py \ + --run-root /home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b \ + --manifest /home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/pilot-manifest.json \ + --output runs/intervention-response-v0/p1-audit.json +``` + +## Data sanity + +- Phase 6: action pairs `n=17`, repeat pairs `n=29`, trials `n=66`; MNS + action size min/max `8/32`, `3` distinct; action-state vectors `n=17`, `17` + distinct; streams `n=12`, bytes min/max `12,745,297/52,957,710`, `12` + distinct. +- P1: action pairs `n=12`, repeat pairs `n=24`, trials `n=36`; MNS action + size min/max `48/56`, `2` distinct; efficacy labels `n=12`, min/max `0/1`, + `2` distinct; streams `n=6`, bytes min/max `17,449,143/29,431,988`, `6` + distinct. +- Checked invariants: exact action request hashes and offered loads match; + all `36/36` P1 trials match the manifest; expected pair counts hold; all + deltas are finite; non-negative counters and bounded ratios hold; per-config + state vectors are not all identical; both efficacy classes are present. No + red flags were observed. diff --git a/docs/intervention-response-v1-p1-protocol-20260714.md b/docs/intervention-response-v1-p1-protocol-20260714.md new file mode 100644 index 0000000..9a3e413 --- /dev/null +++ b/docs/intervention-response-v1-p1-protocol-20260714.md @@ -0,0 +1,58 @@ +# Intervention-response v1 prospective-repeat confirmation + +Status: **FROZEN AFTER PHASE-6 V0 FAILURE AND BEFORE P1 RESPONSE ANALYSIS**. + +Date: 2026-07-14 (Asia/Singapore). + +## Why this is a new confirmation, not a relaxed V0 + +Phase-6 V0 failed its frozen global response gate. Its 29 same-config +confirmations were triggered after disputed outcomes, and the resulting noise +sample contains extreme trajectory divergence by construction. V0 remains +failed and its thresholds are unchanged. + +The already-completed P1 campaign supplies a distinct test: three +prospectively scheduled, disjoint repeat bands for every cell/load. TP1 and +TP4 use identical offered loads and exact request-order hashes across their MNS +endpoints. V1 asks whether an MNS response is identifiable against this +prospective workload-repeat noise, and whether that response predicts action +efficacy beyond exact external prefix outcomes. + +P1 is now development data. No result here is held-out or paper-facing. + +## Frozen pairs + +- Action pairs: TP1 `MNS 8 -> 64` and TP4 `MNS 16 -> 64`, at low/high load and + repeat 1/2/3. Endpoints must have identical TP, offered rate, repeat role, + and request-order hash. Expected `n=12`. +- Repeat-noise pairs: consecutive pre-arranged repeat bands within each of six + cells and low/high load: `rep1 -> rep2`, `rep2 -> rep3`. Expected `n=24`. + Repeat bands intentionally contain different requests and therefore include + workload-sampling noise rather than pretending to be identical trials. + Adjacent differences share the middle run; the gate uses their empirical + magnitude only and does not treat the 24 differences as independent samples + for a p-value or confidence interval. +- Prefix horizons: 5 and 10 seconds. Exact monotonic request completion times + and the same Layer-1 intervals are used. + +## Frozen gates + +The response-identifiability thresholds are exactly the Phase-6 V0 thresholds: +75% sign consistency, 2x median effect/repeat noise, and at least 50% of action +deltas above repeat absolute p95. At least two response features must qualify +at both horizons. + +Action efficacy is one only for an infeasible-to-feasible full-run transition. +The 12 action pairs must contain at least four examples of each class. + +For decision relevance, each individual external-outcome response feature and +each individual telemetry-response feature is evaluated by leave-one-repeat- +band-out threshold fitting. This intentionally avoids a multivariate model on +12 examples. At least one telemetry feature must, at both horizons: + +1. reach balanced accuracy at least 0.75; and +2. exceed the best external-outcome response feature by at least 0.15. + +Only if data validity, response identifiability, and incremental decision +relevance all pass does V1 open a newly registered matched GPU pilot. No +threshold or feature is changed after observing V1. diff --git a/docs/intervention-response-v2-phase-aware-protocol-20260714.md b/docs/intervention-response-v2-phase-aware-protocol-20260714.md new file mode 100644 index 0000000..4c72d37 --- /dev/null +++ b/docs/intervention-response-v2-phase-aware-protocol-20260714.md @@ -0,0 +1,94 @@ +# Phase-aware telemetry intervention-response v2 protocol + +Status: **INVALID OPERATIONAL ATTEMPT; SUPERSEDED BY V3**. + +Date: 2026-07-14 (Asia/Singapore). + +The first MNS=16 session timed out while draining the 3.125 requests/s/GPU +workload after its 300-second arrival window. It produced no high-load result, +and no MNS=64 endpoint was run. No comparative conclusion is drawn from this +attempt; see `intervention-response-v3-two-load-protocol-20260714.md`. + +## Correction to v0/v1 + +The 5/10-second analyses tested an ultra-early verifier. They did not test +whether telemetry observed after the engine has developed queue, batch, and KV +state can guide tuning. The P1 replay lasts 60 seconds after time scaling, and +the 5/10-second prefixes contain only a small fraction of its requests. + +V2 therefore replaces absolute cutoffs with replay phase. The old audits and +their negative decisions remain immutable, but their claim is narrowed to the +first 5/10 seconds. + +## Historical corrective audit + +The historical audit is development-only and cannot become confirmatory after +the horizon concern was observed. + +- Infer each trial's intended replay duration as selected requests divided by + offered requests per second. All trials must agree. +- Find every complete 10% replay decile supported by every trial. Analyze all + such deciles; selecting only the best horizon is forbidden. +- At each decile report both: + - cumulative state from replay start to the checkpoint; and + - the non-overlapping 10%-wide state block ending at the checkpoint. +- Report admitted/completed request coverage, response-versus-repeat statistics, + telemetry versus external-outcome efficacy, and per-feature trajectory drift. +- Reuse the frozen v1 action and repeat pairs and the frozen response and + incremental-efficacy thresholds. These thresholds are descriptive in V2; + passing one post-hoc horizon does not open a contribution claim. + +If early stopping prevents complete observation of the replay phases, the +historical decision is `REQUIRES_UNCENSORED_PHASE_AWARE_PILOT`, independent of +which early decile looks best. + +## Uncensored matched pilot + +The pilot is a mechanism gate, not paper evidence. + +- Hardware/engine/model: solo placement on dash0, 4 NVIDIA H20 GPUs, patched + vLLM `0.24.1.dev3+opprof`, Qwen3-30B-A3B, fixed `TP=4`. +- Action: `MNS 16 -> 64`; topology, model, engine build, workload, arrival + sequence, offered load, and all other settings remain fixed. +- Workload: `chat_w20260312_1000` at replay-time scale `0.5`, hence 300 seconds. +- Offered loads per GPU: `1.5`, `2.125`, and `3.125` requests/s. These supply a + low control and the two already-observed P1 pressure regimes. +- Repetitions: three disjoint session bands, exact request sequence matched + across action endpoints. Endpoint order alternates `A/B`, `B/A`, `A/B`; + load order is counter-rotated across repetitions. +- A fresh server receives the accepted long-request warm-up and a bounded + burn-in before each measured session. +- SLO-unrecoverable early stop is disabled. Every run must observe the full + 300-second arrival window; a separate 360-second safety deadline may mark a + run invalid but cannot manufacture a full-run label. +- Cumulative checkpoints: 10%, 25%, 50%, 75%, and 100%, or 30/75/150/225/300 + seconds. Quarter blocks are analyzed separately from cumulative means. +- A measured Layer-1 interval is complete only when its start-boundary, + end-boundary, and maximum internal record gaps are each at most one second; + timestamps must be monotonic. +- Placement is serialized. Co-location remains forbidden because Phase 6 + observed material co-location-induced outcome shifts. +- Hard cap: 8 H20-hours including startup, warm-up, burn-in, invalid attempts, + and cleanup. + +## Gates + +Data validity requires complete 300-second Layer-1 coverage, zero dropped +records, exact request/arrival/length hashes across action endpoints, monotonic +timestamps, full request accounting, idle GPUs before and after each session, +and no co-resident GPU process. + +Mechanism evidence requires at least two telemetry features whose matched action +response exceeds same-config repeat noise at the same pair of consecutive +checkpoints under the unchanged v1 response thresholds. Those features must +also have a consistent direction in at least two of the three load regimes. + +Decision evidence additionally requires both action-efficacy classes and at +least one of the phase-stable mechanism features to reach +leave-one-repetition-out balanced accuracy at least 0.75 and exceed the best +external prefix outcome by at least 0.15 at two adjacent predeclared +checkpoints from 25% onward. Without label balance the pilot can adjudicate +mechanism evidence only. + +No H20 run is launched if the local analyzer/tests, manifest preflight, GPU +probe, command dry-run, projected cost, or cleanup plan fails. diff --git a/docs/intervention-response-v3-results-20260714.md b/docs/intervention-response-v3-results-20260714.md new file mode 100644 index 0000000..75ac723 --- /dev/null +++ b/docs/intervention-response-v3-results-20260714.md @@ -0,0 +1,160 @@ +# Phase-aware telemetry intervention-response v3 results + +Date: 2026-07-14 (Asia/Singapore). + +Decision: **`STOP_NO_INCREMENTAL_TUNING_SIGNAL`**. + +## Claim tested + +After the replay has developed queue, batch, and KV state, does increasing MNS +from 16 to 64 create telemetry responses that exceed workload-repeat noise, and +does any such response identify whether the action repairs the full-run SLO +better than external prefix outcomes alone? + +The first clause passed. The second clause failed. Long-window telemetry is +mechanistically informative, but this pilot does not support its necessity for +tuning this action. + +## Setup + +- dash0 GPU 0-3: four NVIDIA H20 GPUs; Qwen3-30B-A3B; patched vLLM + `0.24.1.dev3+opprof`; TP=4. +- Action: MNS `16 -> 64` with exact request, arrival, and input-length hashes. +- Trace: `chat_w20260312_1000`, replay-time scale 0.5, 300 seconds. +- Loads: 1.5 and 2.125 requests/s/GPU; three disjoint bands; endpoint order + A/B, B/A, A/B; load order low/mid, mid/low, low/mid. +- Five cumulative checkpoints: 30, 75, 150, 225, and 300 seconds; four + non-overlapping quarter blocks. +- Six fresh-server sessions, 12 measured runs, six action pairs, and eight + same-config repeat pairs. + +The prior three-load attempt is not part of the result. Its MNS=16 workload at +3.125 requests/s/GPU could not drain by the 450-second client timeout and +produced no high-load result. V3 reran every retained point from scratch. + +## End-to-end outcome + +All three low-load pairs remained feasible (`true->true`, label 0). All three +pressure-load pairs changed from infeasible to feasible (`false->true`, label +1). MNS=16 pressure pass rates were 0.5604, 0.3145, and 0.2635; all three +MNS=64 pressure runs reached 1.0. This yielded a balanced 3/3 action label set. + +## Mechanism result + +No telemetry feature passed the action-versus-repeat gate at 10% or 25%. At +50%, graph padding first passed. At both 75% and 100%, graph padding and queue +waiting passed, satisfying the requirement for two features at the same pair +of adjacent checkpoints and with consistent directions in both load regimes. + +| Feature | Direction for MNS 16->64 | 75% effect/repeat median | 75% above repeat p95 | 100% effect/repeat median | 100% above repeat p95 | +|---|---:|---:|---:|---:|---:| +| `queue_waiting_mean` | lower | 1346.22 | 3/6 | 898.70 | 3/6 | +| `graph_padding_fraction` | higher | 5.00 | 4/6 | 5.74 | 5/6 | + +The very large queue effect/median-repeat ratios should not be read alone: its +repeat p95 was much larger than its repeat median, so the independent p95 +coverage criterion remained binding. Full-window queue-waiting deltas were +-0.19 to -0.32 at low load and -20.99 to -32.16 at pressure load. Graph +padding increased in every pair, by 0.00133-0.00217 at low load and +0.00705-0.00873 at pressure load. + +The mechanism is therefore a real tradeoff: larger MNS reduces queueing, +especially under pressure, while increasing CUDA-graph padding. + +## Tuning-signal result + +Leave-one-repetition-out balanced accuracy was evaluated against the best +external prefix-outcome feature at every predeclared checkpoint. + +| Replay phase | Best external BA | Best telemetry BA | Incremental telemetry gate | +|---:|---:|---:|---| +| 10% | 0.833 | 0.833 | fail | +| 25% | 1.000 | 1.000 | fail | +| 50% | 0.833 | 1.000 | pass at this checkpoint only | +| 75% | 1.000 | 1.000 | fail | +| 100% | 1.000 | 1.000 | fail | + +At 50%, several telemetry features exceeded the external baseline by at least +0.15, including the phase-stable mechanism feature +`graph_padding_fraction`. The advantage did not hold at either adjacent +checkpoint. Consequently no feature passed the frozen two-adjacent-phase +requirement. + +The important ordering is that external TTFT already classified the action +perfectly at 25%, whereas the robust two-feature mechanism response did not +emerge until 75%-100%. In this setup telemetry explains *why* MNS helps, but it +does not provide earlier or more reliable action selection than direct prefix +outcomes. + +## Research conclusion + +The 5/10-second negative result was indeed too narrow. It only ruled out an +ultra-early telemetry verifier; it did not rule out engine-state information. +The 300-second pilot finds a clear and reproducible queueing-versus-padding +response. + +However, this does not rescue the direct telemetry-guided tuning claim. For +this action and workload, the external signal is already as good or better +before the telemetry mechanism becomes stable. The project should therefore +not claim that engine instrumentation is necessary for tuning on this evidence, +and should not open an E2E policy test from this pilot. + +The narrower simulator-residual route remains logically open: telemetry may +explain why a simulator misranks real configurations even when direct online +outcomes can guide a tuner. That is a different hypothesis and was not tested +here. + +## Change and verification + +Change: absolute 5/10-second prefixes were replaced by phase-aware 30/75/150/ +225/300-second analysis; SLO early stop was disabled; full Layer-1 coverage, +hash, request-accounting, controller, and stream/footer gates were added. The +mechanism gate was corrected to require two features at the same adjacent +phase pair. + +Expected effect: distinguish “telemetry has not developed yet” from “telemetry +does not identify or improve the action decision.” + +Verification: five local analysis/controller test suites passed; remote +manifest preflight and command dry-run passed; six serialized sessions passed +all stream invariants; the analyzer was rerun and produced byte-identical +output. + +Result: mechanism evidence passed; incremental tuning evidence failed. Audit +SHA256: `45f6f248712f9cbd3ed72036837ff6dc5b5c14c0f2eb6ba5cd5daceb1aa4ddb7`. + +Remaining risk: this is a development pilot with one model, one TP, one action, +two retained loads, three request bands, and six action labels. It is adequate +to reject opening the next direct-policy stage, not to establish a universal +negative claim about telemetry. + +## Research-validity audit + +| Check | Verdict | Evidence / boundary | +|---|---|---| +| Real system and E2E outcome | PASS | Real H20/vLLM replay; full SLO outcome accompanies mechanism telemetry. | +| Matched action baseline | PASS | Exact request/arrival/length hashes for MNS 16 and 64; external prefix outcome is the decision baseline. | +| Repeats and order effects | PASS for pilot | Three disjoint bands; A/B, B/A, A/B endpoint order; counter-rotated load order. | +| Selective load removal | PASS with narrowed claim | The 3.125 load produced no result before any action comparison; the failure and cost are retained, and all kept points were freshly rerun. | +| Significance/generalization | NEEDS EVIDENCE for a paper claim | Only three bands, one model, one TP, one action, and six labels. This is explicitly a stage gate. | +| Calibration versus evaluation | NEEDS EVIDENCE for a positive policy claim | Frozen gates and leave-one-band-out folds reduce leakage, but a new workload/model hold-out is still required. | +| Platform/reproducibility | PASS | Commit, commands, manifest, controller state, platform fingerprint, raw remote paths, and audit hashes are recorded. | + +## Data sanity + +- Measured runs: n=12; elapsed 300.346-317.012 seconds; 12 distinct; pass rate + 0.2635-1.0 with 4 distinct values; selected requests 1800-2550 with 2 + distinct values. +- Sessions: n=6; 0.8413-0.8631 H20-hours; 6 distinct; Layer-1 records + 58,465-64,776; 6 distinct. V3 cost was 5.0924 H20-hours; total including + the invalid attempt was 6.4505, below the 8.0 cap. +- Labels: n=6; min/max 0/1; 2 distinct. Action pairs were 6 and repeat pairs + were 8 at every checkpoint. +- Coverage-gap observations: n=60; start gaps 0.0427-0.1247 seconds; end gaps + 0.00014-0.1705; maximum internal gaps 0.1695-0.6528, all below one second. +- Checked invariants: exact pair hashes and counts, all runs uncensored, full + request accounting, monotonic admitted/completed coverage, monotonic Layer-1 + timestamps, nonnegative counters, bounded ratios, non-identical per-config + states, contiguous step indices, zero drops, footer/sidecar agreement, no + controller failures, all sessions complete, GPU idle after completion. No + red flags were found. diff --git a/docs/intervention-response-v3-two-load-protocol-20260714.md b/docs/intervention-response-v3-two-load-protocol-20260714.md new file mode 100644 index 0000000..f5f83b7 --- /dev/null +++ b/docs/intervention-response-v3-two-load-protocol-20260714.md @@ -0,0 +1,73 @@ +# Phase-aware telemetry intervention-response v3 protocol + +Status: **FROZEN AFTER A NON-COMPARATIVE OPERATIONAL FAILURE AND BEFORE V3 RUNS**. + +Date: 2026-07-14 (Asia/Singapore). + +## Why v2 was invalid + +The first v2 session completed the 300-second arrival windows at 1.5 and 2.125 +requests/s/GPU. At 3.125 requests/s/GPU, MNS=16 could not drain the admitted +requests before the 450-second client timeout. The session produced no result +and no MNS=64 action endpoint was run. V2 is therefore an invalid operational +attempt, not evidence for or against the telemetry hypothesis. + +This failure was observed before any MNS action comparison. V3 excludes only +the unmeasurable overload point and reruns every retained point on fresh +servers; it does not reuse the completed v2 low/mid results. + +## Question and hypothesis + +Question: after enough replay time for queue, batch, and KV state to develop, +does an MNS intervention create telemetry responses that exceed workload-repeat +noise, and does any such response predict whether the intervention repairs the +full-run SLO outcome better than external prefix outcomes alone? + +Hypothesis: increasing MNS from 16 to 64 has little value at the 1.5 +requests/s/GPU control load but can repair the 2.125 requests/s/GPU pressure +load. Queue, running-set, batch, or KV telemetry should expose the difference +at stable replay phases. Label balance is an assumption to test, not a fact. + +## Frozen setup + +- Solo placement on dash0 GPU 0-3: 4 NVIDIA H20 GPUs, Qwen3-30B-A3B, patched + vLLM `0.24.1.dev3+opprof`, fixed TP=4. +- Action: MNS `16 -> 64`; all other engine and workload parameters fixed. +- Workload: `chat_w20260312_1000`, replay-time scale 0.5, hence 300 seconds. +- Loads per GPU: 1.5 control and 2.125 pressure requests/s. The failed 3.125 + overload point is excluded from V3 and retained only as a failure artifact. +- Three disjoint request bands. Each MNS action pair has exact request, + arrival, and input-length hashes. Endpoint order is A/B, B/A, A/B; load + order is low/mid, mid/low, low/mid. +- Every session starts a fresh server, then runs the accepted 16-request long + warm-up and bounded burn-in before measured runs. +- SLO-unrecoverable early stop is disabled. Measured results must cover the + full 300-second arrival window and must not be early-stopped. +- Cumulative checkpoints are 10%, 25%, 50%, 75%, and 100%; non-overlapping + quarter blocks are also reported. +- A Layer-1 interval is complete only if timestamps are monotonic and its + start, end, and maximum internal record gaps are each at most one second. +- Incremental V3 cap is the unused portion of the original 8 H20-hour cap. + The exact prior cost and V3 cap are machine-recorded in the manifest. + +## Frozen gates + +Data validity requires six uncensored sessions, six action pairs, eight +same-config repeat pairs, exact action-pair hashes, full request accounting, +zero Layer-1 drops, continuous coverage, all stream/footer invariants, no +co-resident compute process, idle GPUs before and after sessions, nonnegative +counters, bounded ratios, non-identical per-config state, and monotonic request +coverage across checkpoints. Any red flag stops analysis. + +Mechanism evidence requires at least two telemetry features to exceed the +unchanged v1 repeat-noise thresholds at the same pair of adjacent checkpoints. +Those features must have a direction consistent in both retained load regimes. + +Decision evidence additionally requires at least two positive and two negative +full-run action-efficacy labels, valid leave-one-repetition-out folds, and at +least one phase-stable mechanism feature whose balanced accuracy is at least +0.75 and at least 0.15 above the best external prefix-outcome feature at two +adjacent predeclared checkpoints from 25% onward. + +V3 remains a development mechanism pilot. Even `OPEN_E2E_POLICY_TEST` opens a +held-out tuning-policy experiment; it is not itself a paper performance claim. diff --git a/docs/telemetry-residual-tuning-roadmap-20260714.md b/docs/telemetry-residual-tuning-roadmap-20260714.md new file mode 100644 index 0000000..ba6ceab --- /dev/null +++ b/docs/telemetry-residual-tuning-roadmap-20260714.md @@ -0,0 +1,286 @@ +# Telemetry-conditioned residual tuning roadmap + +Status: **R0 COMPLETE / FAILED; R1 AND R2 CLOSED FOR THIS MODEL**. + +Date: 2026-07-14 (Asia/Singapore). + +## Research question and claim boundary + +The question is whether a small number of real engine observations can correct +a simulator's task-specific error over **unmeasured configurations**, and +whether that correction reduces the real-GPU cost of finding a high +SLO-goodput serving configuration. + +The intended headline claim, if the evidence supports it, is: + +> An engine-state-conditioned residual model turns a simulator prediction into +> a task-specific posterior over unmeasured serving configurations, allowing a +> sequential tuner to reach near-oracle SLO-goodput with materially fewer +> H20-hours than simulator-only and outcome-only tuning. + +Classification accuracy, simulator-error diagnosis, and telemetry overhead are +supporting evidence. None is an end-to-end tuning contribution by itself. + +The following method is closed and will not be revived under another name: +per-candidate five-second accept/reject as the headline contribution. The P1 +result showed only 1.426% cost reduction in the frozen `k=2` workflow. + +## Two models, one evaluation + +Both branches use the same legal candidate set, real measurements, task split, +cost accounting, and acquisition function. + +### Simulator-residual branch (primary) + +For measured anchor `c_t` and unmeasured candidate `c'`: + +```text +y_hat(c') = y_real(c_t) + + [y_sim(c') - y_sim(c_t)] + + f(state_real(c_t) - state_sim(c_t), c' - c_t, workload, SLO) +``` + +The simulator delta is the prior. The learned model may correct it only with +training-supported state/config transitions; uncertainty or distribution shift +must shrink the correction back toward the simulator prior. + +### Telemetry-only branch (mandatory) + +```text +y_hat(c') = y_real(c_t) + + g(state_real(c_t), c' - c_t, workload, SLO) +``` + +This branch tests whether the simulator is actually necessary. It does not +use a hand-authored bottleneck-to-knob rule. + +### Search policy + +Legal configurations are enumerated independently of telemetry. A generic +cost-aware acquisition rule ranks candidates from predicted improvement, +uncertainty, and measured H20 cost. The current production harness's +bottleneck scores, topology-first ordering, and hand-set relief constants are +not consumed by either branch. The validator may enforce legality, +full-config no-repeat, failure accounting, and resource caps only. + +## Hypotheses + +| ID | Hypothesis | Direct test | Failure meaning | +|---|---|---|---| +| H0 | Existing artifacts can express a common, direct-measurement state without heuristic labels. | Engine/simulator extractor coverage and invariants. | Route is not currently implementable. | +| H1 | Simulator errors are predictable from engine/simulator state discrepancy at measured anchors. | Task-held-out pairwise inversion correction and new-inversion rate. | Telemetry is diagnostic but cannot correct the surface. | +| H2 | Telemetry alone predicts useful config transitions beyond outcome-only history. | Telemetry-only versus real-outcome-only sequential replay. | Direct telemetry-guided tuning has no independent value. | +| H3 | Residual correction changes actual tuning decisions and cost. | H20-hours to 95% oracle and regret AUC against the strongest safe baseline. | No system contribution even if H1/H2 prediction metrics improve. | + +## Common-state contract + +Only directly observed or exactly reconstructed quantities are admitted. + +| Quantity | vLLM Layer-1 | Frontier | R0 status | +|---|---|---|---| +| Scheduled requests / batch size | Per scheduler step | Existing per-batch metric, disabled in P1 output | Common after CPU replay | +| Scheduled prefill/decode tokens | Per scheduler step | Existing per-batch metrics | Common after CPU replay | +| Scheduler/batch rate | Monotonic step timestamps | Batch count / simulated duration | Common after CPU replay | +| Waiting queue area | Time-weighted queue gauge | Sum of request waiting times | Common aggregate | +| Running request area | Time-weighted running gauge | Sum of E2E minus waiting time | Common aggregate, semantics audited | +| Preemption count | Per step | Per request | Common | +| KV usage/headroom | Exact blocks and ratio | Not in committed output | Engine-only until exact reconstruction exists | +| CUDA graph mode/padding | Exact per step | Not modeled | Engine-only omitted-mechanism signal | +| Request TTFT/TPOT/pass rate | Exact real outcomes | Exact simulated request metrics | Common outcome, not state | + +Unavailable fields remain null. They cannot be imputed from a human +`prefill/decode/queueing` label. + +Frontier already contains the required detailed batch and timestamped +stage-batch ledger output. P1 disabled it for artifact size. R0 replays the +same immutable fixtures with the existing output flags enabled; it does not +change the simulator model or calibration. + +## Data separation + +- Phase 6 / `chat_w20260311_1000`: development only. +- P1 / `chat_w20260312_1000`: development only. +- R1 / `chat_w20260313_1000`: new development surface. +- R2: trace windows not used for feature, model, threshold, candidate-space, + cutoff, or acquisition decisions. +- Splits are by complete workload/SLO task. Anchor- or pair-level random + splits are prohibited. +- Sequential-policy seeds measure algorithmic variability; they are not + counted as independent system tasks. + +The two existing development tasks have an important limitation: the now- +available SLO-gated simulator reading already retains the real oracle at its +top rank/tie. They therefore cannot establish a positive end-to-end ranking +claim. They are used for plumbing, known false-feasible cases, and negative +evidence. R1 must be run as an unbiased complete surface, not selected after +observing simulator success or failure. + +## Step-by-step roadmap + +### R0.1 — Inventory and roadmap + +Deliverables: + +- this roadmap; +- rolling untracked `ONGOING.md`; +- exact engine/simulator field and artifact inventory. + +Gate: every claimed input has an authoritative file path and provenance. + +### R0.2 — Common-state plumbing + +Deliverables: + +- `runs/telemetry-residual/common_state.py`; +- synthetic correctness tests; +- one exact P1 Frontier replay with individual batch metrics and the full + stage-batch ledger enabled; +- paired engine/simulator state summary for the same fixture. + +Gate: + +- replay request count and SLO scorer exactly agree with the committed replay; +- batch/ledger outputs are non-empty; +- all counters are non-negative, ratios bounded, times monotonic; +- no GPU is visible to Frontier; +- output volume is practical before expanding to twelve replays. + +### R0.3 — Development residual/headroom audit + +Use all frozen P1 primary fixtures and corresponding engine intervals. Produce: + +- common-state residuals per anchor; +- simulator-error labels and continuous SLO/goodput residuals; +- ordered source/target diagnostic that removes both config identities from + both roles in every training fold; +- oracle upper bound for cross-candidate correction; +- explicit comparison with simulator+outcome and telemetry-only features. + +R0 is a feasibility gate, not headline evidence. Proceed to R1 only if: + +1. state features are collected with the measured source anchor, vary across + cells, and are available before any target config is evaluated; +2. at least one known simulator error has a state discrepancy not exposed by + the matched external prefix outcome; +3. a prior-preserving model can correct development errors without introducing + a larger number of new errors under regularization sensitivity; +4. an oracle cross-candidate correction has at least 15% sequential tuning-cost + headroom under full startup/warm-up accounting. + +### R0 result and decision + +R0 completed without a data-validity red flag, but failed condition 3. The +decision is **STOP_BEFORE_R1**; no H20 job was launched for this route. + +- All 12 detailed Frontier CPU replays exactly reproduced their committed SLO + scorers. Runtime was 23.943--54.786 seconds per replay, detailed artifacts + were 4.12--13.53 MB, CUDA visibility was empty, and there were zero failures. +- The paired surface contains 12 real/sim anchors, two known simulator + false-feasible anchors, and 120 legal cross-config ordered transitions. A + fold removes both the source and target TP/MNS identity from source and + target roles; the two offered-load anchors remain part of the same task. +- Raw Frontier feasibility is 83.33% on the repeated transition view. The + structurally correct hybrid model uses + `r_target = r_source + delta_r`; the direct model uses + `y_target = y_source + delta_y` and never reads simulator fields. +- Direct telemetry is not robust relative to real-outcome-only: its accuracy + delta over L2 `{0.1,1,10,100}` is `{-0.83,+1.67,0,-4.17}` percentage points, + and its best absolute accuracy is 54.17%, below the raw simulator's 83.33%. +- Hybrid telemetry raises classification accuracy over the corresponding + simulator+outcome transition regression by 1.67--4.17 percentage points, + but worsens pass-rate RMSE by 0.141--0.201 and MAE by 0.084--0.125. Its full + correction reaches only 46.67--53.33% absolute accuracy. +- Across 24 nonzero `(L2, raw-simulator-prior weight)` combinations, no model + both corrects an existing simulator error without more new errors and avoids + worsening RMSE/MAE. Whenever a correction fixes at least one error, it + corrupts at least 11 previously correct transitions. +- A perfect correction could skip the frozen simulator rank-2 real final and + save 0.043469 H20-hours: 15.45% of the prospective online `k=2` cost, or + 14.40% when the prior failed launch is charged. On this development task the + simulator top-1 already is the real oracle with zero regret, so headroom + versus the observed-safe top-1 baseline is 0%. + +The result does not prove that engine telemetry is useless. It shows that the +current one-task anchor-transition evidence cannot support either a safe +simulator-residual tuner or a simulator-free telemetry tuner. A larger model +or an R1 run would add capacity/data after a failed gate and is therefore not +authorized under this roadmap. + +### R1 — New development surface + +Status: **NOT LAUNCHED; CLOSED BY R0**. + +Frozen starting setup: + +- host: dash0, eight NVIDIA H20 GPUs; +- cells run solo; no co-location for SLO verdicts; +- patched vLLM 0.24.1.dev3, Qwen3-30B-A3B BF16; +- trace: `chat_w20260313_1000`; +- output tokens: exactly 128; +- SLO: stepped TTFT 2/4/6 seconds, TPOT 50 ms, pass rate at least 0.95; +- config surface: TP `{1,2,4}` × MNS `{8,16,32,64}`; +- hard campaign cap: 4 H20-hours. + +The load ladder, repetitions, randomized order, exact commands, expected wall +time, and artifact paths are frozen only after R0. A resolved echo is required +before launch. + +R1 passes only if a frozen sequential replay shows at least 15% E2E H20-hour +headroom over the strongest safe baseline with final regret at most 5%. R1 is +development evidence and cannot be reported as the held-out result. + +### R2 — Held-out sequential tuning + +Status: **NOT LAUNCHED; CLOSED BY R0**. + +Required baselines: + +1. random search; +2. real-outcome-only Bayesian/sequential search; +3. Frontier ranking plus real top-k final; +4. simulator plus real-outcome residual; +5. telemetry-only transition tuner; +6. simulator plus telemetry residual tuner; +7. complete real surface as oracle, not as a cost competitor. + +Primary metric: end-to-end H20-hours to first reach 95% of the real full-surface +SLO-goodput oracle. Secondary metrics are cost-normalized regret AUC, final +regret at fixed budgets, oracle false-prune, wall time, and per-task regressions. + +The route is successful only if the winning telemetry method reduces the +primary cost by at least 20% versus the strongest safe baseline and ends within +5% regret on every headline task. If hybrid beats telemetry-only by at least +10%, simulator residual correction is the primary method. If telemetry-only +is within 5% or better, the simulator dependency is removed. If neither clears +the contribution bar, the route is closed and telemetry remains a diagnostic +facility only. + +## Cost discipline + +- R0 simulator work is CPU-only and must set empty CUDA visibility. +- R1 cannot exceed 4 H20-hours. +- R2 receives no budget until R1 passes. +- Startup, warm-up, burn-in, failed launches, real probes, continuation, and + final validation are charged. Benchmark-only annotation repeats are + reported separately and cannot disappear from campaign accounting. + +## Final R0 sanity block + +| Data | n | Min | Max | Distinct | Checked invariant | +|---|---:|---:|---:|---:|---| +| Phase 6 cells | 12 | TP1/MNS8 | TP4/MNS64 | 12 | Surface not identical; solo SLO tier authoritative | +| Phase 6 Layer-1 primary steps | 37 streams | 343 | 12,103 | 37 | Contiguous; zero drops | +| P1 primary anchors | 12 | infeasible | feasible | 2 labels | 7 feasible / 5 infeasible | +| P1 Frontier runtime | 12 | 24.093 s | 54.575 s | 12 | CPU-only; zero failures | +| Detailed Frontier replay runtime | 12 | 23.943 s | 54.786 s | 12 | Exact committed scorers; CUDA hidden | +| Detailed artifact bytes | 12 | 4,123,724 | 13,527,776 | 12 | Non-negative; practical CPU replay size | +| Cross-config transitions | 120 | real pass 0.1067 | real pass 1.0 | 6 outcomes | Both endpoint config identities held out | +| State residual vectors | 12 | 16 fields | 16 fields | 12 vectors | Finite; no missing common field | +| R0 E2E cost values | 4 | 0.237914 | 0.301935 H20-h | 4 | Non-negative; `k=1/2`, online/conservative | + +Checked invariants: non-negative counts and costs; pass rates in `[0,1]`; +simulator results not all identical; exact request count/hash agreement; Layer-1 +step continuity and zero drops; no co-resident SLO measurements; no calibration +or evaluation split reuse for a future headline claim. No current red flag +invalidates R0 plumbing. The R0 tuning gate itself failed because safe +prior-preserving correction was absent. diff --git a/docs/tuning-core-challenges-cost-audit-20260715.md b/docs/tuning-core-challenges-cost-audit-20260715.md new file mode 100644 index 0000000..200bdae --- /dev/null +++ b/docs/tuning-core-challenges-cost-audit-20260715.md @@ -0,0 +1,391 @@ +# AITuner tuning：核心挑战、统一成本口径与研究路线 + +日期：2026-07-15（Asia/Singapore） + +状态：**问题定义与历史成本审计完成；新的 tuner 贡献尚未建立。** + +## 结论先行 + +我们不应该把 tuning 定义成“根据当前 telemetry 判断哪个 cap 满了，再调对应 knob”。这个定义同时遗漏了 knob interaction、反事实识别、实验成本和跨任务失配。更准确的问题是： + +> 给定模型、engine version、hardware、workload、SLO 和一个声明好的合法配置空间，tuner 如何用最少的真实 GPU 成本，依次选择可能包含多个 knob 的 intervention，找到 SLO-goodput regret 不超过 `epsilon` 的配置？ + +AITuner 可以形成的系统贡献应当是： + +> **一个 intervention-calibrated、action-conditioned、cost-aware 的 tuner：它从真实 engine trajectory 和已测 intervention 中学习联合 config action 的反事实收益分布，并以 cost-to-oracle 而非规则命中率作为目标。Harness 只负责实验语义、合法性、配对、记账和可复现性，不负责用人工 bottleneck rule 决定 action。** + +现有结果支持这个问题值得做，但不支持宣称它已经解决： + +- 在真实 `TP x MNS` surface 上，one-knob-at-a-time 会停在比 oracle 低 **25.6%** 的 coordinate-wise local optimum。 +- 在 action-aware pilot 中，增加 MBBT 在“几乎从未独占打满 MBBT cap”的情况下仍把 source goodput 提高 **48.0%--77.1%**；因此 `cap -> knob` 不是完整模型。 +- 同一 dash0 任务上，当前 guided harness 到 5% empirical regret 只比纯 LLM 少 **5.85%** H20-hours；到 2% regret 则少 **61.09%**。这说明必须比较完整 cost--regret curve，不能只比较最终最好值。 +- Frontier 的 decision-bearing throughput top-1 在 12-cell surface 上有 **30.46%** real regret。Simulator 本身的边际 GPU cost 是 0，但通过 real-final 恢复 oracle 需要 tie-expanded 4 个真实 cell，即 **0.7828 reconstructed H20-hours**。 + +## 1. Tuning 问题和成功标准 + +固定 task context： + +```text +T = {model, engine build, hardware, workload/trace, SLO, legal config space C} +``` + +每个完整配置 `c in C` 的目标为： + +```text +f_T(c) = max request_rate_per_gpu + subject to request SLO pass rate >= target +``` + +有限空间 oracle 为： + +```text +f*_T = max_{c in C} f_T(c) +regret(c) = 1 - f_T(c) / f*_T +``` + +顺序 tuner 在第 `t` 步基于历史 `D_t` 选择一个完整 config intervention： + +```text +a_t = c_t -> c_{t+1} +``` + +成功不是“最后找到一个不错的值”，而是同时满足： + +1. `regret(best_t) <= epsilon`； +2. 达到该点之前的 all-in H20-hours 最小； +3. launch、correctness、SLO 和失败率约束不退化； +4. 结论在 held-out task 上成立，而不是在用于设计规则的 task 上成立。 + +### 1.1 GPU cost 的统一定义 + +未来实验的 task-marginal cost 应定义为： + +```text +C_task = sum_j allocated_GPU_count_j + * (GPU_idle_or_release_time_j - allocation_start_time_j) +``` + +它包括 method 实际触发的 startup、warm-up、prefix/full replay、confirmation、failure、cleanup；如果 LLM 思考期间 GPU 仍被占用，也计入。Simulator/模型的一次性 onboarding 成本单独报告： + +```text +C_e2e(N tasks) = C_profile_or_training / N + C_task +``` + +另外报告 CPU-hours、LLM API latency/cost，但不把它们伪装成 GPU-hours。构建 benchmark oracle 的 exhaustive annotation cost 是公共评测成本，单独报告，不计入任何方法；同时可给一个将其等量加回所有方法的 conservative view。 + +历史记录没有 allocation start/release timestamp。本次只能从每个 `engine.log` 的首末时间戳重建： + +```text +C_engine_lower_bound = parallel_size * engine_log_span / 3600 +``` + +因此下面所有历史 H20-hour 数字都是 **engine-lifetime lower bound**，不是 all-in cost。尤其 simulator 的一次性 H20 operator profiling 成本没有记录，不能称为完全免费。 + +### 1.2 两种 oracle 必须分开 + +- **Exact finite-surface oracle**：声明好的 12-cell `TP x MNS` 空间全部真实测量，oracle 是 `TP2/MNS32 = 3.2833 req/s/GPU`。 +- **Broader empirical reference**：dash0 两个 sequential run 中观察到的最好值 `3.35 req/s/GPU`。它包含 surface 外的 MBBT/chunk/GMU action，但只是 best observed，不是全局 oracle。 + +不能把 empirical best 写成 global oracle，也不能让每个方法使用不同的 oracle 定义。 + +## 2. 现有方案的 cost-to-oracle 审计 + +可复算输入和完整结果在： + +- `runs/tuning-cost/manifest.json` +- `runs/tuning-cost/analyze.py` +- `runs/tuning-cost/metrics.json` + +### 2.1 严格同任务对照：纯 LLM vs 当前 guided harness + +两组均为 dash0、Qwen3-30B-A3B、community-vLLM 0.20.0、8xH20 可见、`chat_w20260311_1000`、input 0--8k、output 128、replay scale 0.1、TTFT 2/4/6s、TPOT 50ms、pass rate 0.95。除 tuner method 和服务端口外，固定 task spec 相同。 + +Reference 是两组中 best observed `3.35 req/s/GPU`： + +| Method | 到 <=5% regret | 到 <=2% regret | 到 <=1% regret | 完整 run 成本 | 最终 best | +|---|---:|---:|---:|---:|---:| +| Pure LLM, no harness | 0.2847 H20h，trial 2，regret 2.736% | 1.1458，trial 6，regret 1.493% | 1.3719，trial 7，regret 0% | 2.2825 | 3.35 | +| Guided harness v2 | 0.2681 H20h，trial 2，regret 2.736% | 0.4458，trial 3，regret 1.990% | 未达到 | 0.6231 | 3.30，regret 1.493% | + +直接结论： + +- 5% endpoint：guided 比 pure LLM 少 **5.85%**，不是 material contribution。 +- 2% endpoint：guided 比 pure LLM 少 **61.09%**，有明显 headroom signal，但只有一个 task，不能外推。 +- Pure LLM 在 trial 7 已找到 best observed，之后又花了 `2.2825 - 1.3719 = 0.9106 H20h` 而没有改进，说明 trustworthy stopping 本身就是成本来源。 +- Pure LLM 的 trial 3 使用当前 binary 不支持的 `--expert-parallel-size` 并在 launch 前失败。当前 harness 的 legality/version contract 有实际价值，但它仍不是性能 action-ranking 贡献。 + +### 2.2 Simulator：零边际 GPU cost 不等于零 tuning cost + +Frontier fidelity suite 在 CPU 上执行 184 个 simulation，耗时 **2.055 CPU-hours**，simulation 本身为 0 marginal H20-hours。其对应的 exact dash1 12-cell real surface annotation lower bound 为 **3.5953 H20-hours**。 + +Decision-bearing `frozen-calibrated/throughput-proxy`： + +| Policy | Real cells evaluated | Real-final H20h lower bound | Selected real regret | +|---|---:|---:|---:| +| Simulator-only top-1 | 0 | 0 | **30.46%**，选 TP1/MNS64 | +| Throughput top-1 + real final | 1 | 0.1353 | **30.46%** | +| Throughput top-2 + real final | 2 | 0.2672 | **30.46%** | +| Throughput nominal top-3 + real final | tie-expanded 4 | 0.7828 | 0%，找到 TP2/MNS32 | + +Post-hoc `SLO-gated` reading 把 `{TP2/MNS32, TP2/MNS64}` 放在 top tie bucket；测两个 cell 需 **0.5156 H20h** 并能找到 oracle。但它不是 preregistered decision-bearing policy，而且 anchor verdict 中有 21 个 false-feasible、7 个 false-infeasible，只能作为诊断上界，不能反写成 prospective simulator 结果。 + +Pure LLM/harness 数据来自 dash0，simulator exact surface 来自 dash1。模型、engine、trace、GPU type 匹配，但 host 和 campaign 不同。因此两块内部可以直接比较，跨块只能做 development-level 指示；paper 结论必须在同 host、同 task execution protocol 下重跑。 + +### 2.3 我们要达到的成本目标 + +在当前 reconstructed lower-bound 口径下，一个有意义的单任务 development bar 是： + +| Endpoint | 当前最强同任务 baseline | 20% reduction bar | 兼顾 post-hoc sim+real 的 30% bar | 暂定目标 | +|---|---:|---:|---:|---:| +| <=5% empirical regret | guided 0.2681 | 0.2144 | 0.3609 | **<=0.2144 H20h** | +| <=2% empirical regret | guided 0.4458 | 0.3567 | 0.3609 | **<=0.3567 H20h** | + +这两个数字不是 paper result，只用于检查 proposed method 是否有足够 headroom： + +- 5% endpoint 已经由 baseline + TP2 两个完整 trial 达到。任何必须先跑 source 再跑 target 的 telemetry tuner 都不能靠减少 trial count 获得 20% 优势；它必须能够 one-shot warm-start、跳过 baseline，或安全地缩短其中一次测量。 +- 2% endpoint 有更合理的结构性空间：从一个 source 直接选择 joint `TP2 + MBBT/chunk` target，可能跳过当前中间 trial；如果仍按当前三次完整 trial 顺序执行，就不会达到 bar。 + +Paper-facing gate 不使用这些跨 campaign 绝对数，而使用 prospective same-host all-in cost：在每个 held-out task 上 regret <=5%，相对最强 safe outcome-only/current harness 至少省 20%，相对 frozen simulator+real 至少省 30%，并报告 task-level paired confidence interval。 + +## 3. 四个最核心的 tuning challenge + +### Challenge 1：响应面是联合、条件化且 regime-dependent 的 + +#### 问题本质 + +一般情况下： + +```text +f(c) != base + sum_k effect_k(c_k) +``` + +一个 knob 的 effect 是当前完整 context 的函数： + +```text +Delta_x(c, workload, engine state) +``` + +它可能随 topology、另一个 runtime knob、load、SLO 或 engine version 改变大小甚至改变符号。因此不能先分别求每个 knob 的最优值再 merge，也不能固定一个低质量 context 去判断另一个 knob。 + +#### 已有真实证据 + +在 C1 12-cell real surface： + +- `MNS 8 -> 32` 在 TP1/TP2/TP4 下分别提升约 **8.7% / 44.3% / 90.3%**。 +- 从同一 `TP1/MNS8` 起点，先 tune MNS 再 TP 会停在 `TP4/MNS16 = 2.4417`；该点沿任一单维都没有 strictly improving move，但 joint/global surface oracle `TP2/MNS32 = 3.2833` 高 **34.5% relative to the local point**，即 local point 对 oracle 有 **25.6% regret**。 +- C3 中 `MBT 256 -> 384` 的 effect 根据 topology/MNS 从 0 到约 -9.2%；`MNS 64 -> 128` 从 0 到约 +10.1%。 +- Action-aware Regime A 中 MBBT 几乎从不作为 exclusive cap，但 MBBT action 仍把 source goodput 提高 48.0%--77.1%。它通过 chunk size、prefill packing 和 scarce MNS slot residency 的联合变化获得收益。 + +这直接否定两类通用策略：OAT/coordinate greedy，以及 `which cap is full -> tune that knob`。 + +#### Tuner 必须具备的能力 + +- Action 的基本单位是完整 `config delta`，允许 sparse joint action，而不是孤立 knob/value。 +- 对 topology/runtime family 使用 crossed anchors 或信息增益设计，主动测 interaction；不是默认所有 interaction 都强。 +- 能从数据判断 task 是 topology-dominant、runtime-interaction-dominant 还是 flat/noisy，并据此分配实验，而不是把固定 search order 写进规则。 + +### Challenge 2：当前状态是 observational signal，tuning 需要 counterfactual identification + +#### 问题本质 + +一次 telemetry trace 只能告诉我们： + +```text +P(engine trajectory | current config, workload) +``` + +Tuning 真正需要的是： + +```text +P(Delta SLO-goodput, failure, cost + | source trajectory, proposed full-config action) +``` + +Queue、KV、padding、split prefill 等状态既可能是原因，也可能是 workload/config 的结果。看见某种状态，不等于知道哪个 action 能修复它。一个 action 也可能同时改变多条机制；例如 MBBT 同时改变总 token budget、per-request chunk 和 multi-request packing，现有 telemetry 的解释是 mechanism-consistent，不是已完成的 causal decomposition。 + +#### 已有真实证据 + +- 5/10 秒 telemetry 确实太短；300 秒 phase-aware experiment 中，MNS action 的 queue/padding 机制直到 replay 75%--100% 才稳定出现。 +- 但 external TTFT outcome 在 25% 已完美区分该 action 是否修复 SLO。Telemetry 解释了 why，却没有比 outcome 更早或更可靠地指导 tuning。 +- 3.125 req/s/GPU 的 source 无法在 timeout 内 drain；另一组 source 已达 offered ceiling 的 99.1%--100%，数学上不可能通过 10% improvement gate。没有 exposure/headroom 和 censoring control，模型学到的不是 action response。 +- Same-config repeats 与 matched intervention 的波动不可忽略；只比较两个未经配对的 run 会混入 arrival/order/warm-state noise。 + +#### Tuner 必须具备的能力 + +- 训练样本必须是 exact-workload paired intervention：`(source trajectory, action) -> target delta`，保留失败和 censoring。 +- 使用 phase-binned continuous trajectory，而不是人工 bottleneck label 或 threshold rule。 +- 输出 response distribution 和 uncertainty；证据不足时 abstain，而不是强行给 diagnosis。 +- Telemetry 的价值必须通过同 cutoff、同 model capacity 的 outcome-only ablation 证明。若不能降低 end-to-end H20-hours，instrumentation 只保留为 debugging/解释工具。 + +### Challenge 3：这是异构成本下的 sequential experimental design，不是静态 ranking + +#### 问题本质 + +每个 trial 的成本不同：TP4 是 TP1 的四倍 GPU multiplier，startup/warm-up 可能主导短 probe，失败也有成本；同时 tuner 不知道 oracle，只能在 exploitation、information gain 和 cost 之间权衡。选对 top-1 的 accuracy 不能代表 tuning 效果。 + +必须回答三个连续问题： + +1. 下一次测哪个联合 action？ +2. 测多久，何时 continuation/confirmation？ +3. 什么证据允许停止，并声称 best 已在 `epsilon` 内？ + +#### 已有真实证据 + +- Pure LLM 达到 best observed 后仍浪费 0.9106 reconstructed H20h。 +- Simulator top-1 虽然 0 marginal GPUh，却因 rank error 损失 30.46%；real-final 的 k 增大又迅速增加 H20h。 +- 5% endpoint 上两个方法都只需两个 trial，selection-count headroom 很小；2% endpoint 才暴露 action quality 和 stopping 的巨大差异。 +- Prefix 不是天然便宜：如果 startup、warm-up 和稳定状态形成占主要成本，缩短 replay window 未必带来等比例 H20h reduction。 + +#### Tuner 必须具备的能力 + +- Acquisition 直接优化 expected regret reduction / predicted H20 cost，并把 failure probability 纳入约束。 +- 在 run 前做与 tuning policy 分离的 workload admissibility check：避免 outcome ceiling、无法 drain、无请求或 measurement cap。 +- 使用 uncertainty-aware continuation 和 stop；stop criterion 针对声明的 candidate set 中“仍存在 >epsilon improvement 的概率”，而不是连续几次没提升。 +- 主结果报告 H20-hours-to-5%/2%/1%、fixed-budget regret 和 cost-normalized regret AUC，不 metric shopping。 + +### Challenge 4：任何 mechanism model 都有 fidelity 和 transfer boundary + +#### 问题本质 + +Simulator、learned surrogate、LLM prior 都是近似。Workload、SLO、model、hardware、engine version 改变后，operator cost、scheduler state transition、合法 flag 和 response surface 都可能变化。模型在 calibration task 上解释得好，不表示能在 held-out task 上排序正确。 + +#### 已有真实证据 + +- Frontier throughput reading 在完全匹配的 12-cell task 上仍把 real oracle 排错，top-1 regret 30.46%。这说明预测绝对 throughput 还不够，局部 rank fidelity 才是 tuning 关键。 +- Post-hoc SLO reading 的 top bucket 正确，但有大量 anchor feasibility error，也没有 prospective policy status。 +- Pure LLM 提出了当前 community-vLLM binary 不支持的 flag；engine/API version knowledge 本身会漂移。 +- 已有 cross-version experiment 中 vLLM 0.20 的强配置在 0.24 上出现大幅退化，说明 response prior 不能无条件迁移。 + +#### Tuner 必须具备的能力 + +- Simulator 只能作为 prior mean 或 candidate prior；真实 outcome 是 authoritative update。 +- 学习 simulator residual：把 `sim prediction + source state + action` 映射到 real response，而不是用 telemetry 重新实现另一个无校准 simulator。 +- 对 task-level OOD 显式提高 uncertainty/abstain；train/test 按完整 task 分割，不能按 request、anchor 或同一 surface cell 随机分割。 +- 分开报告 cold-start profile/training cost 与 per-task marginal cost，并在 N=1/10/100 等 amortization horizon 下展示。 + +## 4. 对应的系统设计 + +### 4.1 Harness：从 rule-based tuner 收缩成 experimental control plane + +Harness 保留以下确定性职责： + +- engine-version-aware config schema、合法性和资源约束； +- 完整 config/action canonicalization，禁止隐式 merge 和重复试验； +- exact trace/request/arrival/length hash，配对、随机化和 counter-rotation； +- engine trajectory、external outcome、failure/censoring 的统一时间轴； +- all-in GPU cost ledger、oracle annotation 分账、budget enforcement； +- data sanity、coverage、SLO/correctness 和 stop-proof audit。 + +Harness **不**包含 `queue > N -> increase MNS`、`cap full -> tune knob` 或人工 diagnosis-to-action mapping。这里的规则是实验语义和安全 invariant，不是性能决策 heuristic。 + +### 4.2 Action-conditioned response model + +每条学习记录为： + +```text +x = {source full config, + workload/SLO context, + source external outcome, + phase-binned engine trajectory} +a = normalized full-config delta +y = {Delta SLO-goodput, target feasibility/failure, measured H20 cost} +``` + +学习： + +```text +p_theta(y | x, a, optional simulator prediction) +``` + +第一版应使用适合小数据且有 uncertainty 的 action-conditioned Gaussian-process/bootstrapped surrogate；kernel/feature ablation包括： + +1. config + external outcome； +2. 同样输入 + telemetry trajectory； +3. simulator + config + outcome； +4. 同样输入 + telemetry residual features。 + +Telemetry 保留 continuous phase distributions：queue/running residency、MNS/token slack、prefill/decode composition、partial/split prefill、step duration、KV、graph/padding。模型学习它们与 action 的 interaction；不先压成 bottleneck label。 + +### 4.3 Cost-aware policy + +在合法的 single/joint candidate set 上选择： + +```text +a* = argmax_a + expected constrained improvement(a) + / expected all-in H20 cost(a) +``` + +探索项来自 posterior uncertainty/information gain；launch/SLO failure 有显式 penalty。Simulator 可提供 prior mean，但 simulator 与 real discrepancy 会被 posterior residual 更新。一次 target measurement 后更新 response model，并重新计算下一步 action 或停止概率。 + +LLM 在这个 tuning core 中不是 telemetry classifier。它最多作为可移除的 candidate/prior source，提出 schema 内的 sparse joint actions 或检索 engine mechanism；每个 proposal 都由同一个 response model、cost acquisition 和 real validator 评分。只有 `with LLM` 相对 `same tuner without LLM` 在 held-out tasks 上继续降低 cost-to-oracle，才能讨论 LLM 必要性。 + +### 4.4 Stop 条件 + +对一个预先声明的有限 candidate set，满足以下条件才 stop： + +```text +P(exists c: f(c) > best_observed / (1 - epsilon) | D_t) < alpha +``` + +并且 best config 通过独立 confirmation、SLO/correctness gate，remaining candidate 的 cost-aware value of information 低于阈值。停止原因、posterior coverage 和未测区域必须写入 audit。 + +## 5. 下一阶段如何证明，而不是再次构造 heuristic + +### R0：已有数据 retrospective premise check + +- 用 C1/C3 response surfaces 检查 joint model 是否能避免 OAT trap。 +- 用 action-aware paired records 比较 outcome-only 与 +telemetry 的 action-delta calibration。 +- 用 SimFid surface 比较 direct model 与 simulator-residual model 的 rank/regret。 +- 所有 feature、kernel、candidate encoding 在 held-out task 结果之前冻结。 + +R0 只能筛选 model family，不能作为 paper result，因为现有 tasks 已参与路线设计。 + +### R1：prospective same-host cost-to-oracle pilot + +- dash0 8xH20，固定 engine build/model；serialized placement，禁止共置干扰。 +- 至少一个未参与 feature/threshold 选择的新 trace window；选择非 ceiling、可 drain 的 offered load。 +- 声明一个可穷举的小 surface，至少包含 topology/runtime crossed actions，而不是只有一个 MNS ladder。 +- Oracle annotation 与 tuner online actions 分开记账；method 只能看到当时可用的数据。 +- 运行 random/search、OAT、纯 LLM、当前 guided harness、frozen simulator+real、outcome-only response、+telemetry response、sim-residual +telemetry。 +- 比较完整 H20 cost-to-regret curve，而不是 action classification accuracy。 + +Pilot opening gate： + +1. telemetry model 相对相同 response model 去掉 telemetry，确实改变至少一个正确的 prospective action ranking； +2. 最终 regret <=5%，无 false-safe accept； +3. all-in H20-hours 相对 strongest safe outcome-only 至少下降 20%； +4. 如果使用 simulator，需相对 frozen simulator+real 至少下降 30%； +5. instrumentation overhead <=1%，所有成本和失败均计入。 + +若 1--5 任一失败，就不能把 telemetry/harness 写成 tuning contribution；保留其 debugging/measurement 价值即可。 + +### R2：task-held-out replication + +至少 3 个 workload window x 2 个 SLO regime，按完整 task 做 leave-one-task-out 或固定 train/test split。报告每个 task 的 regret、安全和成本，以及 task-level paired bootstrap CI。只有 R2 通过，才能把单 task 的 61.09% lower-bound saving 升级为项目贡献。 + +## 6. 当前能说与不能说的贡献 + +当前能说： + +- 我们有真实反例证明 OAT 和 cap-to-knob mapping 不是通用 tuning strategy。 +- Harness 的 legality、exact replay、failure/cost accounting 有必要的实验基础设施价值。 +- 当前 guided sequence 在一个严格同任务比较中显著减少了达到 2% empirical regret 的 reconstructed engine cost。 +- Simulator 的边际计算便宜，但 rank error 会转化成显著 real regret 或更多 real-final 成本。 + +当前不能说： + +- telemetry 已经对 end-to-end tuning 提供独立增益；现有 direct pilot 对此为 negative。 +- 当前 harness 的 heuristic action ranking 是系统贡献；5% endpoint 只省 5.85%。 +- LLM 是必要组件；尚无同 policy 的 with/without LLM held-out ablation。 +- simulator 总 tuning cost 是 0；profile GPU cost 未审计，real verification 不能忽略。 +- 3.35 是 global oracle，或 dash0 与 dash1 数字是完全 controlled comparison。 + +## Data sanity + +- Dash0 sequential numeric scores：n=9，min/max `1.1042/3.35`，distinct=7；两组 config outcome 不全相同。 +- Exact surface scores：n=12，min/max `1.2833/3.2833`，distinct=8；12 cells 完整且与 simulator metrics 中的 real scores 一致。 +- Reconstructed trial/cell attempts 包括 4 个无 engine timestamp 的失败：n=32，min/max `0/0.49778 H20h`，distinct=26；所有可重建成本均非负。 +- Sequential regret observations：n=16，min/max `0/0.34328`，distinct=6，全部在 `[0,1]`。 +- Checked invariants：dash0 fixed task contexts 相同（除 method/port）；trial counts 与 manifest 相符；engine log timestamps monotonic；surface cell 唯一且 MBT=8192；simulator 无失败且 predictions 不全相同；scores/results 不全相同；cost 非负；regret bounded。 +- Measurement limitation：primary 12-cell campaign 的 4 个 TP4 pre-ready failure 没有 engine timestamp，随后由 companion campaign 完整重跑；其失败成本在 engine-lifetime reconstruction 中为 0。因此 `3.5953 H20h` 是 completed annotation lower bound，不能作为 all-in annotation cost。这个缺口已显式保留，没有在其上建立 total-cost claim。 diff --git a/runs/action-aware-v0/action_aware_client.py b/runs/action-aware-v0/action_aware_client.py new file mode 100644 index 0000000..58fbe27 --- /dev/null +++ b/runs/action-aware-v0/action_aware_client.py @@ -0,0 +1,63 @@ +#!/usr/bin/env python3 +"""Add explicit MBBT/config provenance to the accepted Phase-6 replay client.""" + +from __future__ import annotations + +import argparse +import json +import sys +from pathlib import Path + + +PHASE6 = Path(__file__).resolve().parents[1] / "opprof-phase6" +sys.path.insert(0, str(PHASE6)) + +import opprof_phase6_client as base # noqa: E402 + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("command", choices=("warmup", "run-anchor")) + result.add_argument("--study", required=True) + result.add_argument("--cell", required=True) + result.add_argument("--anchor", type=float, required=True) + result.add_argument("--tp", type=int, required=True) + result.add_argument("--mns", type=int, required=True) + result.add_argument("--mbbt", type=int, required=True) + result.add_argument("--base-url", required=True) + result.add_argument("--result-dir", required=True) + result.add_argument("--disable-slo-early-stop", action="store_true") + return result + + +def main() -> None: + args = parser().parse_args() + result = base.run_replay(args, warmup=args.command == "warmup") + result.update( + { + "schema": "action-aware-pilot-result-v0", + "config_id": args.cell, + "mbbt": args.mbbt, + } + ) + base.atomic_json(Path(args.result_dir) / "result.json", result) + print( + json.dumps( + { + key: result[key] + for key in ( + "config_id", + "mns", + "mbbt", + "kind", + "pass_rate", + "feasible", + ) + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/action-aware-v0/analyze_pilot.py b/runs/action-aware-v0/analyze_pilot.py new file mode 100644 index 0000000..18e295d --- /dev/null +++ b/runs/action-aware-v0/analyze_pilot.py @@ -0,0 +1,697 @@ +#!/usr/bin/env python3 +"""Audit source-only constraint signals against crossed real interventions.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import statistics +import sys +from pathlib import Path +from typing import Any, Iterable, Mapping + + +HERE = Path(__file__).resolve().parent +COMMON_STATE = HERE.parent / "telemetry-residual" +sys.path.insert(0, str(COMMON_STATE)) + +from common_state import summarize_engine # noqa: E402 + + +SCHEMA = "action-aware-constraint-pilot-audit-v0" + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def numeric(values: Iterable[float]) -> dict[str, Any]: + finite = [float(value) for value in values] + if not finite: + raise ValueError("numeric summary requires values") + if any(not math.isfinite(value) for value in finite): + raise ValueError("numeric summary received non-finite values") + return { + "n": len(finite), + "min": min(finite), + "max": max(finite), + "distinct_n": len(set(finite)), + } + + +def distribution(values: Iterable[float]) -> dict[str, Any]: + finite = [float(value) for value in values] + summary = numeric(finite) + return { + **summary, + "mean": statistics.fmean(finite), + "p50": quantile(finite, 0.50), + "p95": quantile(finite, 0.95), + "p99": quantile(finite, 0.99), + } + + +def quantile(values: Iterable[float], probability: float) -> float: + ordered = sorted(float(value) for value in values) + if not ordered: + raise ValueError("quantile requires values") + position = probability * (len(ordered) - 1) + lower = math.floor(position) + upper = math.ceil(position) + if lower == upper: + return ordered[lower] + weight = position - lower + return ordered[lower] * (1.0 - weight) + ordered[upper] * weight + + +def load_jsonl(path: Path) -> list[dict[str, Any]]: + records = [] + with path.open(encoding="utf-8") as source: + for line_number, line in enumerate(source, 1): + try: + records.append(json.loads(line)) + except json.JSONDecodeError as error: + raise ValueError(f"{path}:{line_number}: invalid JSON") from error + return records + + +def binding_summary( + records: list[Mapping[str, Any]], *, mns: int, mbbt: int +) -> dict[str, Any]: + if not records: + raise ValueError("binding summary requires scheduler records") + counts = { + "mns_exclusive": 0, + "mbbt_exclusive": 0, + "both": 0, + "waiting_unresolved": 0, + "waiting": 0, + } + running_utilization = [] + token_utilization = [] + kv_usage = [] + preemptions = 0 + for record in records: + waiting = int(record["queues"]["waiting"]) + int( + record["queues"]["deferred"] + ) + running = int(record["queues"]["running"]) + scheduled_tokens = int(record["prefill_tokens"]) + int( + record["decode_tokens"] + ) + if running > mns: + raise ValueError("running requests exceed configured MNS") + if scheduled_tokens > mbbt: + raise ValueError("scheduled tokens exceed configured MBBT") + mns_hit = waiting > 0 and running == mns + mbbt_hit = waiting > 0 and scheduled_tokens == mbbt + if waiting > 0: + counts["waiting"] += 1 + if mns_hit and mbbt_hit: + counts["both"] += 1 + elif mns_hit: + counts["mns_exclusive"] += 1 + elif mbbt_hit: + counts["mbbt_exclusive"] += 1 + else: + counts["waiting_unresolved"] += 1 + running_utilization.append(running / mns) + token_utilization.append(scheduled_tokens / mbbt) + kv_usage.append(float(record["kv"]["usage"])) + preemptions += int(record["preemptions"]) + count = len(records) + return { + "records": count, + **{f"{name}_count": value for name, value in counts.items()}, + **{f"{name}_fraction": value / count for name, value in counts.items()}, + "running_utilization_mean": statistics.fmean(running_utilization), + "running_utilization_max": max(running_utilization), + "token_utilization_mean": statistics.fmean(token_utilization), + "token_utilization_max": max(token_utilization), + "kv_usage_mean": statistics.fmean(kv_usage), + "kv_usage_max": max(kv_usage), + "preemptions": preemptions, + } + + +def telemetry_coverage( + records: list[Mapping[str, Any]], *, start_ns: int, end_ns: int +) -> tuple[dict[str, float], bool]: + if not records: + raise ValueError("telemetry coverage requires records") + submit_gaps = [ + (int(right["submit_mono_ns"]) - int(left["submit_mono_ns"])) / 1e9 + for left, right in zip(records, records[1:], strict=False) + ] + uncovered_gaps = [ + max( + 0, + int(right["submit_mono_ns"]) - int(left["complete_mono_ns"]), + ) + / 1e9 + for left, right in zip(records, records[1:], strict=False) + ] + coverage = { + "start_gap_s": (int(records[0]["submit_mono_ns"]) - start_ns) / 1e9, + "end_gap_s": (end_ns - int(records[-1]["submit_mono_ns"])) / 1e9, + "max_internal_submit_gap_s": max(submit_gaps, default=0.0), + "max_uncovered_gap_s": max(uncovered_gaps, default=0.0), + } + covered = ( + 0.0 <= coverage["start_gap_s"] <= 1.0 + and 0.0 <= coverage["end_gap_s"] <= 1.0 + and 0.0 <= coverage["max_uncovered_gap_s"] <= 1.0 + ) + return coverage, covered + + +def mechanism_summary(records: list[Mapping[str, Any]]) -> dict[str, Any]: + executed = [record for record in records if bool(record["model_executed"])] + if not executed: + raise ValueError("mechanism summary requires executed steps") + prefill = [record for record in executed if int(record["prefill_tokens"]) > 0] + decode_only = [ + record for record in executed if int(record["prefill_tokens"]) == 0 + ] + if not prefill or not decode_only: + raise ValueError("mechanism summary requires prefill and decode-only steps") + + def durations_ms(selected: list[Mapping[str, Any]]) -> list[float]: + values = [ + (int(record["complete_mono_ns"]) - int(record["submit_mono_ns"])) + / 1e6 + for record in selected + ] + if any(value < 0.0 for value in values): + raise ValueError("engine step duration must be non-negative") + return values + + chunk_keys = ("first", "middle", "final", "unsplit", "tokens") + chunks = { + key: sum(int(record["chunked_prefill"][key]) for record in executed) + for key in chunk_keys + } + prefill_tokens = [int(record["prefill_tokens"]) for record in prefill] + prefill_requests = sum(int(record["prefill_requests"]) for record in prefill) + prefix_queries = sum( + int(record["prefix"]["local"]["queries"]) for record in executed + ) + prefix_hits = sum( + int(record["prefix"]["local"]["hits"]) for record in executed + ) + invariants = { + "nonnegative_counts": all( + value >= 0 + for value in ( + *chunks.values(), + prefill_requests, + prefix_queries, + prefix_hits, + ) + ), + "chunk_tokens_match_prefill_tokens": chunks["tokens"] + == sum(prefill_tokens), + "prefix_hits_bounded": 0 <= prefix_hits <= prefix_queries, + } + return { + "executed_steps": len(executed), + "step_duration_ms": distribution(durations_ms(executed)), + "prefill_steps": len(prefill), + "prefill_step_duration_ms": distribution(durations_ms(prefill)), + "decode_only_steps": len(decode_only), + "decode_only_step_duration_ms": distribution(durations_ms(decode_only)), + "prefill": { + "requests": prefill_requests, + "requests_per_step": prefill_requests / len(prefill), + "tokens": sum(prefill_tokens), + "tokens_per_step": distribution(prefill_tokens), + "chunks": chunks, + }, + "prefix": { + "queries": prefix_queries, + "hits": prefix_hits, + "hit_rate": prefix_hits / prefix_queries if prefix_queries else 0.0, + }, + "sanity": {"invariants": invariants}, + } + + +def request_summary(path: Path, expected_count: int) -> dict[str, Any]: + rows = load_jsonl(path) + if len(rows) != expected_count: + raise ValueError(f"request row count mismatch: {path}") + ttft = [float(row["ttft_ms"]) for row in rows if row["ttft_ms"] is not None] + tpot = [float(row["tpot_ms"]) for row in rows if row["tpot_ms"] is not None] + if not ttft or not tpot: + raise ValueError(f"missing request latency values: {path}") + return { + "ttft_ms": {f"p{int(p * 100)}": quantile(ttft, p) for p in (0.5, 0.95, 0.99)}, + "tpot_ms": {f"p{int(p * 100)}": quantile(tpot, p) for p in (0.5, 0.95, 0.99)}, + } + + +def load_stream(session_root: Path) -> tuple[list[dict[str, Any]], dict[str, Any]]: + streams = sorted((session_root / "opprof").glob("*.jsonl")) + sidecars = sorted((session_root / "opprof").glob("*.jsonl.footer.json")) + if len(streams) != 1 or len(sidecars) != 1: + raise ValueError(f"expected one OpProf stream and sidecar: {session_root}") + decoded = load_jsonl(streams[0]) + records = [row for row in decoded if "step_index" in row] + footers = [row for row in decoded if row.get("record_type") == "footer"] + sidecar = json.loads(sidecars[0].read_text(encoding="utf-8")) + indexes = [int(row["step_index"]) for row in records] + invariants = { + "one_footer_last": len(footers) == 1 and decoded[-1] is footers[0], + "sidecar_final": sidecar.get("final") is True, + "zero_drops": sidecar.get("dropped_records") == 0, + "written_matches_records": sidecar.get("written_records") == len(records), + "contiguous_step_indexes": indexes == list(range(len(indexes))), + "monotonic_timestamps": all( + int(right["submit_mono_ns"]) >= int(left["submit_mono_ns"]) + for left, right in zip(records, records[1:], strict=False) + ), + } + return records, { + "stream": str(streams[0]), + "stream_sha256": sha256_file(streams[0]), + "records": len(records), + "invariants": invariants, + } + + +def analyze_run( + *, + run_root: Path, + config: Mapping[str, Any], + repetition: int, + expected: Mapping[str, Any], + stream_records: list[Mapping[str, Any]], + duration_s: float, + phase_fractions: list[float], +) -> dict[str, Any]: + result_root = run_root / "sessions" / str(config["id"]) / f"rep{repetition}" + result_path = result_root / "result.json" + result = json.loads(result_path.read_text(encoding="utf-8")) + selection = result["selection"] + invariants = { + "result_schema": result.get("schema") == "action-aware-pilot-result-v0", + "config_id": result.get("config_id") == config["id"], + "tp": int(result.get("tp", -1)) == 4, + "mns": int(result.get("mns", -1)) == int(config["mns"]), + "mbbt": int(result.get("mbbt", -1)) == int(config["mbbt"]), + "uncensored": not bool(result.get("early_stopped", True)), + "slo_early_stop_disabled": result.get("slo_early_stop_disabled") is True, + "selection_count": int(selection["count"]) == int(expected["selected_count"]), + "request_accounting": int(result["observed_count"]) + == int(expected["selected_count"]), + "request_hash": selection["request_id_order_sha256"] + == expected["request_id_order_sha256"], + "arrival_hash": selection["arrival_order_sha256"] + == expected["arrival_order_sha256"], + "length_hash": selection["raw_length_order_sha256"] + == expected["input_length_order_sha256"], + } + start_ns = int(result["interval"]["start_mono_ns"]) + arrival_end_ns = start_ns + round(duration_s * 1e9) + full_records = [ + record + for record in stream_records + if start_ns <= int(record["submit_mono_ns"]) <= arrival_end_ns + ] + if not full_records: + raise ValueError(f"no telemetry records in measured window: {result_path}") + coverage, invariants["telemetry_coverage"] = telemetry_coverage( + full_records, start_ns=start_ns, end_ns=arrival_end_ns + ) + binding = binding_summary( + full_records, mns=int(config["mns"]), mbbt=int(config["mbbt"]) + ) + mechanism = mechanism_summary(full_records) + invariants["mechanism_summary"] = all( + mechanism["sanity"]["invariants"].values() + ) + phases = {} + for fraction in phase_fractions: + phase_end = start_ns + round(duration_s * fraction * 1e9) + phase_records = [ + record + for record in full_records + if int(record["submit_mono_ns"]) <= phase_end + ] + phases[f"{fraction:.2f}"] = binding_summary( + phase_records, mns=int(config["mns"]), mbbt=int(config["mbbt"]) + ) + state = summarize_engine( + full_records, + start_ns=start_ns, + end_ns=arrival_end_ns, + request_count=int(result["observed_count"]), + ) + latency = request_summary( + result_root / "requests.jsonl", int(result["observed_count"]) + ) + return { + "config_id": config["id"], + "mns": int(config["mns"]), + "mbbt": int(config["mbbt"]), + "repetition": repetition, + "result_path": str(result_path), + "result_sha256": sha256_file(result_path), + "selection": { + "count": int(selection["count"]), + "request_id_order_sha256": selection["request_id_order_sha256"], + "arrival_order_sha256": selection["arrival_order_sha256"], + "raw_length_order_sha256": selection["raw_length_order_sha256"], + }, + "outcome": { + "pass_rate": float(result["pass_rate"]), + "feasible": bool(result["feasible"]), + "slo_pass_count": int(result["slo_pass_count"]), + "slo_goodput_req_s": int(result["slo_pass_count"]) / duration_s, + "elapsed_s": float(result["interval"]["elapsed_s"]), + **latency, + }, + "binding": binding, + "mechanism": mechanism, + "phases": phases, + "state": state, + "coverage": coverage, + "invariants": invariants, + } + + +def median(values: Iterable[float]) -> float: + return float(statistics.median(float(value) for value in values)) + + +def evaluate_decisions( + runs: list[Mapping[str, Any]], manifest: Mapping[str, Any] +) -> dict[str, Any]: + by_key = { + (str(run["config_id"]), int(run["repetition"])): run for run in runs + } + repetitions = sorted(int(key) for key in manifest["repetitions"]) + regime_results = {} + all_predictions = [] + crossed_pass = True + binding_pass = True + material_ambiguity = False + for regime_name, regime in manifest["regimes"].items(): + rows = [] + source_runs = [] + for repetition in repetitions: + source = by_key[(str(regime["source"]), repetition)] + mns_target = by_key[(str(regime["actions"]["mns"]), repetition)] + mbbt_target = by_key[(str(regime["actions"]["mbbt"]), repetition)] + source_runs.append(source) + source_goodput = float(source["outcome"]["slo_goodput_req_s"]) + mns_goodput = float(mns_target["outcome"]["slo_goodput_req_s"]) + mbbt_goodput = float(mbbt_target["outcome"]["slo_goodput_req_s"]) + observed = ( + "mns" + if mns_goodput > mbbt_goodput + else "mbbt" + if mbbt_goodput > mns_goodput + else "tie" + ) + mns_score = float(source["binding"]["mns_exclusive_fraction"]) + mbbt_score = float(source["binding"]["mbbt_exclusive_fraction"]) + predicted = ( + "mns" + if mns_score > mbbt_score + else "mbbt" + if mbbt_score > mns_score + else "tie" + ) + phase_predictions = {} + for phase, summary in source["phases"].items(): + left = float(summary["mns_exclusive_fraction"]) + right = float(summary["mbbt_exclusive_fraction"]) + phase_predictions[phase] = ( + "mns" if left > right else "mbbt" if right > left else "tie" + ) + margin = ( + abs(mns_goodput - mbbt_goodput) / source_goodput + if source_goodput > 0 + else None + ) + row = { + "repetition": repetition, + "source_goodput_req_s": source_goodput, + "mns_target_goodput_req_s": mns_goodput, + "mbbt_target_goodput_req_s": mbbt_goodput, + "observed_winner": observed, + "predicted_winner": predicted, + "prediction_correct": predicted == observed, + "relative_winner_margin_over_source": margin, + "mns_exclusive_fraction": mns_score, + "mbbt_exclusive_fraction": mbbt_score, + "phase_predictions": phase_predictions, + "phase_stable": all(value == predicted for value in phase_predictions.values()), + } + rows.append(row) + all_predictions.append(row) + + expected_winner = "mns" if regime_name == "A" else "mbbt" + minimum_margin = float(manifest["gates"]["minimum_relative_winner_margin"]) + regime_crossed = all( + row["observed_winner"] == expected_winner + and row["relative_winner_margin_over_source"] is not None + and row["relative_winner_margin_over_source"] >= minimum_margin + for row in rows + ) + crossed_pass &= regime_crossed + winning_key = f"{expected_winner}_exclusive_fraction" + losing_key = ( + "mbbt_exclusive_fraction" if expected_winner == "mns" else "mns_exclusive_fraction" + ) + winning_median = median(row[winning_key] for row in rows) + losing_median = median(row[losing_key] for row in rows) + ratio_pass = winning_median >= float( + manifest["gates"]["minimum_exclusive_ratio"] + ) * losing_median + regime_binding = ( + all(row["prediction_correct"] and row["phase_stable"] for row in rows) + and winning_median + >= float(manifest["gates"]["minimum_exclusive_fraction"]) + and ratio_pass + ) + binding_pass &= regime_binding + ambiguity_median = median( + float(run["binding"]["both_fraction"]) + + float(run["binding"]["waiting_unresolved_fraction"]) + for run in source_runs + ) + score_gap_median = median( + abs( + float(run["binding"]["mns_exclusive_fraction"]) + - float(run["binding"]["mbbt_exclusive_fraction"]) + ) + for run in source_runs + ) + kv_max_median = median( + float(run["binding"]["kv_usage_max"]) for run in source_runs + ) + any_preemption = any( + int(run["binding"]["preemptions"]) > 0 for run in source_runs + ) + regime_material = ( + ambiguity_median >= score_gap_median + or kv_max_median >= float(manifest["gates"]["material_kv_usage"]) + or any_preemption + ) + material_ambiguity |= regime_material + regime_results[regime_name] = { + "source": regime["source"], + "actions": regime["actions"], + "expected_winner": expected_winner, + "crossed_response_pass": regime_crossed, + "binding_pass": regime_binding, + "winning_exclusive_median": winning_median, + "losing_exclusive_median": losing_median, + "exclusive_ratio_pass": ratio_pass, + "ambiguity_median": ambiguity_median, + "exclusive_gap_median": score_gap_median, + "kv_usage_max_median": kv_max_median, + "any_preemption": any_preemption, + "material_ambiguity": regime_material, + "repetitions": rows, + } + + if not crossed_pass: + decision = "STOP_WORKLOAD_NOT_CROSSED" + elif not binding_pass: + decision = "STOP_BINDING_NOT_PREDICTIVE" + elif material_ambiguity: + decision = "OPEN_EXACT_ATTRIBUTION_ABLATION" + else: + decision = "STOP_NO_NEW_INSTRUMENTATION_NEEDED" + correct = sum(int(row["prediction_correct"]) for row in all_predictions) + return { + "decision": decision, + "crossed_response_pass": crossed_pass, + "binding_pass": binding_pass, + "material_ambiguity": material_ambiguity, + "regimes": regime_results, + "baselines": { + "always_mns_correct": sum( + int(row["observed_winner"] == "mns") for row in all_predictions + ), + "always_mbbt_correct": sum( + int(row["observed_winner"] == "mbbt") for row in all_predictions + ), + "binding_correct": correct, + "decision_count": len(all_predictions), + }, + } + + +def analyze(run_root: Path, manifest_path: Path) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + if manifest.get("schema") not in { + "action-aware-constraint-pilot-manifest-v0", + "action-aware-constraint-pilot-manifest-v1", + }: + raise ValueError("unexpected manifest schema") + duration_s = float(manifest["engine"]["duration_s"]) + phase_fractions = [float(value) for value in manifest["gates"]["phase_fractions"]] + runs = [] + stream_audits = [] + for config in manifest["configs"]: + session_root = run_root / "sessions" / str(config["id"]) + stream_records, stream_audit = load_stream(session_root) + stream_audit["config_id"] = config["id"] + stream_audits.append(stream_audit) + for repetition in sorted(int(key) for key in manifest["repetitions"]): + runs.append( + analyze_run( + run_root=run_root, + config=config, + repetition=repetition, + expected=manifest["repetitions"][str(repetition)]["selection"], + stream_records=stream_records, + duration_s=duration_s, + phase_fractions=phase_fractions, + ) + ) + invariants = { + "fifteen_runs": len(runs) == 15, + "five_streams": len(stream_audits) == 5, + "all_run_invariants": all( + all(bool(value) for value in run["invariants"].values()) for run in runs + ), + "all_stream_invariants": all( + all(bool(value) for value in stream["invariants"].values()) + for stream in stream_audits + ), + "nonnegative_counters": all( + all( + float(run["binding"][key]) >= 0 + for key in ( + "mns_exclusive_count", + "mbbt_exclusive_count", + "both_count", + "waiting_unresolved_count", + "preemptions", + ) + ) + for run in runs + ), + "ratios_bounded": all( + all( + 0.0 <= float(run["binding"][key]) <= 1.0 + for key in ( + "mns_exclusive_fraction", + "mbbt_exclusive_fraction", + "both_fraction", + "waiting_unresolved_fraction", + "kv_usage_mean", + "kv_usage_max", + ) + ) + for run in runs + ), + "per_config_results_not_all_identical": len( + {float(run["outcome"]["pass_rate"]) for run in runs} + ) + > 1, + } + red_flags = [name for name, passed in invariants.items() if not passed] + decisions = ( + evaluate_decisions(runs, manifest) + if not red_flags + else { + "decision": "STOP_DATA_INVALID", + "crossed_response_pass": False, + "binding_pass": False, + "material_ambiguity": False, + "regimes": {}, + "baselines": {}, + } + ) + payload = { + "schema": SCHEMA, + "decision": decisions["decision"], + "manifest": str(manifest_path), + "manifest_sha256": sha256_file(manifest_path), + "run_root": str(run_root), + "runs": runs, + "streams": stream_audits, + "decision_audit": decisions, + "sanity": { + "runs": len(runs), + "pass_rate": numeric(run["outcome"]["pass_rate"] for run in runs), + "slo_goodput_req_s": numeric( + run["outcome"]["slo_goodput_req_s"] for run in runs + ), + "telemetry_records_per_run": numeric( + run["binding"]["records"] for run in runs + ), + "mns_values": numeric(run["mns"] for run in runs), + "mbbt_values": numeric(run["mbbt"] for run in runs), + "invariants": invariants, + "red_flags": red_flags, + }, + } + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = analyze(args.run_root, args.manifest) + atomic_json(args.output, payload) + print( + json.dumps( + { + "decision": payload["decision"], + "sanity": payload["sanity"], + "decision_audit": payload["decision_audit"], + }, + indent=2, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/action-aware-v0/pilot-manifest-v1.json b/runs/action-aware-v0/pilot-manifest-v1.json new file mode 100644 index 0000000..14558e7 --- /dev/null +++ b/runs/action-aware-v0/pilot-manifest-v1.json @@ -0,0 +1,227 @@ +{ + "budget": { + "expected_h20_hours": [ + 6.0, + 7.2 + ], + "expected_wall_minutes": [ + 90, + 110 + ], + "global_hard_cap_h20_hours": 8.0, + "hard_cap_h20_hours": 7.614013100465138, + "prior_attempt_artifact": "/home/admin/cpfs/wjh/action-aware-constraint-v0-20260714/operational-stop-v0.json", + "prior_attempt_h20_hours": 0.38598689953486126, + "safety_h20_hours": 0.25, + "session_estimate_h20_hours": 1.35 + }, + "burnin": { + "anchor": 0.18919793755240089, + "arrival_order_sha256": "6c0ac4cb9a30ef501eeeacc8e6cc631c345e976db5ccf530ea5a1ec706d62a24", + "input_length_order_sha256": "7939cc20e1a00d1031d27d71508789f38decbbbb6ea59a1df18b2ec342fd2ef8", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "84f4809acbc8acd3b1d14dfa357134a1dc0b9287341624b33f598dafeef54dc7", + "selected_count": 510, + "study": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/studies/burnin-tp4.json", + "study_sha256": "5d6c2098042909a863efd3112818fbee9bafe96f22898ac98b66846dbe1fef0f" + }, + "configs": [ + { + "id": "b_base", + "mbbt": 2048, + "mns": 64, + "repetition_order": [ + 1, + 2, + 3 + ] + }, + { + "id": "a_base", + "mbbt": 8192, + "mns": 16, + "repetition_order": [ + 2, + 3, + 1 + ] + }, + { + "id": "shared", + "mbbt": 8192, + "mns": 64, + "repetition_order": [ + 3, + 1, + 2 + ] + }, + { + "id": "b_mns", + "mbbt": 2048, + "mns": 128, + "repetition_order": [ + 1, + 3, + 2 + ] + }, + { + "id": "a_mbbt", + "mbbt": 16384, + "mns": 16, + "repetition_order": [ + 2, + 1, + 3 + ] + } + ], + "engine": { + "burnin_max_elapsed_s": 90.0, + "client_timeout_s": 450.0, + "disable_slo_early_stop": true, + "duration_s": 300.0, + "tp": 4 + }, + "gates": { + "material_kv_usage": 0.9, + "minimum_exclusive_fraction": 0.1, + "minimum_exclusive_ratio": 5.0, + "minimum_relative_winner_margin": 0.1, + "phase_fractions": [ + 0.25, + 0.5, + 0.75, + 1.0 + ] + }, + "regimes": { + "A": { + "actions": { + "mbbt": "a_mbbt", + "mns": "shared" + }, + "source": "a_base" + }, + "B": { + "actions": { + "mbbt": "shared", + "mns": "b_mns" + }, + "source": "b_base" + } + }, + "repetitions": { + "1": { + "merged_trace": { + "bytes": 337429767, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep1.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9420, + "sha256": "68983266aa0e66aa589562f7c08edbd966f9ba4405e20c105adb43777d2dfbf5", + "source_sha256": [ + "b242d1d9086df3accab57b4c92445d5edd581e12f47e12cea227aa63964c6930", + "d23b549f7b69af3647308677bbf76f818a3c226a1c98f9a9f93f09ceee46be87" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low1.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high1.jsonl" + ] + }, + "selection": { + "anchor": 0.48686986110831465, + "arrival_order_sha256": "c2ad99986ce558da5901a9c5ec0a00bd69f198c981d8779235f2773a5c87f1c0", + "input_length_order_sha256": "9442bfebdc3fab5062dc1f4d688dc28c02afe3fd806c56dd8159f0ac7e6d0b94", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep1-tp4.json", + "study_sha256": "ecfff96e33d458eb1e3b9a6d24386f00cc6f1b19ff926e2ec6320b3f671a7ae3" + }, + "2": { + "merged_trace": { + "bytes": 337509330, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep2.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9457, + "sha256": "f38e8938f6a481fc6725b71b21aa04ff7eaf79783cdfd6e41aa2f074156f00c2", + "source_sha256": [ + "4cbb0baac082bd54af562ce2f39104c5c23b4671672da365a67b1e8c146adf9f", + "bb0bcd2564a88000f435f12feb21c7c902eafc9ea5fe916adfe9d1eae47f3f9a" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low2.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high2.jsonl" + ] + }, + "selection": { + "anchor": 0.4825698948735577, + "arrival_order_sha256": "b9fc12cf3f86bc8a79bee65296e65aa2b8bf2aeca46b2887094c669adcbb9a00", + "input_length_order_sha256": "d8d4bd6fc8ba852a45605b673b6b3e4f33b58f459e69f2a032d226ee175b074e", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep2-tp4.json", + "study_sha256": "d92a576db031db24bb58f354ea725d7f7567cb76699d387117ac5a6c9317bbb9" + }, + "3": { + "merged_trace": { + "bytes": 337450256, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep3.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9431, + "sha256": "3094084b0bb20cc02eecf465091a5c919b4e5b112f704cdc36a563d1efdcee46", + "source_sha256": [ + "1f7ececb142f9a363d2d1ca25eb7b8488b2cc319a51b55faa384f2a3d51f2142", + "6f326234791e1cff4ff866bface0d097d0d6e3844eebb1c97653d8e9c35e9397" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low3.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high3.jsonl" + ] + }, + "selection": { + "anchor": 0.48664343020532463, + "arrival_order_sha256": "efce7339e22d3618cb4d55e6b55bfddb2c563c18faba2a992d5829c13e3f55e9", + "input_length_order_sha256": "0792b05fff6729fbd92ab2bb4cb6d31bea7799e232ad42772936bc06efbafb54", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep3-tp4.json", + "study_sha256": "fb8ffe256dace32f4ca8a8d49b662d98c3b69b94ecc8fa826e43068b238884ab" + } + }, + "sanity": { + "invariants": { + "all_repetition_orders_are_permutations": true, + "five_unique_configs": true, + "same_load_all_repetitions": true, + "shared_endpoint_reused_by_both_regimes": true, + "three_disjoint_repetitions": true + }, + "red_flags": [] + }, + "schema": "action-aware-constraint-pilot-manifest-v1", + "source": { + "base_manifest": "/home/gahow/phd/aituner/runs/intervention-response-v2/pilot-manifest-v3.json", + "base_manifest_sha256": "273db1181dcc9d6b64439650d0642ebe553b12e6aa9adebfbe3758a7977e5611", + "source_trace": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/chat_w20260312_1000.jsonl", + "source_trace_sha256": "875ba869775deb78086477919f03b322da14e2673c7d070e26528c4190912757", + "window_id": "chat_w20260312_1000" + }, + "status": "PASS" +} diff --git a/runs/action-aware-v0/pilot-manifest-v2.json b/runs/action-aware-v0/pilot-manifest-v2.json new file mode 100644 index 0000000..47908c7 --- /dev/null +++ b/runs/action-aware-v0/pilot-manifest-v2.json @@ -0,0 +1,227 @@ +{ + "budget": { + "expected_h20_hours": [ + 6.0, + 7.2 + ], + "expected_wall_minutes": [ + 90, + 110 + ], + "global_hard_cap_h20_hours": 8.0, + "hard_cap_h20_hours": 7.295602157380846, + "prior_attempt_artifact": "/home/admin/cpfs/wjh/aituner/aituner-action-aware-20260714/runs/action-aware-v0/prior-attempts-v2.json", + "prior_attempt_h20_hours": 0.7043978426191542, + "safety_h20_hours": 0.25, + "session_estimate_h20_hours": 1.35 + }, + "burnin": { + "anchor": 0.18919793755240089, + "arrival_order_sha256": "6c0ac4cb9a30ef501eeeacc8e6cc631c345e976db5ccf530ea5a1ec706d62a24", + "input_length_order_sha256": "7939cc20e1a00d1031d27d71508789f38decbbbb6ea59a1df18b2ec342fd2ef8", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "84f4809acbc8acd3b1d14dfa357134a1dc0b9287341624b33f598dafeef54dc7", + "selected_count": 510, + "study": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/studies/burnin-tp4.json", + "study_sha256": "5d6c2098042909a863efd3112818fbee9bafe96f22898ac98b66846dbe1fef0f" + }, + "configs": [ + { + "id": "b_base", + "mbbt": 2048, + "mns": 64, + "repetition_order": [ + 1, + 2, + 3 + ] + }, + { + "id": "a_base", + "mbbt": 8192, + "mns": 16, + "repetition_order": [ + 2, + 3, + 1 + ] + }, + { + "id": "shared", + "mbbt": 8192, + "mns": 64, + "repetition_order": [ + 3, + 1, + 2 + ] + }, + { + "id": "b_mns", + "mbbt": 2048, + "mns": 128, + "repetition_order": [ + 1, + 3, + 2 + ] + }, + { + "id": "a_mbbt", + "mbbt": 16384, + "mns": 16, + "repetition_order": [ + 2, + 1, + 3 + ] + } + ], + "engine": { + "burnin_max_elapsed_s": 90.0, + "client_timeout_s": 450.0, + "disable_slo_early_stop": true, + "duration_s": 300.0, + "tp": 4 + }, + "gates": { + "material_kv_usage": 0.9, + "minimum_exclusive_fraction": 0.1, + "minimum_exclusive_ratio": 5.0, + "minimum_relative_winner_margin": 0.1, + "phase_fractions": [ + 0.25, + 0.5, + 0.75, + 1.0 + ] + }, + "regimes": { + "A": { + "actions": { + "mbbt": "a_mbbt", + "mns": "shared" + }, + "source": "a_base" + }, + "B": { + "actions": { + "mbbt": "shared", + "mns": "b_mns" + }, + "source": "b_base" + } + }, + "repetitions": { + "1": { + "merged_trace": { + "bytes": 337429767, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep1.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9420, + "sha256": "68983266aa0e66aa589562f7c08edbd966f9ba4405e20c105adb43777d2dfbf5", + "source_sha256": [ + "b242d1d9086df3accab57b4c92445d5edd581e12f47e12cea227aa63964c6930", + "d23b549f7b69af3647308677bbf76f818a3c226a1c98f9a9f93f09ceee46be87" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low1.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high1.jsonl" + ] + }, + "selection": { + "anchor": 0.48686986110831465, + "arrival_order_sha256": "c2ad99986ce558da5901a9c5ec0a00bd69f198c981d8779235f2773a5c87f1c0", + "input_length_order_sha256": "9442bfebdc3fab5062dc1f4d688dc28c02afe3fd806c56dd8159f0ac7e6d0b94", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep1-tp4.json", + "study_sha256": "ecfff96e33d458eb1e3b9a6d24386f00cc6f1b19ff926e2ec6320b3f671a7ae3" + }, + "2": { + "merged_trace": { + "bytes": 337509330, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep2.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9457, + "sha256": "f38e8938f6a481fc6725b71b21aa04ff7eaf79783cdfd6e41aa2f074156f00c2", + "source_sha256": [ + "4cbb0baac082bd54af562ce2f39104c5c23b4671672da365a67b1e8c146adf9f", + "bb0bcd2564a88000f435f12feb21c7c902eafc9ea5fe916adfe9d1eae47f3f9a" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low2.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high2.jsonl" + ] + }, + "selection": { + "anchor": 0.4825698948735577, + "arrival_order_sha256": "b9fc12cf3f86bc8a79bee65296e65aa2b8bf2aeca46b2887094c669adcbb9a00", + "input_length_order_sha256": "d8d4bd6fc8ba852a45605b673b6b3e4f33b58f459e69f2a032d226ee175b074e", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep2-tp4.json", + "study_sha256": "d92a576db031db24bb58f354ea725d7f7567cb76699d387117ac5a6c9317bbb9" + }, + "3": { + "merged_trace": { + "bytes": 337450256, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep3.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9431, + "sha256": "3094084b0bb20cc02eecf465091a5c919b4e5b112f704cdc36a563d1efdcee46", + "source_sha256": [ + "1f7ececb142f9a363d2d1ca25eb7b8488b2cc319a51b55faa384f2a3d51f2142", + "6f326234791e1cff4ff866bface0d097d0d6e3844eebb1c97653d8e9c35e9397" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low3.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high3.jsonl" + ] + }, + "selection": { + "anchor": 0.48664343020532463, + "arrival_order_sha256": "efce7339e22d3618cb4d55e6b55bfddb2c563c18faba2a992d5829c13e3f55e9", + "input_length_order_sha256": "0792b05fff6729fbd92ab2bb4cb6d31bea7799e232ad42772936bc06efbafb54", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep3-tp4.json", + "study_sha256": "fb8ffe256dace32f4ca8a8d49b662d98c3b69b94ecc8fa826e43068b238884ab" + } + }, + "sanity": { + "invariants": { + "all_repetition_orders_are_permutations": true, + "five_unique_configs": true, + "same_load_all_repetitions": true, + "shared_endpoint_reused_by_both_regimes": true, + "three_disjoint_repetitions": true + }, + "red_flags": [] + }, + "schema": "action-aware-constraint-pilot-manifest-v1", + "source": { + "base_manifest": "/home/gahow/phd/aituner/runs/intervention-response-v2/pilot-manifest-v3.json", + "base_manifest_sha256": "273db1181dcc9d6b64439650d0642ebe553b12e6aa9adebfbe3758a7977e5611", + "source_trace": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/chat_w20260312_1000.jsonl", + "source_trace_sha256": "875ba869775deb78086477919f03b322da14e2673c7d070e26528c4190912757", + "window_id": "chat_w20260312_1000" + }, + "status": "PASS" +} diff --git a/runs/action-aware-v0/pilot-manifest.json b/runs/action-aware-v0/pilot-manifest.json new file mode 100644 index 0000000..170c02f --- /dev/null +++ b/runs/action-aware-v0/pilot-manifest.json @@ -0,0 +1,223 @@ +{ + "budget": { + "expected_h20_hours": [ + 6.0, + 7.2 + ], + "expected_wall_minutes": [ + 90, + 110 + ], + "hard_cap_h20_hours": 8.0, + "safety_h20_hours": 0.25, + "session_estimate_h20_hours": 1.35 + }, + "burnin": { + "anchor": 0.18919793755240089, + "arrival_order_sha256": "6c0ac4cb9a30ef501eeeacc8e6cc631c345e976db5ccf530ea5a1ec706d62a24", + "input_length_order_sha256": "7939cc20e1a00d1031d27d71508789f38decbbbb6ea59a1df18b2ec342fd2ef8", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "84f4809acbc8acd3b1d14dfa357134a1dc0b9287341624b33f598dafeef54dc7", + "selected_count": 510, + "study": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/studies/burnin-tp4.json", + "study_sha256": "5d6c2098042909a863efd3112818fbee9bafe96f22898ac98b66846dbe1fef0f" + }, + "configs": [ + { + "id": "b_base", + "mbbt": 256, + "mns": 64, + "repetition_order": [ + 1, + 2, + 3 + ] + }, + { + "id": "a_base", + "mbbt": 8192, + "mns": 16, + "repetition_order": [ + 2, + 3, + 1 + ] + }, + { + "id": "shared", + "mbbt": 8192, + "mns": 64, + "repetition_order": [ + 3, + 1, + 2 + ] + }, + { + "id": "b_mns", + "mbbt": 256, + "mns": 128, + "repetition_order": [ + 1, + 3, + 2 + ] + }, + { + "id": "a_mbbt", + "mbbt": 16384, + "mns": 16, + "repetition_order": [ + 2, + 1, + 3 + ] + } + ], + "engine": { + "client_timeout_s": 450.0, + "disable_slo_early_stop": true, + "duration_s": 300.0, + "tp": 4 + }, + "gates": { + "material_kv_usage": 0.9, + "minimum_exclusive_fraction": 0.1, + "minimum_exclusive_ratio": 5.0, + "minimum_relative_winner_margin": 0.1, + "phase_fractions": [ + 0.25, + 0.5, + 0.75, + 1.0 + ] + }, + "regimes": { + "A": { + "actions": { + "mbbt": "a_mbbt", + "mns": "shared" + }, + "source": "a_base" + }, + "B": { + "actions": { + "mbbt": "shared", + "mns": "b_mns" + }, + "source": "b_base" + } + }, + "repetitions": { + "1": { + "merged_trace": { + "bytes": 337429767, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep1.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9420, + "sha256": "68983266aa0e66aa589562f7c08edbd966f9ba4405e20c105adb43777d2dfbf5", + "source_sha256": [ + "b242d1d9086df3accab57b4c92445d5edd581e12f47e12cea227aa63964c6930", + "d23b549f7b69af3647308677bbf76f818a3c226a1c98f9a9f93f09ceee46be87" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low1.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high1.jsonl" + ] + }, + "selection": { + "anchor": 0.48686986110831465, + "arrival_order_sha256": "c2ad99986ce558da5901a9c5ec0a00bd69f198c981d8779235f2773a5c87f1c0", + "input_length_order_sha256": "9442bfebdc3fab5062dc1f4d688dc28c02afe3fd806c56dd8159f0ac7e6d0b94", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep1-tp4.json", + "study_sha256": "ecfff96e33d458eb1e3b9a6d24386f00cc6f1b19ff926e2ec6320b3f671a7ae3" + }, + "2": { + "merged_trace": { + "bytes": 337509330, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep2.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9457, + "sha256": "f38e8938f6a481fc6725b71b21aa04ff7eaf79783cdfd6e41aa2f074156f00c2", + "source_sha256": [ + "4cbb0baac082bd54af562ce2f39104c5c23b4671672da365a67b1e8c146adf9f", + "bb0bcd2564a88000f435f12feb21c7c902eafc9ea5fe916adfe9d1eae47f3f9a" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low2.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high2.jsonl" + ] + }, + "selection": { + "anchor": 0.4825698948735577, + "arrival_order_sha256": "b9fc12cf3f86bc8a79bee65296e65aa2b8bf2aeca46b2887094c669adcbb9a00", + "input_length_order_sha256": "d8d4bd6fc8ba852a45605b673b6b3e4f33b58f459e69f2a032d226ee175b074e", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep2-tp4.json", + "study_sha256": "d92a576db031db24bb58f354ea725d7f7567cb76699d387117ac5a6c9317bbb9" + }, + "3": { + "merged_trace": { + "bytes": 337450256, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep3.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9431, + "sha256": "3094084b0bb20cc02eecf465091a5c919b4e5b112f704cdc36a563d1efdcee46", + "source_sha256": [ + "1f7ececb142f9a363d2d1ca25eb7b8488b2cc319a51b55faa384f2a3d51f2142", + "6f326234791e1cff4ff866bface0d097d0d6e3844eebb1c97653d8e9c35e9397" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low3.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high3.jsonl" + ] + }, + "selection": { + "anchor": 0.48664343020532463, + "arrival_order_sha256": "efce7339e22d3618cb4d55e6b55bfddb2c563c18faba2a992d5829c13e3f55e9", + "input_length_order_sha256": "0792b05fff6729fbd92ab2bb4cb6d31bea7799e232ad42772936bc06efbafb54", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + }, + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep3-tp4.json", + "study_sha256": "fb8ffe256dace32f4ca8a8d49b662d98c3b69b94ecc8fa826e43068b238884ab" + } + }, + "sanity": { + "invariants": { + "all_repetition_orders_are_permutations": true, + "five_unique_configs": true, + "same_load_all_repetitions": true, + "shared_endpoint_reused_by_both_regimes": true, + "three_disjoint_repetitions": true + }, + "red_flags": [] + }, + "schema": "action-aware-constraint-pilot-manifest-v0", + "source": { + "base_manifest": "/home/gahow/phd/aituner/runs/intervention-response-v2/pilot-manifest-v3.json", + "base_manifest_sha256": "273db1181dcc9d6b64439650d0642ebe553b12e6aa9adebfbe3758a7977e5611", + "source_trace": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/chat_w20260312_1000.jsonl", + "source_trace_sha256": "875ba869775deb78086477919f03b322da14e2673c7d070e26528c4190912757", + "window_id": "chat_w20260312_1000" + }, + "status": "PASS" +} diff --git a/runs/action-aware-v0/pilot_controller.py b/runs/action-aware-v0/pilot_controller.py new file mode 100644 index 0000000..7ecddb3 --- /dev/null +++ b/runs/action-aware-v0/pilot_controller.py @@ -0,0 +1,638 @@ +#!/usr/bin/env python3 +"""Serialized controller for the crossed-constraint action-aware pilot.""" + +from __future__ import annotations + +import argparse +import json +import os +import shlex +import signal +import subprocess +import sys +import time +from pathlib import Path +from typing import Any, Mapping + + +HERE = Path(__file__).resolve().parent +PHASE6 = HERE.parent / "opprof-phase6" +sys.path.insert(0, str(PHASE6)) + +import opprof_phase6_controller as base # noqa: E402 + + +SCHEMA = "action-aware-constraint-pilot-state-v0" + + +def atomic_json(path: Path, payload: Any) -> None: + base.atomic_json(path, payload) + + +def wait_all_idle(timeout_s: float = 30.0) -> None: + deadline = time.monotonic() + timeout_s + last_error: Exception | None = None + while time.monotonic() < deadline: + try: + base.assert_all_idle() + return + except RuntimeError as error: + last_error = error + time.sleep(1.0) + raise last_error or RuntimeError("GPU idle timeout") + + +def configure(args: argparse.Namespace, manifest: Mapping[str, Any]) -> None: + base.WORKDIR = args.run_root.parent + base.RUN_ROOT = args.run_root + base.STATE = args.run_root / "controller-state.json" + base.SOURCE = args.vllm_source + base.VENV = args.venv + base.AITUNER = args.aituner_root + base.MODEL = args.model + base.CLIENT = args.client + base.GPU_LIMIT = float(manifest["budget"]["hard_cap_h20_hours"]) + base.MARKER = "action-aware-constraint-pilot-v0" + + +def validate_inputs(args: argparse.Namespace, manifest: Mapping[str, Any]) -> None: + if manifest.get("schema") not in { + "action-aware-constraint-pilot-manifest-v0", + "action-aware-constraint-pilot-manifest-v1", + }: + raise RuntimeError("unexpected action-aware manifest schema") + if manifest.get("status") != "PASS": + raise RuntimeError("action-aware manifest did not pass preflight") + red_flags = manifest.get("sanity", {}).get("red_flags", []) + if red_flags: + raise RuntimeError(f"manifest red flags: {red_flags}") + + required = { + "manifest": args.manifest, + "aituner_root": args.aituner_root, + "vllm_source": args.vllm_source, + "venv_python": args.venv / "bin/python", + "venv_vllm": args.venv / "bin/vllm", + "model": args.model, + "client": args.client, + "burnin_study": Path(manifest["burnin"]["study"]), + } + for repetition, item in manifest["repetitions"].items(): + required[f"rep{repetition}_study"] = Path(item["study"]) + required[f"rep{repetition}_trace"] = Path(item["merged_trace"]["path"]) + missing = {name: str(path) for name, path in required.items() if not path.exists()} + if missing: + raise RuntimeError(f"action-aware input paths missing: {missing}") + + +def config_map(manifest: Mapping[str, Any]) -> dict[str, dict[str, Any]]: + return {str(item["id"]): dict(item) for item in manifest["configs"]} + + +def server_command( + config: Mapping[str, Any], *, gpus: tuple[int, ...], port: int +) -> list[str]: + return [ + "taskset", + "-c", + base.cpu_mask(gpus), + str(base.VENV / "bin/vllm"), + "serve", + str(base.MODEL), + "--host", + "127.0.0.1", + "--port", + str(port), + "--served-model-name", + "qwen3-30b-a3b-community", + "--max-num-batched-tokens", + str(config["mbbt"]), + "--max-num-seqs", + str(config["mns"]), + "--tensor-parallel-size", + "4", + "--shutdown-timeout", + "120", + ] + + +def client_command( + entry: Mapping[str, Any], + config: Mapping[str, Any], + *, + study: str, + anchor: float, + output: Path, + warmup: bool, +) -> list[str]: + command = [ + "taskset", + "-c", + base.cpu_mask(entry["gpus"]), + str(base.VENV / "bin/python"), + str(base.CLIENT), + "warmup" if warmup else "run-anchor", + "--study", + study, + "--cell", + str(config["id"]), + "--anchor", + str(anchor), + "--tp", + "4", + "--mns", + str(config["mns"]), + "--mbbt", + str(config["mbbt"]), + "--base-url", + f"http://127.0.0.1:{entry['port']}", + "--result-dir", + str(output), + "--disable-slo-early-stop", + ] + return command + + +def remaining_projection( + manifest: Mapping[str, Any], *, completed_sessions: int +) -> float: + remaining = len(manifest["configs"]) - completed_sessions + return ( + remaining * float(manifest["budget"]["session_estimate_h20_hours"]) + + float(manifest["budget"]["safety_h20_hours"]) + ) + + +def dry_run_plan( + args: argparse.Namespace, manifest: Mapping[str, Any] +) -> dict[str, Any]: + sessions = [] + for index, config in enumerate(manifest["configs"]): + entry = {"gpus": (0, 1, 2, 3), "port": 9050 + index} + session_root = args.run_root / "sessions" / str(config["id"]) + first_repetition = str(config["repetition_order"][0]) + first = manifest["repetitions"][first_repetition] + commands = { + "server": server_command(config, gpus=entry["gpus"], port=entry["port"]), + "warmup": client_command( + entry, + config, + study=first["study"], + anchor=float(first["selection"]["anchor"]), + output=session_root / "warmup", + warmup=True, + ), + "burnin": client_command( + entry, + config, + study=manifest["burnin"]["study"], + anchor=float(manifest["burnin"]["anchor"]), + output=session_root / "burnin", + warmup=False, + ), + } + for repetition in config["repetition_order"]: + item = manifest["repetitions"][str(repetition)] + commands[f"rep{repetition}"] = client_command( + entry, + config, + study=item["study"], + anchor=float(item["selection"]["anchor"]), + output=session_root / f"rep{repetition}", + warmup=False, + ) + sessions.append( + { + "config": config["id"], + "mns": config["mns"], + "mbbt": config["mbbt"], + "port": entry["port"], + "repetition_order": config["repetition_order"], + "commands": { + role: shlex.join(command) for role, command in commands.items() + }, + } + ) + return { + "schema": "action-aware-constraint-pilot-dry-run-v0", + "status": "PASS", + "manifest": str(args.manifest), + "run_root": str(args.run_root), + "projected_h20_hours": remaining_projection( + manifest, completed_sessions=0 + ), + "hard_cap_h20_hours": manifest["budget"]["hard_cap_h20_hours"], + "sessions": sessions, + } + + +def load_state(path: Path, hard_cap: float) -> dict[str, Any]: + if path.exists(): + return json.loads(path.read_text(encoding="utf-8")) + return { + "schema": SCHEMA, + "status": "initialized", + "hard_cap_h20_hours": hard_cap, + "gpu_hours_total": 0.0, + "completed_sessions": 0, + "sessions": {}, + "failures": [], + "started_at": time.time(), + } + + +def append_echo(run_root: Path, line: str) -> None: + run_root.mkdir(parents=True, exist_ok=True) + with (run_root / "launch-echo.log").open("a", encoding="utf-8") as target: + target.write(line + "\n") + print(line, flush=True) + + +def start_server( + *, + args: argparse.Namespace, + config: Mapping[str, Any], + index: int, +) -> dict[str, Any]: + gpus = (0, 1, 2, 3) + session_root = args.run_root / "sessions" / str(config["id"]) + session_root.mkdir(parents=True, exist_ok=True) + port = 9050 + index + command = server_command(config, gpus=gpus, port=port) + with (session_root / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"SERVER {shlex.join(command)}\n") + server_log = (session_root / "server.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update( + { + "CUDA_VISIBLE_DEVICES": "0,1,2,3", + "VLLM_OPPROF_DIR": str(session_root / "opprof"), + "OPPROF_PHASE6_MARKER": base.MARKER, + "AITUNER_ROOT": str(base.AITUNER), + "HF_HUB_OFFLINE": "1", + "TRANSFORMERS_OFFLINE": "1", + "PYTHONUNBUFFERED": "1", + } + ) + server = subprocess.Popen( + command, + cwd=base.SOURCE, + env=environment, + stdout=server_log, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + base.OWNED_PGIDS.add(server.pid) + return { + "cell": str(config["id"]), + "gpus": gpus, + "port": port, + "dir": session_root, + "server": server, + "server_handle": server_log, + "spawned_at": time.time(), + "results": [], + } + + +def validate_result( + result: Mapping[str, Any], + *, + config: Mapping[str, Any], + selection: Mapping[str, Any], + role: str, + warmup: bool, +) -> None: + if result.get("schema") != "action-aware-pilot-result-v0": + raise RuntimeError(f"unexpected result schema: {role}") + if result.get("config_id") != config["id"]: + raise RuntimeError(f"config id mismatch: {role}") + if int(result["tp"]) != 4: + raise RuntimeError(f"TP mismatch: {role}") + if int(result["mns"]) != int(config["mns"]): + raise RuntimeError(f"MNS mismatch: {role}") + if int(result["mbbt"]) != int(config["mbbt"]): + raise RuntimeError(f"MBBT mismatch: {role}") + if result.get("slo_early_stop_disabled") is not True: + raise RuntimeError(f"SLO early stop was not disabled: {role}") + if warmup: + if result["kind"] != "warmup" or int(result["selection"]["count"]) != 16: + raise RuntimeError(f"invalid warmup: {role}") + return + if bool(result["early_stopped"]): + raise RuntimeError(f"uncensored run early-stopped: {role}") + if int(result["selection"]["count"]) != int(selection["selected_count"]): + raise RuntimeError(f"selection count mismatch: {role}") + if int(result["observed_count"]) != int(selection["selected_count"]): + raise RuntimeError(f"request accounting mismatch: {role}") + for result_key, selection_key in ( + ("request_id_order_sha256", "request_id_order_sha256"), + ("arrival_order_sha256", "arrival_order_sha256"), + ("raw_length_order_sha256", "input_length_order_sha256"), + ): + if result["selection"][result_key] != selection[selection_key]: + raise RuntimeError(f"selection hash mismatch {result_key}: {role}") + + +def burnin_gate( + result: Mapping[str, Any], + *, + expected_count: int, + maximum_elapsed_s: float, +) -> dict[str, Any]: + if result.get("kind") != "anchor": + raise RuntimeError("burnin gate received a non-anchor result") + if int(result["selection"]["count"]) != expected_count: + raise RuntimeError("burnin gate received the wrong request set") + elapsed_s = float(result["interval"]["elapsed_s"]) + summary = { + "elapsed_s": elapsed_s, + "pass_rate": float(result["pass_rate"]), + "feasible": bool(result["feasible"]), + } + if elapsed_s > maximum_elapsed_s: + raise RuntimeError( + f"burnin throughput gate failed: {elapsed_s:.3f}s > " + f"{maximum_elapsed_s:.3f}s" + ) + return summary + + +def run_client( + *, + entry: dict[str, Any], + config: Mapping[str, Any], + role: str, + study: str, + selection: Mapping[str, Any], + output: Path, + state: Mapping[str, Any], + timeout_s: float, + warmup: bool = False, +) -> dict[str, Any]: + command = client_command( + entry, + config, + study=study, + anchor=float(selection["anchor"]), + output=output, + warmup=warmup, + ) + with (entry["dir"] / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"CLIENT role={role} {shlex.join(command)}\n") + handle = (output.parent / f"{output.name}.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update({"AITUNER_ROOT": str(base.AITUNER), "PYTHONUNBUFFERED": "1"}) + process = subprocess.Popen( + command, + cwd=base.WORKDIR, + env=environment, + stdout=handle, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + deadline = time.monotonic() + timeout_s + try: + while process.poll() is None: + if time.monotonic() > deadline: + raise TimeoutError(f"client timeout: {config['id']} {role}") + if entry["server"].poll() is not None: + raise RuntimeError(f"server exited during {config['id']} {role}") + base.assert_no_other_compute() + if state["gpu_hours_total"] + base.live_gpu_hours([entry]) >= base.GPU_LIMIT: + raise RuntimeError("action-aware pilot H20-hour hard cap reached") + time.sleep(1.0) + except Exception: + try: + os.killpg(process.pid, signal.SIGTERM) + except ProcessLookupError: + pass + try: + process.wait(timeout=10.0) + except subprocess.TimeoutExpired: + try: + os.killpg(process.pid, signal.SIGKILL) + except ProcessLookupError: + pass + process.wait(timeout=10.0) + raise + finally: + handle.close() + if process.returncode: + raise RuntimeError( + f"client failed: config={config['id']} role={role} rc={process.returncode}" + ) + result = json.loads((output / "result.json").read_text(encoding="utf-8")) + validate_result( + result, + config=config, + selection=selection, + role=role, + warmup=warmup, + ) + entry["results"].append( + {"anchor": float(selection["anchor"]), "dir": str(output), "kind": result["kind"]} + ) + return result + + +def execute_session( + *, + args: argparse.Namespace, + manifest: Mapping[str, Any], + config: Mapping[str, Any], + index: int, + state: dict[str, Any], + state_path: Path, +) -> None: + name = str(config["id"]) + if state["sessions"].get(name, {}).get("status") == "complete": + return + projection = remaining_projection( + manifest, completed_sessions=int(state["completed_sessions"]) + ) + if float(state["gpu_hours_total"]) + projection > base.GPU_LIMIT: + raise RuntimeError(f"projected cost exceeds cap before {name}") + load_values = { + float(item["selection"]["offered_req_s_per_gpu"]) + for item in manifest["repetitions"].values() + } + load_text = ( + f"{next(iter(load_values)):.6g}" + if len(load_values) == 1 + else ",".join(f"{value:.6g}" for value in sorted(load_values)) + ) + echo = ( + f"ACTION_AWARE_SESSION_ECHO host=dash0 config={name} tp=4 " + f"mns={config['mns']} mbbt={config['mbbt']} gpus=0-3 " + f"workload={manifest['source']['window_id']} load_per_gpu={load_text} " + f"duration_s={manifest['engine']['duration_s']} " + f"repetitions={','.join(map(str, config['repetition_order']))} " + f"source={args.manifest} output={args.run_root / 'sessions' / name} " + f"spent_h20h={state['gpu_hours_total']:.6f} " + f"remaining_projection_h20h={projection:.3f} cap_h20h={base.GPU_LIMIT:.1f}" + ) + append_echo(args.run_root, echo) + wait_all_idle() + session_state = { + "status": "starting", + "mns": int(config["mns"]), + "mbbt": int(config["mbbt"]), + "repetition_order": list(config["repetition_order"]), + "started_at": time.time(), + "runs": [], + } + state["status"] = "running" + state["sessions"][name] = session_state + atomic_json(state_path, state) + entry = start_server(args=args, config=config, index=index) + failure: Exception | None = None + try: + base.wait_ready(entry) + first = manifest["repetitions"][str(config["repetition_order"][0])] + session_state["status"] = "warmup" + atomic_json(state_path, state) + run_client( + entry=entry, + config=config, + role="warmup", + study=first["study"], + selection=first["selection"], + output=entry["dir"] / "warmup", + state=state, + timeout_s=180.0, + warmup=True, + ) + session_state["status"] = "burnin" + atomic_json(state_path, state) + burnin = manifest["burnin"] + burnin_result = run_client( + entry=entry, + config=config, + role="burnin", + study=burnin["study"], + selection=burnin, + output=entry["dir"] / "burnin", + state=state, + timeout_s=float(manifest["engine"]["client_timeout_s"]), + ) + session_state["burnin"] = burnin_gate( + burnin_result, + expected_count=int(burnin["selected_count"]), + maximum_elapsed_s=float(manifest["engine"]["burnin_max_elapsed_s"]), + ) + atomic_json(state_path, state) + session_state["status"] = "measured" + atomic_json(state_path, state) + for repetition in config["repetition_order"]: + item = manifest["repetitions"][str(repetition)] + role = f"rep{repetition}" + result = run_client( + entry=entry, + config=config, + role=role, + study=item["study"], + selection=item["selection"], + output=entry["dir"] / role, + state=state, + timeout_s=float(manifest["engine"]["client_timeout_s"]), + ) + session_state["runs"].append( + { + "repetition": int(repetition), + "pass_rate": result["pass_rate"], + "feasible": result["feasible"], + "slo_pass_count": result["slo_pass_count"], + "elapsed_s": result["interval"]["elapsed_s"], + } + ) + atomic_json(state_path, state) + session_state["status"] = "stopping" + atomic_json(state_path, state) + except Exception as error: # noqa: BLE001 + failure = error + finally: + try: + base.stop_entry(entry) + except Exception as error: # noqa: BLE001 + failure = failure or error + time.sleep(2.0) + try: + wait_all_idle() + except Exception as error: # noqa: BLE001 + failure = failure or error + + session_hours = base.live_gpu_hours([entry]) + state["gpu_hours_total"] += session_hours + session_state["gpu_hours"] = session_hours + if failure is not None: + session_state["status"] = "failed" + session_state["failure"] = repr(failure) + state["status"] = "failed" + state["failures"].append({"session": name, "failure": repr(failure)}) + atomic_json(state_path, state) + raise failure + validation = base.validate_cell(entry) + session_state["validation"] = validation + session_state["status"] = "complete" + session_state["completed_at"] = time.time() + state["completed_sessions"] += 1 + atomic_json(state_path, state) + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--manifest", type=Path, required=True) + result.add_argument("--run-root", type=Path, required=True) + result.add_argument("--aituner-root", type=Path, required=True) + result.add_argument("--vllm-source", type=Path, required=True) + result.add_argument("--venv", type=Path, required=True) + result.add_argument("--model", type=Path, required=True) + result.add_argument("--client", type=Path, required=True) + result.add_argument("--dry-run", action="store_true") + return result + + +def main() -> None: + args = parser().parse_args() + manifest = json.loads(args.manifest.read_text(encoding="utf-8")) + validate_inputs(args, manifest) + configure(args, manifest) + if args.dry_run: + print(json.dumps(dry_run_plan(args, manifest), indent=2, sort_keys=True)) + return + args.run_root.mkdir(parents=True, exist_ok=True) + copied_manifest = args.run_root / "pilot-manifest.json" + if not copied_manifest.exists(): + atomic_json(copied_manifest, manifest) + state_path = args.run_root / "controller-state.json" + state = load_state(state_path, base.GPU_LIMIT) + state["status"] = "running" + atomic_json(state_path, state) + for index, config in enumerate(manifest["configs"]): + execute_session( + args=args, + manifest=manifest, + config=config, + index=index, + state=state, + state_path=state_path, + ) + state["status"] = "complete" + state["completed_at"] = time.time() + atomic_json(state_path, state) + wait_all_idle() + print( + json.dumps( + { + "status": state["status"], + "completed_sessions": state["completed_sessions"], + "gpu_hours_total": state["gpu_hours_total"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/action-aware-v0/prepare_pilot.py b/runs/action-aware-v0/prepare_pilot.py new file mode 100644 index 0000000..c7f6aaf --- /dev/null +++ b/runs/action-aware-v0/prepare_pilot.py @@ -0,0 +1,195 @@ +#!/usr/bin/env python3 +"""Freeze the crossed-constraint action-aware development pilot.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import os +from pathlib import Path +from typing import Any + + +SCHEMA_V0 = "action-aware-constraint-pilot-manifest-v0" +SCHEMA_V1 = "action-aware-constraint-pilot-manifest-v1" + + +def configs(token_source_mbbt: int) -> tuple[dict[str, Any], ...]: + return ( + { + "id": "b_base", + "mns": 64, + "mbbt": token_source_mbbt, + "repetition_order": [1, 2, 3], + }, + {"id": "a_base", "mns": 16, "mbbt": 8192, "repetition_order": [2, 3, 1]}, + {"id": "shared", "mns": 64, "mbbt": 8192, "repetition_order": [3, 1, 2]}, + { + "id": "b_mns", + "mns": 128, + "mbbt": token_source_mbbt, + "repetition_order": [1, 3, 2], + }, + {"id": "a_mbbt", "mns": 16, "mbbt": 16384, "repetition_order": [2, 1, 3]}, + ) + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def build( + base_path: Path, + *, + token_source_mbbt: int = 256, + prior_attempt_h20_hours: float = 0.0, + prior_attempt_artifact: str | None = None, +) -> dict[str, Any]: + if token_source_mbbt <= 0: + raise ValueError("token source MBBT must be positive") + if prior_attempt_h20_hours < 0.0 or prior_attempt_h20_hours >= 8.0: + raise ValueError("prior attempt cost must be in [0, 8)") + base = json.loads(base_path.read_text(encoding="utf-8")) + if base.get("schema") != "intervention-response-phase-aware-pilot-manifest-v3": + raise ValueError("unexpected base manifest schema") + if base.get("status") != "PASS": + raise ValueError("base manifest did not pass its preflight") + if sorted(int(key) for key in base["repetitions"]) != [1, 2, 3]: + raise ValueError("base manifest must contain exactly three repetitions") + + repetitions = {} + selection_hashes = [] + for repetition in (1, 2, 3): + source = base["repetitions"][str(repetition)] + selection = dict(source["selections"]["mid"]) + selection_hashes.append(selection["request_id_order_sha256"]) + repetitions[str(repetition)] = { + "study": source["study"], + "study_sha256": source["study_sha256"], + "selection": selection, + "merged_trace": source["merged_trace"], + } + + frozen_configs = configs(token_source_mbbt) + config_ids = [str(config["id"]) for config in frozen_configs] + schema = ( + SCHEMA_V0 + if token_source_mbbt == 256 and prior_attempt_h20_hours == 0.0 + else SCHEMA_V1 + ) + payload = { + "schema": schema, + "status": "PASS", + "source": { + "base_manifest": str(base_path.resolve()), + "base_manifest_sha256": sha256_file(base_path), + "window_id": base["source"]["window_id"], + "source_trace": base["source"]["source_trace"], + "source_trace_sha256": base["source"]["source_trace_sha256"], + }, + "engine": { + "tp": 4, + "duration_s": 300.0, + "disable_slo_early_stop": True, + "client_timeout_s": 450.0, + "burnin_max_elapsed_s": 90.0, + }, + "burnin": base["burnin"], + "repetitions": repetitions, + "configs": [dict(config) for config in frozen_configs], + "regimes": { + "A": { + "source": "a_base", + "actions": {"mns": "shared", "mbbt": "a_mbbt"}, + }, + "B": { + "source": "b_base", + "actions": {"mns": "b_mns", "mbbt": "shared"}, + }, + }, + "budget": { + "global_hard_cap_h20_hours": 8.0, + "hard_cap_h20_hours": 8.0 - prior_attempt_h20_hours, + "prior_attempt_h20_hours": prior_attempt_h20_hours, + "prior_attempt_artifact": prior_attempt_artifact, + "session_estimate_h20_hours": 1.35, + "safety_h20_hours": 0.25, + "expected_h20_hours": [6.0, 7.2], + "expected_wall_minutes": [90, 110], + }, + "gates": { + "minimum_relative_winner_margin": 0.10, + "minimum_exclusive_fraction": 0.10, + "minimum_exclusive_ratio": 5.0, + "phase_fractions": [0.25, 0.50, 0.75, 1.0], + "material_kv_usage": 0.90, + }, + "sanity": { + "invariants": { + "five_unique_configs": len(config_ids) == len(set(config_ids)) == 5, + "three_disjoint_repetitions": len(set(selection_hashes)) == 3, + "same_load_all_repetitions": len( + { + float(item["selection"]["offered_req_s_per_gpu"]) + for item in repetitions.values() + } + ) + == 1, + "all_repetition_orders_are_permutations": all( + sorted(config["repetition_order"]) == [1, 2, 3] + for config in frozen_configs + ), + } + }, + } + payload["sanity"]["invariants"]["shared_endpoint_reused_by_both_regimes"] = ( + payload["regimes"]["A"]["actions"]["mns"] + == payload["regimes"]["B"]["actions"]["mbbt"] + == "shared" + ) + payload["sanity"]["red_flags"] = [ + name + for name, passed in payload["sanity"]["invariants"].items() + if not passed + ] + if payload["sanity"]["red_flags"]: + payload["status"] = "FAIL" + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--base-manifest", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--token-source-mbbt", type=int, default=256) + parser.add_argument("--prior-attempt-h20-hours", type=float, default=0.0) + parser.add_argument("--prior-attempt-artifact") + args = parser.parse_args() + payload = build( + args.base_manifest, + token_source_mbbt=args.token_source_mbbt, + prior_attempt_h20_hours=args.prior_attempt_h20_hours, + prior_attempt_artifact=args.prior_attempt_artifact, + ) + atomic_json(args.output, payload) + print(json.dumps(payload["sanity"], sort_keys=True)) + if payload["status"] != "PASS": + raise SystemExit("manifest preflight failed") + + +if __name__ == "__main__": + main() diff --git a/runs/action-aware-v0/prior-attempts-v2.json b/runs/action-aware-v0/prior-attempts-v2.json new file mode 100644 index 0000000..3faf060 --- /dev/null +++ b/runs/action-aware-v0/prior-attempts-v2.json @@ -0,0 +1,24 @@ +{ + "global_hard_cap_h20_hours": 8.0, + "invariants": { + "all_gpus_idle_after_each_stop": true, + "no_completed_measured_runs": true, + "no_prior_runtime_data_reused": true + }, + "prior_attempt_h20_hours": 0.7043978426191542, + "schema": "action-aware-prior-attempts-v2", + "stops": [ + { + "artifact": "/home/admin/cpfs/wjh/action-aware-constraint-v0-20260714/operational-stop-v0.json", + "h20_hours": 0.38598689953486126, + "reason": "MBBT256 burn-in remained throughput-backlogged", + "stage": "burnin" + }, + { + "artifact": "/home/admin/cpfs/wjh/action-aware-constraint-v1-20260714/operational-stop-v1.json", + "h20_hours": 0.31841094308429296, + "reason": "controller passed the warmup result to the burn-in gate", + "stage": "first measured run in flight; zero measured results completed" + } + ] +} diff --git a/runs/action-aware-v0/test_pilot.py b/runs/action-aware-v0/test_pilot.py new file mode 100644 index 0000000..3b25456 --- /dev/null +++ b/runs/action-aware-v0/test_pilot.py @@ -0,0 +1,292 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import copy +import importlib.util +from pathlib import Path +from types import SimpleNamespace + + +HERE = Path(__file__).resolve().parent +ROOT = HERE.parents[1] + + +def load(name: str, filename: str): + spec = importlib.util.spec_from_file_location(name, HERE / filename) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def record(*, waiting: int, running: int, tokens: int) -> dict: + return { + "queues": {"waiting": waiting, "deferred": 0, "running": running}, + "prefill_tokens": tokens, + "decode_tokens": 0, + "kv": {"usage": 0.5}, + "preemptions": 0, + } + + +def fake_run( + config: str, + repetition: int, + *, + goodput: float, + mns_score: float = 0.0, + mbbt_score: float = 0.0, + ambiguous: float = 0.0, +) -> dict: + binding = { + "mns_exclusive_fraction": mns_score, + "mbbt_exclusive_fraction": mbbt_score, + "both_fraction": ambiguous, + "waiting_unresolved_fraction": 0.0, + "kv_usage_max": 0.5, + "preemptions": 0, + } + phases = { + phase: { + "mns_exclusive_fraction": mns_score, + "mbbt_exclusive_fraction": mbbt_score, + } + for phase in ("0.25", "0.50", "0.75", "1.00") + } + return { + "config_id": config, + "repetition": repetition, + "outcome": {"slo_goodput_req_s": goodput}, + "binding": binding, + "phases": phases, + } + + +def main() -> None: + analysis = load("action_aware_analysis", "analyze_pilot.py") + summary = analysis.binding_summary( + [ + record(waiting=1, running=16, tokens=8), + record(waiting=1, running=8, tokens=32), + record(waiting=1, running=16, tokens=32), + record(waiting=1, running=8, tokens=8), + record(waiting=0, running=8, tokens=8), + ], + mns=16, + mbbt=32, + ) + assert summary["mns_exclusive_count"] == 1 + assert summary["mbbt_exclusive_count"] == 1 + assert summary["both_count"] == 1 + assert summary["waiting_unresolved_count"] == 1 + assert summary["waiting_count"] == 4 + + # A per-step stream may have a submit gap above one second when the + # preceding model execution itself spans that interval. Such a gap is + # covered telemetry, not a dropped-record interval. + asynchronous = [ + {"submit_mono_ns": 0, "complete_mono_ns": 1_200_000_000}, + {"submit_mono_ns": 1_100_000_000, "complete_mono_ns": 1_300_000_000}, + ] + coverage, covered = analysis.telemetry_coverage( + asynchronous, start_ns=0, end_ns=1_100_000_000 + ) + assert coverage["max_internal_submit_gap_s"] == 1.1 + assert coverage["max_uncovered_gap_s"] == 0.0 + assert covered + missing = copy.deepcopy(asynchronous) + missing[0]["complete_mono_ns"] = 0 + assert not analysis.telemetry_coverage( + missing, start_ns=0, end_ns=1_100_000_000 + )[1] + + mechanism = analysis.mechanism_summary( + [ + { + "model_executed": True, + "submit_mono_ns": 0, + "complete_mono_ns": 2_000_000, + "prefill_tokens": 8, + "prefill_requests": 2, + "chunked_prefill": { + "first": 1, + "middle": 0, + "final": 0, + "unsplit": 1, + "tokens": 8, + }, + "prefix": {"local": {"queries": 10, "hits": 2}}, + }, + { + "model_executed": True, + "submit_mono_ns": 2_000_000, + "complete_mono_ns": 3_000_000, + "prefill_tokens": 0, + "prefill_requests": 0, + "chunked_prefill": { + "first": 0, + "middle": 0, + "final": 0, + "unsplit": 0, + "tokens": 0, + }, + "prefix": {"local": {"queries": 0, "hits": 0}}, + }, + ] + ) + assert mechanism["prefill"]["requests_per_step"] == 2.0 + assert mechanism["prefill"]["chunks"]["first"] == 1 + assert mechanism["prefix"]["hit_rate"] == 0.2 + assert all(mechanism["sanity"]["invariants"].values()) + + manifest = { + "repetitions": {str(index): {} for index in (1, 2, 3)}, + "regimes": { + "A": { + "source": "a_base", + "actions": {"mns": "shared", "mbbt": "a_mbbt"}, + }, + "B": { + "source": "b_base", + "actions": {"mns": "b_mns", "mbbt": "shared"}, + }, + }, + "gates": { + "minimum_relative_winner_margin": 0.10, + "minimum_exclusive_fraction": 0.10, + "minimum_exclusive_ratio": 5.0, + "material_kv_usage": 0.90, + }, + } + runs = [] + for repetition in (1, 2, 3): + runs.extend( + [ + fake_run( + "a_base", + repetition, + goodput=1.0, + mns_score=0.8, + mbbt_score=0.01, + ), + fake_run( + "b_base", + repetition, + goodput=1.0, + mns_score=0.01, + mbbt_score=0.7, + ), + fake_run("shared", repetition, goodput=3.0), + fake_run("a_mbbt", repetition, goodput=1.5), + fake_run("b_mns", repetition, goodput=1.2), + ] + ) + result = analysis.evaluate_decisions(runs, manifest) + assert result["decision"] == "STOP_NO_NEW_INSTRUMENTATION_NEEDED" + assert result["baselines"] == { + "always_mns_correct": 3, + "always_mbbt_correct": 3, + "binding_correct": 6, + "decision_count": 6, + } + + ambiguous = copy.deepcopy(runs) + for run in ambiguous: + if run["config_id"] == "b_base": + run["binding"]["both_fraction"] = 0.8 + assert ( + analysis.evaluate_decisions(ambiguous, manifest)["decision"] + == "OPEN_EXACT_ATTRIBUTION_ABLATION" + ) + + wrong = copy.deepcopy(runs) + for run in wrong: + if run["config_id"] == "b_base": + run["binding"]["mns_exclusive_fraction"] = 0.8 + run["binding"]["mbbt_exclusive_fraction"] = 0.01 + for phase in run["phases"].values(): + phase["mns_exclusive_fraction"] = 0.8 + phase["mbbt_exclusive_fraction"] = 0.01 + assert ( + analysis.evaluate_decisions(wrong, manifest)["decision"] + == "STOP_BINDING_NOT_PREDICTIVE" + ) + + prepare = load("action_aware_prepare", "prepare_pilot.py") + frozen = prepare.build( + ROOT / "runs/intervention-response-v2/pilot-manifest-v3.json" + ) + assert frozen["status"] == "PASS" + assert frozen["sanity"]["red_flags"] == [] + assert [config["id"] for config in frozen["configs"]] == [ + "b_base", + "a_base", + "shared", + "b_mns", + "a_mbbt", + ] + + controller = load("action_aware_controller", "pilot_controller.py") + args = SimpleNamespace( + manifest=Path("/tmp/manifest.json"), + run_root=Path("/tmp/action-aware"), + aituner_root=Path("/tmp/aituner"), + vllm_source=Path("/tmp/vllm"), + venv=Path("/tmp/venv"), + model=Path("/tmp/model"), + client=Path("/tmp/client.py"), + ) + controller.configure(args, frozen) + plan = controller.dry_run_plan(args, frozen) + assert plan["status"] == "PASS" + assert len(plan["sessions"]) == 5 + assert plan["projected_h20_hours"] == 7.0 + assert "--max-num-batched-tokens 256" in plan["sessions"][0]["commands"]["server"] + revised = prepare.build( + ROOT / "runs/intervention-response-v2/pilot-manifest-v3.json", + token_source_mbbt=2048, + prior_attempt_h20_hours=0.38598689953486126, + prior_attempt_artifact="/tmp/operational-stop-v0.json", + ) + assert revised["schema"] == "action-aware-constraint-pilot-manifest-v1" + assert revised["configs"][0]["mbbt"] == 2048 + assert revised["configs"][3]["mbbt"] == 2048 + assert revised["budget"]["hard_cap_h20_hours"] < 8.0 + controller.configure(args, revised) + revised_plan = controller.dry_run_plan(args, revised) + assert revised_plan["projected_h20_hours"] < revised_plan["hard_cap_h20_hours"] + assert ( + "--max-num-batched-tokens 2048" + in revised_plan["sessions"][0]["commands"]["server"] + ) + accepted_burnin = { + "kind": "anchor", + "selection": {"count": 510}, + "interval": {"elapsed_s": 61.25}, + "pass_rate": 0.5, + "feasible": False, + } + assert controller.burnin_gate( + accepted_burnin, expected_count=510, maximum_elapsed_s=90.0 + )["elapsed_s"] == 61.25 + warmup = copy.deepcopy(accepted_burnin) + warmup["kind"] = "warmup" + try: + controller.burnin_gate(warmup, expected_count=510, maximum_elapsed_s=90.0) + except RuntimeError as error: + assert "non-anchor" in str(error) + else: + raise AssertionError("warmup incorrectly passed the burnin gate") + slow = copy.deepcopy(accepted_burnin) + slow["interval"]["elapsed_s"] = 91.0 + try: + controller.burnin_gate(slow, expected_count=510, maximum_elapsed_s=90.0) + except RuntimeError as error: + assert "throughput gate failed" in str(error) + else: + raise AssertionError("slow burnin incorrectly passed the throughput gate") + print("action-aware constraint pilot: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/analyze_prospective.py b/runs/active-intervention-v0/analyze_prospective.py new file mode 100644 index 0000000..3442769 --- /dev/null +++ b/runs/active-intervention-v0/analyze_prospective.py @@ -0,0 +1,325 @@ +#!/usr/bin/env python3 +"""Audit held-out action/measurement choices against the exact 2x2 surface.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import statistics +from pathlib import Path +from typing import Any, Mapping + + +SCHEMA = "active-intervention-prospective-audit-v0" + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def numeric(values: list[float]) -> dict[str, Any]: + finite = [float(value) for value in values] + if not finite or any(not math.isfinite(value) for value in finite): + raise ValueError("numeric summary requires finite values") + return { + "n": len(finite), + "min": min(finite), + "max": max(finite), + "distinct_n": len(set(finite)), + } + + +def load_surface( + manifest: Mapping[str, Any], run_root: Path +) -> tuple[dict[str, Any], list[dict[str, Any]]]: + rows = [] + aggregate = {} + duration_s = float(manifest["engine"]["duration_s"]) + tp = int(manifest["engine"]["tp"]) + for config in manifest["configs"]: + config_id = str(config["id"]) + values = [] + for repetition in sorted(int(key) for key in manifest["repetitions"]): + expected = manifest["repetitions"][str(repetition)]["selection"] + result_path = ( + run_root / "sessions" / config_id / f"rep{repetition}" / "result.json" + ) + result = json.loads(result_path.read_text(encoding="utf-8")) + if result["selection"]["request_id_order_sha256"] != expected[ + "request_id_order_sha256" + ]: + raise ValueError(f"request hash mismatch: {config_id} rep{repetition}") + offered_total = float(expected["offered_req_s_per_gpu"]) * tp + normalized = float(result["slo_pass_count"]) / duration_s / offered_total + values.append(normalized) + rows.append( + { + "config_id": config_id, + "mns": int(config["mns"]), + "mbbt": int(config["mbbt"]), + "repetition": repetition, + "normalized_slo_goodput": normalized, + "slo_goodput_req_s": float(result["slo_pass_count"]) / duration_s, + "pass_rate": float(result["pass_rate"]), + "elapsed_s": float(result["interval"]["elapsed_s"]), + "result": str(result_path), + "result_sha256": sha256_file(result_path), + } + ) + aggregate[config_id] = { + "normalized_slo_goodput_values": values, + "median_normalized_slo_goodput": float(statistics.median(values)), + "sanity": numeric(values), + } + return aggregate, rows + + +def source_cost_estimate( + *, + source_session: Mapping[str, Any], + source_rows: list[Mapping[str, Any]], + cutoff_s: float, + tp: int, +) -> dict[str, float]: + actual_h20_hours = float(source_session["gpu_hours"]) + measured_replay_h20_hours = ( + tp * sum(float(row["elapsed_s"]) for row in source_rows) / 3600.0 + ) + fixed_h20_hours = max(0.0, actual_h20_hours - measured_replay_h20_hours) + prefix_replay_h20_hours = tp * len(source_rows) * cutoff_s / 3600.0 + return { + "actual_full_session_h20_hours": actual_h20_hours, + "fixed_startup_warmup_burnin_cleanup_h20_hours": fixed_h20_hours, + "prefix_replay_h20_hours_lower_bound": prefix_replay_h20_hours, + "counterfactual_all_in_h20_hours_lower_bound": fixed_h20_hours + + prefix_replay_h20_hours, + } + + +def replay_policy( + *, + mode: str, + manifest: Mapping[str, Any], + decision: Mapping[str, Any], + surface: Mapping[str, Any], + session_costs: Mapping[str, float], + source_cost: Mapping[str, float], +) -> dict[str, Any]: + acceptable_regret = float(manifest["gates"]["acceptable_regret"]) + source_id = str(manifest["source_config_id"]) + oracle = max( + float(item["median_normalized_slo_goodput"]) for item in surface.values() + ) + cumulative = float(source_cost["counterfactual_all_in_h20_hours_lower_bound"]) + source_score = float(surface[source_id]["median_normalized_slo_goodput"]) + source_regret = 1.0 - source_score / oracle if oracle > 0 else 0.0 + points = [ + { + "action_id": "noop", + "config_id": source_id, + "score": source_score, + "regret": source_regret, + "cumulative_h20_hours_lower_bound": cumulative, + } + ] + hit = points[0] if source_regret <= acceptable_regret + 1e-12 else None + seen = {source_id} + for action_id in decision["decisions"][mode]["intervention_order"]: + config_id = str(manifest["actions"][action_id]) + if config_id in seen: + continue + seen.add(config_id) + cumulative += float(session_costs[config_id]) + score = float(surface[config_id]["median_normalized_slo_goodput"]) + regret = 1.0 - score / oracle if oracle > 0 else 0.0 + point = { + "action_id": action_id, + "config_id": config_id, + "score": score, + "regret": regret, + "cumulative_h20_hours_lower_bound": cumulative, + } + points.append(point) + if hit is None and regret <= acceptable_regret + 1e-12: + hit = point + return { + "mode": mode, + "measurement_cutoff_s": float( + decision["decisions"][mode]["selected_cutoff_s"] + ), + "selected_action": decision["decisions"][mode]["selected_action"], + "decision_kind": decision["decisions"][mode]["decision_kind"], + "intervention_order": decision["decisions"][mode]["intervention_order"], + "source_cost": dict(source_cost), + "oracle_normalized_slo_goodput": oracle, + "cost_to_acceptable": hit, + "reached_acceptable": hit is not None, + "points": points, + } + + +def build_audit( + *, manifest_path: Path, decision_path: Path, run_root: Path +) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + decision = json.loads(decision_path.read_text(encoding="utf-8")) + state_path = run_root / "controller-state.json" + state = json.loads(state_path.read_text(encoding="utf-8")) + if manifest.get("schema") != "active-intervention-prospective-manifest-v0": + raise ValueError("unexpected prospective manifest schema") + if decision.get("schema") != "active-intervention-prospective-decision-v0": + raise ValueError("unexpected prospective decision schema") + if decision["manifest_sha256"] != sha256_file(manifest_path): + raise ValueError("decision does not match prospective manifest") + surface, rows = load_surface(manifest, run_root) + source_id = str(manifest["source_config_id"]) + sessions = state["sessions"] + session_costs = { + config_id: float(sessions[config_id]["gpu_hours"]) + for config_id in surface + } + source_rows = [row for row in rows if row["config_id"] == source_id] + policies = {} + for mode in ("outcome_only", "telemetry"): + cost = source_cost_estimate( + source_session=sessions[source_id], + source_rows=source_rows, + cutoff_s=float(decision["decisions"][mode]["selected_cutoff_s"]), + tp=int(manifest["engine"]["tp"]), + ) + policies[mode] = replay_policy( + mode=mode, + manifest=manifest, + decision=decision, + surface=surface, + session_costs=session_costs, + source_cost=cost, + ) + outcome_hit = policies["outcome_only"]["cost_to_acceptable"] + telemetry_hit = policies["telemetry"]["cost_to_acceptable"] + if outcome_hit is None or telemetry_hit is None: + reduction = None + else: + outcome_cost = float(outcome_hit["cumulative_h20_hours_lower_bound"]) + telemetry_cost = float(telemetry_hit["cumulative_h20_hours_lower_bound"]) + reduction = 1.0 - telemetry_cost / outcome_cost if outcome_cost > 0 else 0.0 + confirmation_trigger = bool( + reduction is not None + and reduction + >= float(manifest["gates"]["confirmation_trigger_gpu_cost_reduction"]) + and policies["telemetry"]["reached_acceptable"] + ) + contribution_gate = bool( + reduction is not None + and reduction >= float(manifest["gates"]["contribution_gpu_cost_reduction"]) + and policies["telemetry"]["reached_acceptable"] + ) + status = ( + "TRIGGER_ACTUAL_EARLY_STOP_CONFIRMATION" + if confirmation_trigger + else "STOP_NO_PROSPECTIVE_GPU_COST_SIGNAL" + ) + normalized_values = [float(row["normalized_slo_goodput"]) for row in rows] + costs = list(session_costs.values()) + invariants = { + "controller_complete": state.get("status") == "complete", + "four_sessions_complete": len(sessions) == 4 + and all(item.get("status") == "complete" for item in sessions.values()), + "twelve_surface_outcomes": len(rows) == 12, + "nonnegative_goodput": all(value >= 0.0 for value in normalized_values), + "normalized_goodput_bounded": all(value <= 1.0 + 1e-12 for value in normalized_values), + "surface_not_all_identical": len(set(normalized_values)) > 1, + "nonnegative_session_costs": all(value >= 0.0 for value in costs), + "policy_replay_reaches_oracle_surface": all( + policy["reached_acceptable"] for policy in policies.values() + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + if red_flags: + status = "STOP_SANITY" + return { + "schema": SCHEMA, + "status": status, + "claim_boundary": ( + "Prospective exact-surface replay. Prefix source costs reconstruct the " + "measured fixed overhead plus selected replay seconds; actual early-stop " + "confirmation is required before claiming GPU-cost reduction." + ), + "manifest": str(manifest_path), + "manifest_sha256": sha256_file(manifest_path), + "decision": str(decision_path), + "decision_sha256": sha256_file(decision_path), + "controller_state": str(state_path), + "controller_state_sha256": sha256_file(state_path), + "surface": surface, + "rows": rows, + "session_costs_h20_hours": session_costs, + "annotation_campaign_h20_hours": float(state["gpu_hours_total"]), + "policies": policies, + "comparison": { + "telemetry_gpu_cost_reduction_fraction": reduction, + "confirmation_trigger": confirmation_trigger, + "contribution_gate": contribution_gate, + "confirmation_trigger_threshold": manifest["gates"][ + "confirmation_trigger_gpu_cost_reduction" + ], + "contribution_threshold": manifest["gates"][ + "contribution_gpu_cost_reduction" + ], + "action_changed": policies["outcome_only"]["selected_action"] + != policies["telemetry"]["selected_action"], + "measurement_changed": policies["outcome_only"]["measurement_cutoff_s"] + != policies["telemetry"]["measurement_cutoff_s"], + }, + "sanity": { + "invariants": invariants, + "red_flags": red_flags, + "normalized_slo_goodput": numeric(normalized_values), + "session_h20_hours": numeric(costs), + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--decision", type=Path, required=True) + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + audit = build_audit( + manifest_path=args.manifest, + decision_path=args.decision, + run_root=args.run_root, + ) + atomic_json(args.output, audit) + print( + json.dumps( + { + "status": audit["status"], + "comparison": audit["comparison"], + "sanity": audit["sanity"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/extract_training.py b/runs/active-intervention-v0/extract_training.py new file mode 100644 index 0000000..8f7c8bf --- /dev/null +++ b/runs/active-intervention-v0/extract_training.py @@ -0,0 +1,324 @@ +#!/usr/bin/env python3 +"""Extract paired source/action examples from the accepted action-aware run.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import sys +from pathlib import Path +from statistics import fmean +from typing import Any, Mapping + + +PHASES = ("0.25", "0.50", "0.75", "1.00") +HERE = Path(__file__).resolve().parent +COMMON_STATE = HERE.parent / "telemetry-residual" +sys.path.insert(0, str(COMMON_STATE)) + +from common_state import summarize_engine # noqa: E402 + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def load_jsonl(path: Path) -> list[dict[str, Any]]: + records = [] + with path.open(encoding="utf-8") as source: + for line_number, line in enumerate(source, 1): + if not line.strip(): + continue + try: + records.append(json.loads(line)) + except json.JSONDecodeError as error: + raise ValueError(f"{path}:{line_number}: invalid JSON") from error + if not records: + raise ValueError(f"{path}: no request records") + return records + + +def prefix_outcome( + requests: list[Mapping[str, Any]], *, cutoff_s: float, offered_total: float +) -> dict[str, float]: + admitted = [request for request in requests if float(request["arrival_s"]) <= cutoff_s] + completed = [ + request + for request in requests + if request.get("completed_elapsed_s") is not None + and float(request["completed_elapsed_s"]) <= cutoff_s + ] + if not admitted: + raise ValueError("prefix has no admitted requests") + admitted_ids = {str(request["request_id"]) for request in admitted} + if any(str(request["request_id"]) not in admitted_ids for request in completed): + raise ValueError("prefix completion precedes admission") + passed = sum(bool(request["slo_pass"]) for request in completed) + ttft = [float(request["ttft_ms"]) for request in completed] + tpot = [float(request["tpot_ms"]) for request in completed] + total = len(requests) + return { + "normalized_slo_goodput": passed / cutoff_s / offered_total, + "admitted_fraction": len(admitted) / total, + "completed_over_admitted": len(completed) / len(admitted), + "completed_pass_rate": passed / max(1, len(completed)), + "completed_fail_fraction_of_total": (len(completed) - passed) / total, + "outstanding_over_admitted": (len(admitted) - len(completed)) / len(admitted), + "ttft_max_over_slo_max": max(ttft, default=0.0) / 6000.0, + "ttft_mean_over_slo_max": fmean(ttft) / 6000.0 if ttft else 0.0, + "tpot_max_over_slo": max(tpot, default=0.0) / 50.0, + "tpot_mean_over_slo": fmean(tpot) / 50.0 if tpot else 0.0, + "admitted_input_tokens_mean_over_limit": fmean( + float(request["raw_input_tokens"]) for request in admitted + ) + / 8192.0, + } + + +def telemetry_record(state: Mapping[str, Any]) -> dict[str, float]: + common = state["common"] + engine = state["engine_only"] + executed_steps = int(state["sanity"]["executed_steps"]) + if executed_steps <= 0: + raise ValueError("telemetry phase contains no executed engine steps") + return { + "scheduler_steps_per_s": float(common["scheduler_steps_per_s"]), + "batch_size_mean": float(common["batch_size"]["mean"]), + "batch_size_cv": float(common["batch_size"]["cv"]), + "batch_tokens_mean": float(common["batch_tokens"]["mean"]), + "batch_tokens_cv": float(common["batch_tokens"]["cv"]), + "decode_batch_size_mean": float(common["decode_batch_size"]["mean"]), + "decode_batch_size_cv": float(common["decode_batch_size"]["cv"]), + "prefill_token_fraction": float(common["prefill_token_fraction"]), + "queue_waiting_mean": float(common["queue_waiting_mean"]), + "queue_running_mean": float(common["queue_running_mean"]), + "preemptions_per_step": float(common["preemptions"]) / executed_steps, + "kv_usage_mean": float(engine["kv_usage_mean"]), + "kv_usage_max": float(engine["kv_usage_max"]), + "kv_usage_end_minus_start": float(engine["kv_usage_end_minus_start"]), + "graph_none_share": float(engine["graph_none_share"]), + "graph_full_share": float(engine["graph_full_share"]), + "graph_padding_fraction": float(engine["graph_padding_fraction"]), + } + + +def load_stream(path: Path, *, expected_sha256: str) -> list[dict[str, Any]]: + if sha256_file(path) != expected_sha256: + raise ValueError(f"engine stream hash mismatch: {path}") + decoded = load_jsonl(path) + records = [row for row in decoded if "step_index" in row] + if not records: + raise ValueError(f"engine stream has no Layer-1 records: {path}") + return records + + +def build_dataset( + *, audit_path: Path, manifest_path: Path, run_root: Path +) -> dict[str, Any]: + audit = json.loads(audit_path.read_text(encoding="utf-8")) + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + if audit.get("schema") != "action-aware-constraint-pilot-audit-v0": + raise ValueError("unexpected action-aware audit schema") + if audit["sanity"]["red_flags"]: + raise ValueError(f"action-aware audit red flags: {audit['sanity']['red_flags']}") + configs = {str(item["id"]): item for item in manifest["configs"]} + runs = { + (str(run["config_id"]), int(run["repetition"])): run + for run in audit["runs"] + } + source_ids = {str(regime["source"]) for regime in manifest["regimes"].values()} + stream_entries = { + str(item["config_id"]): item + for item in audit["streams"] + if str(item["config_id"]) in source_ids + } + if set(stream_entries) != source_ids: + raise ValueError("audit is missing a source config engine stream") + streams = { + config_id: load_stream( + Path(item["stream"]), expected_sha256=str(item["stream_sha256"]) + ) + for config_id, item in stream_entries.items() + } + examples = [] + request_hashes = [] + for regime_name, regime in sorted(manifest["regimes"].items()): + source_id = str(regime["source"]) + for repetition in sorted(int(value) for value in manifest["repetitions"]): + source_run = runs[(source_id, repetition)] + source_config = configs[source_id] + request_path = run_root / "sessions" / source_id / f"rep{repetition}" / "requests.jsonl" + requests = load_jsonl(request_path) + request_hashes.append(sha256_file(request_path)) + offered_rate_per_gpu = float( + manifest["repetitions"][str(repetition)]["selection"][ + "offered_req_s_per_gpu" + ] + ) + offered_total = offered_rate_per_gpu * int(manifest["engine"]["tp"]) + source_goodput = float(source_run["outcome"]["slo_goodput_req_s"]) + source_normalized = min(1.0, source_goodput / offered_total) + decision_id = f"{regime_name}-rep{repetition}" + for phase in PHASES: + cutoff_s = float(manifest["engine"]["duration_s"]) * float(phase) + outcome = prefix_outcome( + requests, cutoff_s=cutoff_s, offered_total=offered_total + ) + admitted_count = sum( + float(request["arrival_s"]) <= cutoff_s for request in requests + ) + start_ns = int(source_run["state"]["interval"]["start_ns"]) + phase_state = summarize_engine( + streams[source_id], + start_ns=start_ns, + end_ns=start_ns + round(cutoff_s * 1e9), + request_count=admitted_count, + ) + if not all(phase_state["sanity"]["invariants"].values()): + raise ValueError( + f"engine state invariant failed: {decision_id} phase {phase}" + ) + telemetry = telemetry_record(phase_state) + actions = {"noop": source_id, **regime["actions"]} + for action_name, target_id in sorted(actions.items()): + target_run = runs[(str(target_id), repetition)] + target_config = configs[str(target_id)] + target_goodput = float(target_run["outcome"]["slo_goodput_req_s"]) + normalized = target_goodput / offered_total + if not 0.0 <= normalized <= 1.0 + 1e-12: + raise ValueError("target normalized goodput is outside [0, 1]") + examples.append( + { + "phase": phase, + "cutoff_s": cutoff_s, + "decision_id": decision_id, + "regime": regime_name, + "repetition": repetition, + "source": { + "config_id": source_id, + "mns": int(source_config["mns"]), + "mbbt": int(source_config["mbbt"]), + "offered_rate_per_gpu": offered_rate_per_gpu, + "outcome": outcome, + "telemetry": telemetry, + }, + "action": { + "id": action_name, + "target_config_id": str(target_id), + "target_mns": int(target_config["mns"]), + "target_mbbt": int(target_config["mbbt"]), + }, + "target_slo_goodput_req_s": target_goodput, + "target_normalized_goodput": min(1.0, normalized), + "source_normalized_goodput": source_normalized, + "target_delta_normalized_goodput": min(1.0, normalized) + - source_normalized, + } + ) + invariants = { + "expected_examples": len(examples) == len(PHASES) * 2 * 3 * 3, + "four_phases": sorted({example["phase"] for example in examples}) + == sorted(PHASES), + "six_decisions": len({example["decision_id"] for example in examples}) == 6, + "three_actions_per_decision_phase": all( + sum( + item["decision_id"] == decision + and item["phase"] == phase + for item in examples + ) + == 3 + for decision in {item["decision_id"] for item in examples} + for phase in PHASES + ), + "targets_not_all_identical": len( + {example["target_normalized_goodput"] for example in examples} + ) + > 1, + "bounded_prefix_ratios": all( + 0.0 <= float(value) <= 1.0 + for example in examples + for key, value in example["source"]["outcome"].items() + if key + in { + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + } + ), + "direct_telemetry_without_binding_labels": all( + not any(token in key for token in ("exclusive", "unresolved", "both")) + for example in examples + for key in example["source"]["telemetry"] + ), + "treatment_effects_bounded": all( + -1.0 <= float(example["target_delta_normalized_goodput"]) <= 1.0 + for example in examples + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + if red_flags: + raise RuntimeError(f"training dataset sanity failed: {red_flags}") + return { + "schema": "active-intervention-training-v0", + "status": "VALID", + "provenance": { + "audit": str(audit_path), + "audit_sha256": sha256_file(audit_path), + "manifest": str(manifest_path), + "manifest_sha256": sha256_file(manifest_path), + "run_root": str(run_root), + "source_request_sha256": sorted(set(request_hashes)), + "source_stream_sha256": sorted( + str(item["stream_sha256"]) for item in stream_entries.values() + ), + }, + "examples": examples, + "sanity": {"invariants": invariants, "red_flags": red_flags}, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--audit", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + dataset = build_dataset( + audit_path=args.audit, + manifest_path=args.manifest, + run_root=args.run_root, + ) + atomic_json(args.output, dataset) + print( + json.dumps( + { + "status": dataset["status"], + "examples": len(dataset["examples"]), + "sanity": dataset["sanity"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/model.py b/runs/active-intervention-v0/model.py new file mode 100644 index 0000000..fe815e6 --- /dev/null +++ b/runs/active-intervention-v0/model.py @@ -0,0 +1,287 @@ +#!/usr/bin/env python3 +"""Small-data action-response model for the active intervention pilot. + +The model predicts the paired normalized SLO-goodput treatment effect from a +source measurement and a full MNS/MBBT action. Telemetry features are direct, +continuous engine measurements; there is no diagnosis-to-action rule here. +""" + +from __future__ import annotations + +import math +from dataclasses import dataclass +from typing import Any, Iterable, Mapping, Sequence + +import numpy as np + + +PREFIX_FEATURES = ( + "normalized_slo_goodput", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit", +) + +TELEMETRY_FEATURES = ( + "scheduler_steps_per_s", + "batch_size_mean", + "batch_size_cv", + "batch_tokens_mean", + "batch_tokens_cv", + "decode_batch_size_mean", + "decode_batch_size_cv", + "prefill_token_fraction", + "queue_waiting_mean", + "queue_running_mean", + "preemptions_per_step", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "graph_padding_fraction", +) + + +def _finite(value: Any, name: str) -> float: + result = float(value) + if not math.isfinite(result): + raise ValueError(f"{name} must be finite") + return result + + +def feature_vector( + example: Mapping[str, Any], *, include_telemetry: bool +) -> tuple[list[str], np.ndarray]: + source = example["source"] + action = example["action"] + source_log_mns = math.log2(_finite(source["mns"], "source MNS")) + source_log_mbbt = math.log2(_finite(source["mbbt"], "source MBBT")) + target_log_mns = math.log2(_finite(action["target_mns"], "target MNS")) + target_log_mbbt = math.log2(_finite(action["target_mbbt"], "target MBBT")) + delta_mns = target_log_mns - source_log_mns + delta_mbbt = target_log_mbbt - source_log_mbbt + names = [ + "source_log2_mns", + "source_log2_mbbt", + "target_log2_mns", + "target_log2_mbbt", + "delta_log2_mns", + "delta_log2_mbbt", + "delta_product", + "offered_rate_per_gpu", + ] + values = [ + source_log_mns, + source_log_mbbt, + target_log_mns, + target_log_mbbt, + delta_mns, + delta_mbbt, + delta_mns * delta_mbbt, + _finite(source["offered_rate_per_gpu"], "offered rate"), + ] + for name in PREFIX_FEATURES: + names.append(f"outcome.{name}") + values.append(_finite(source["outcome"][name], name)) + if include_telemetry: + for name in TELEMETRY_FEATURES: + value = _finite(source["telemetry"][name], name) + names.extend( + ( + f"telemetry.{name}", + f"telemetry.{name}*delta_mns", + f"telemetry.{name}*delta_mbbt", + ) + ) + values.extend((value, value * delta_mns, value * delta_mbbt)) + vector = np.asarray(values, dtype=np.float64) + if not np.all(np.isfinite(vector)): + raise ValueError("feature vector contains a non-finite value") + return names, vector + + +@dataclass(frozen=True) +class RidgeModel: + feature_names: tuple[str, ...] + mean: np.ndarray + scale: np.ndarray + weights: np.ndarray + intercept: float + regularization: float + + def predict(self, values: np.ndarray) -> float: + if values.shape != self.mean.shape: + raise ValueError("ridge prediction feature shape mismatch") + normalized = (values - self.mean) / self.scale + return float(self.intercept + normalized @ self.weights) + + def to_json(self) -> dict[str, Any]: + return { + "feature_names": list(self.feature_names), + "mean": self.mean.tolist(), + "scale": self.scale.tolist(), + "weights": self.weights.tolist(), + "intercept": self.intercept, + "regularization": self.regularization, + } + + @classmethod + def from_json(cls, payload: Mapping[str, Any]) -> "RidgeModel": + return cls( + feature_names=tuple(str(value) for value in payload["feature_names"]), + mean=np.asarray(payload["mean"], dtype=np.float64), + scale=np.asarray(payload["scale"], dtype=np.float64), + weights=np.asarray(payload["weights"], dtype=np.float64), + intercept=float(payload["intercept"]), + regularization=float(payload["regularization"]), + ) + + +def fit_ridge( + examples: Sequence[Mapping[str, Any]], + *, + include_telemetry: bool, + regularization: float, +) -> RidgeModel: + if not examples: + raise ValueError("ridge fit requires examples") + if regularization <= 0: + raise ValueError("ridge regularization must be positive") + encoded = [ + feature_vector(example, include_telemetry=include_telemetry) + for example in examples + ] + names = encoded[0][0] + if any(item[0] != names for item in encoded): + raise ValueError("feature names changed across examples") + x = np.stack([item[1] for item in encoded]) + y = np.asarray( + [ + _finite(example["target_delta_normalized_goodput"], "target effect") + for example in examples + ], + dtype=np.float64, + ) + mean = x.mean(axis=0) + scale = x.std(axis=0) + scale[scale < 1e-12] = 1.0 + normalized = (x - mean) / scale + intercept = float(y.mean()) + centered = y - intercept + system = normalized.T @ normalized + regularization * np.eye(x.shape[1]) + weights = np.linalg.solve(system, normalized.T @ centered) + return RidgeModel( + feature_names=tuple(names), + mean=mean, + scale=scale, + weights=weights, + intercept=intercept, + regularization=regularization, + ) + + +def fit_jackknife_ensemble( + examples: Sequence[Mapping[str, Any]], + *, + include_telemetry: bool, + regularization: float, + group_key: str = "decision_id", +) -> list[RidgeModel]: + groups = sorted({str(example[group_key]) for example in examples}) + if len(groups) < 3: + raise ValueError("jackknife ensemble requires at least three groups") + models = [] + for held_out in groups: + training = [ + example for example in examples if str(example[group_key]) != held_out + ] + models.append( + fit_ridge( + training, + include_telemetry=include_telemetry, + regularization=regularization, + ) + ) + return models + + +def ensemble_predict( + models: Sequence[RidgeModel], + example: Mapping[str, Any], + *, + include_telemetry: bool, +) -> dict[str, float]: + if not models: + raise ValueError("ensemble prediction requires models") + source = example["source"] + action = example["action"] + if ( + int(action["target_mns"]) == int(source["mns"]) + and int(action["target_mbbt"]) == int(source["mbbt"]) + ): + return {"mean": 0.0, "std": 0.0, "min": 0.0, "max": 0.0, "distinct_n": 1} + names, values = feature_vector(example, include_telemetry=include_telemetry) + if any(model.feature_names != tuple(names) for model in models): + raise ValueError("ensemble feature schema mismatch") + raw = np.asarray([model.predict(values) for model in models], dtype=np.float64) + clipped = np.clip(raw, -1.0, 1.0) + return { + "mean": float(clipped.mean()), + "std": float(clipped.std(ddof=0)), + "min": float(clipped.min()), + "max": float(clipped.max()), + "distinct_n": len(set(float(value) for value in clipped)), + } + + +def select_action( + models: Sequence[RidgeModel], + candidates: Sequence[Mapping[str, Any]], + *, + include_telemetry: bool, + confidence_z: float = 1.0, + minimum_margin: float = 0.02, +) -> dict[str, Any]: + if len(candidates) < 2: + raise ValueError("action selection requires at least two candidates") + rows = [] + for example in candidates: + prediction = ensemble_predict( + models, example, include_telemetry=include_telemetry + ) + rows.append( + { + "action_id": str(example["action"]["id"]), + "prediction": prediction, + "lower": prediction["mean"] - confidence_z * prediction["std"], + "upper": prediction["mean"] + confidence_z * prediction["std"], + } + ) + rows.sort(key=lambda row: (-row["prediction"]["mean"], row["action_id"])) + best, second = rows[:2] + margin = float(best["prediction"]["mean"] - second["prediction"]["mean"]) + confident = bool( + margin >= minimum_margin and best["lower"] > second["upper"] + ) + return { + "selected_action": best["action_id"], + "confident": confident, + "predicted_margin": margin, + "candidates": rows, + } + + +def models_to_json(models: Iterable[RidgeModel]) -> list[dict[str, Any]]: + return [model.to_json() for model in models] + + +def models_from_json(payload: Iterable[Mapping[str, Any]]) -> list[RidgeModel]: + return [RidgeModel.from_json(item) for item in payload] diff --git a/runs/active-intervention-v0/prepare_prospective.py b/runs/active-intervention-v0/prepare_prospective.py new file mode 100644 index 0000000..7750acb --- /dev/null +++ b/runs/active-intervention-v0/prepare_prospective.py @@ -0,0 +1,363 @@ +#!/usr/bin/env python3 +"""Freeze the unseen-trace 2x2 active intervention development surface.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import sys +from pathlib import Path +from typing import Any + + +AITUNER_ROOT = Path(os.environ.get("AITUNER_ROOT", Path(__file__).resolve().parents[2])) +sys.path.insert(0, str(AITUNER_ROOT / "src")) + +from aituner.spec import load_study_spec # noqa: E402 +from aituner.trace import load_trace_requests, select_requests_for_threshold # noqa: E402 + + +SCHEMA = "active-intervention-prospective-manifest-v0" +TP = 4 +REPETITIONS = (1, 2, 3) +DURATION_S = 300.0 +REPLAY_TIME_SCALE = 0.5 +OFFERED_RATE_PER_GPU = 2.75 +TARGET_COUNT = round(OFFERED_RATE_PER_GPU * DURATION_S * TP) +WINDOW_ID = "chat_w20260313_1000" +ENGINE_VERSION = "0.24.1.dev3+g668cfb7e2" + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def order_hash(values: list[str]) -> str: + return hashlib.sha256("\n".join(values).encode()).hexdigest() + + +def configs() -> list[dict[str, Any]]: + return [ + { + "id": "source_mns32_mbbt4096", + "mns": 32, + "mbbt": 4096, + "repetition_order": [1, 2, 3], + }, + { + "id": "mns64_mbbt4096", + "mns": 64, + "mbbt": 4096, + "repetition_order": [2, 3, 1], + }, + { + "id": "mns32_mbbt8192", + "mns": 32, + "mbbt": 8192, + "repetition_order": [3, 1, 2], + }, + { + "id": "joint_mns64_mbbt8192", + "mns": 64, + "mbbt": 8192, + "repetition_order": [1, 3, 2], + }, + ] + + +def partition_trace(source: Path, output_root: Path) -> dict[str, Any]: + source_sha = sha256_file(source) + output_root.mkdir(parents=True, exist_ok=True) + paths = {rep: output_root / f"rep{rep}.jsonl" for rep in REPETITIONS} + temporary = {rep: path.with_suffix(".jsonl.tmp") for rep, path in paths.items()} + handles = {rep: temporary[rep].open("w", encoding="utf-8") for rep in REPETITIONS} + counts = {rep: 0 for rep in REPETITIONS} + id_digests = {rep: hashlib.sha256() for rep in REPETITIONS} + total = 0 + try: + with source.open(encoding="utf-8") as input_file: + for line_number, line in enumerate(input_file, start=1): + if not line.strip(): + continue + row = json.loads(line) + original_id = str(row.get("request_id") or row.get("id") or line_number) + digest = hashlib.sha256( + f"{source_sha}:{line_number}:{original_id}".encode() + ).hexdigest() + repetition = int(digest[:16], 16) % len(REPETITIONS) + 1 + row["request_id"] = f"active-r{repetition}-{digest}" + handles[repetition].write(json.dumps(row, ensure_ascii=False) + "\n") + counts[repetition] += 1 + total += 1 + id_digests[repetition].update(row["request_id"].encode() + b"\n") + finally: + for handle in handles.values(): + handle.close() + for repetition in REPETITIONS: + os.replace(temporary[repetition], paths[repetition]) + partitions = { + str(rep): { + "path": str(paths[rep]), + "rows": counts[rep], + "bytes": paths[rep].stat().st_size, + "sha256": sha256_file(paths[rep]), + "request_id_order_sha256": id_digests[rep].hexdigest(), + } + for rep in REPETITIONS + } + return { + "source": str(source), + "source_sha256": source_sha, + "source_rows": total, + "partition_rule": "sha256(source_sha:line_number:original_id) modulo 3", + "partitions": partitions, + } + + +def materialize_study( + base_study: Path, + target: Path, + *, + repetition: int, + trace_path: Path, + windows_path: Path, +) -> None: + payload = json.loads(base_study.read_text(encoding="utf-8")) + payload["study_id"] = f"active-intervention-trace13-rep{repetition}" + payload["hardware"]["host_candidates"] = ["dash0"] + payload["engine"]["engine_version"] = ENGINE_VERSION + trace = payload["trace"] + trace.update( + { + "windows_path": str(windows_path), + "window_id": WINDOW_ID, + "trace_file_override": str(trace_path), + "completion_tokens_override": 128, + "replay_time_scale": REPLAY_TIME_SCALE, + "early_stop_max_lag_s": None, + "early_stop_max_elapsed_s": 360.0, + "restart_engine_after_early_stop": False, + "adaptive_stop": {"enabled": False}, + } + ) + atomic_json(target, payload) + + +def attainable_anchor(requests: list[Any], target_count: int) -> tuple[float, list[Any]]: + ordered = sorted(float(request.sampling_u) for request in requests) + if target_count <= 0 or target_count > len(ordered): + raise ValueError( + f"target count {target_count} is outside available range 1..{len(ordered)}" + ) + candidates = [] + for index in sorted({target_count - 1, min(target_count, len(ordered) - 1)}): + anchor = ordered[index] + selected = select_requests_for_threshold(requests, threshold=anchor) + candidates.append((abs(len(selected) - target_count), len(selected), anchor, selected)) + _error, _count, anchor, selected = min( + candidates, key=lambda item: (item[0], item[1], item[2]) + ) + return anchor, selected + + +def selection_record(selected: list[Any]) -> dict[str, Any]: + return { + "anchor": max(float(request.sampling_u) for request in selected), + "selected_count": len(selected), + "target_count": TARGET_COUNT, + "offered_req_s": len(selected) / DURATION_S, + "offered_req_s_per_gpu": len(selected) / DURATION_S / TP, + "request_id_order_sha256": order_hash([request.row_id for request in selected]), + "arrival_order_sha256": order_hash( + [f"{request.arrival_s:.12f}" for request in selected] + ), + "input_length_order_sha256": order_hash( + [str(request.prompt_tokens_hint) for request in selected] + ), + } + + +def build( + *, + base_study: Path, + base_action_manifest: Path, + source_trace: Path, + windows_path: Path, + private_root: Path, + policy_path: Path, +) -> dict[str, Any]: + base_manifest = json.loads(base_action_manifest.read_text(encoding="utf-8")) + if base_manifest.get("status") != "PASS": + raise ValueError("base action-aware manifest did not pass") + policy = json.loads(policy_path.read_text(encoding="utf-8")) + if policy.get("schema") != "active-intervention-policy-v0": + raise ValueError("unexpected frozen policy schema") + if policy.get("sanity", {}).get("red_flags"): + raise ValueError("frozen policy contains red flags") + + partition = partition_trace(source_trace, private_root / "traces") + repetitions = {} + selected_sets: list[set[str]] = [] + for repetition in REPETITIONS: + trace_path = Path(partition["partitions"][str(repetition)]["path"]) + study_path = private_root / "studies" / f"rep{repetition}-tp4.json" + materialize_study( + base_study, + study_path, + repetition=repetition, + trace_path=trace_path, + windows_path=windows_path, + ) + study = load_study_spec(study_path) + window, requests = load_trace_requests(study, study_spec_path=study_path) + duration_s = float(window.window_end - window.window_start) + if not math.isclose(duration_s, DURATION_S, abs_tol=1e-9): + raise ValueError(f"rep{repetition}: duration {duration_s} != {DURATION_S}") + _anchor, selected = attainable_anchor(requests, TARGET_COUNT) + record = selection_record(selected) + selected_sets.append({request.row_id for request in selected}) + repetitions[str(repetition)] = { + "study": str(study_path), + "study_sha256": sha256_file(study_path), + "trace": partition["partitions"][str(repetition)], + "available_filtered_requests": len(requests), + "selection": record, + } + + frozen_configs = configs() + config_ids = {str(config["id"]) for config in frozen_configs} + invariants = { + "three_nonempty_trace_partitions": all( + int(item["rows"]) > 0 for item in partition["partitions"].values() + ), + "partition_rows_conserved": sum( + int(item["rows"]) for item in partition["partitions"].values() + ) + == int(partition["source_rows"]), + "selected_sets_disjoint": all( + not selected_sets[left] & selected_sets[right] + for left in range(len(selected_sets)) + for right in range(left + 1, len(selected_sets)) + ), + "target_count_attained": all( + abs(int(item["selection"]["selected_count"]) - TARGET_COUNT) <= 1 + for item in repetitions.values() + ), + "four_unique_configs": len(config_ids) == 4, + "two_by_two_surface": { + (int(config["mns"]), int(config["mbbt"])) + for config in frozen_configs + } + == {(32, 4096), (64, 4096), (32, 8192), (64, 8192)}, + "repetition_orders_are_permutations": all( + sorted(config["repetition_order"]) == list(REPETITIONS) + for config in frozen_configs + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + return { + "schema": SCHEMA, + "status": "PASS" if not red_flags else "STOP", + "source": { + "window_id": WINDOW_ID, + "source_trace": str(source_trace), + "source_trace_sha256": partition["source_sha256"], + "windows_path": str(windows_path), + "base_study": str(base_study), + "base_study_sha256": sha256_file(base_study), + "base_action_manifest": str(base_action_manifest), + "base_action_manifest_sha256": sha256_file(base_action_manifest), + }, + "policy": { + "path": str(policy_path), + "sha256": sha256_file(policy_path), + "status": policy["status"], + "training": policy["training"], + "measurement_policy": policy["measurement_policy"], + "launch_reason": ( + "bounded unseen-trace joint-action test after a negative narrow " + "retrospective replay" + ), + }, + "engine": { + "tp": TP, + "duration_s": DURATION_S, + "client_timeout_s": 450.0, + "burnin_max_elapsed_s": 90.0, + "disable_slo_early_stop": True, + }, + "burnin": base_manifest["burnin"], + "private": {"trace_partition": partition}, + "repetitions": repetitions, + "configs": frozen_configs, + "source_config_id": "source_mns32_mbbt4096", + "actions": { + "noop": "source_mns32_mbbt4096", + "mns": "mns64_mbbt4096", + "mbbt": "mns32_mbbt8192", + "joint": "joint_mns64_mbbt8192", + }, + "checkpoints": { + "fractions": [0.25, 0.50, 0.75, 1.0], + "seconds": [75.0, 150.0, 225.0, 300.0], + }, + "gates": { + "acceptable_regret": 0.02, + "source_ceiling_normalized_goodput": 0.98, + "confirmation_trigger_gpu_cost_reduction": 0.10, + "contribution_gpu_cost_reduction": 0.20, + "maximum_task_regret": 0.05, + }, + "budget": { + "hard_cap_h20_hours": 6.0, + "session_estimate_h20_hours": 1.3, + "safety_h20_hours": 0.3, + "expected_h20_hours": [4.6, 5.5], + "expected_wall_minutes": [75, 100], + }, + "sanity": {"invariants": invariants, "red_flags": red_flags}, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--base-study", type=Path, required=True) + parser.add_argument("--base-action-manifest", type=Path, required=True) + parser.add_argument("--source-trace", type=Path, required=True) + parser.add_argument("--windows-path", type=Path, required=True) + parser.add_argument("--private-root", type=Path, required=True) + parser.add_argument("--policy", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = build( + base_study=args.base_study, + base_action_manifest=args.base_action_manifest, + source_trace=args.source_trace, + windows_path=args.windows_path, + private_root=args.private_root, + policy_path=args.policy, + ) + atomic_json(args.output, payload) + print(json.dumps({"status": payload["status"], "sanity": payload["sanity"]})) + if payload["status"] != "PASS": + raise SystemExit("prospective manifest preflight failed") + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/prospective_controller.py b/runs/active-intervention-v0/prospective_controller.py new file mode 100644 index 0000000..086ea93 --- /dev/null +++ b/runs/active-intervention-v0/prospective_controller.py @@ -0,0 +1,198 @@ +#!/usr/bin/env python3 +"""Run source first, select the next intervention, then annotate the 2x2 surface.""" + +from __future__ import annotations + +import argparse +import json +import sys +from pathlib import Path +from typing import Any, Mapping + + +HERE = Path(__file__).resolve().parent +ACTION_DIR = HERE.parent / "action-aware-v0" +sys.path.insert(0, str(ACTION_DIR)) +sys.path.insert(0, str(HERE)) + +import pilot_controller as action_controller # noqa: E402 +import prospective_decision # noqa: E402 + + +SCHEMA = "active-intervention-prospective-state-v0" + + +def validate_inputs(args: argparse.Namespace, manifest: Mapping[str, Any]) -> None: + if manifest.get("schema") != "active-intervention-prospective-manifest-v0": + raise RuntimeError("unexpected active intervention manifest schema") + if manifest.get("status") != "PASS" or manifest["sanity"]["red_flags"]: + raise RuntimeError("active intervention manifest did not pass preflight") + required = { + "manifest": args.manifest, + "policy": args.policy, + "aituner_root": args.aituner_root, + "vllm_source": args.vllm_source, + "venv_python": args.venv / "bin/python", + "venv_vllm": args.venv / "bin/vllm", + "model": args.model, + "client": args.client, + "burnin_study": Path(manifest["burnin"]["study"]), + } + for repetition, item in manifest["repetitions"].items(): + required[f"rep{repetition}_study"] = Path(item["study"]) + required[f"rep{repetition}_trace"] = Path(item["trace"]["path"]) + missing = {name: str(path) for name, path in required.items() if not path.exists()} + if missing: + raise RuntimeError(f"active intervention input paths missing: {missing}") + if prospective_decision.sha256_file(args.policy) != manifest["policy"]["sha256"]: + raise RuntimeError("active intervention policy hash mismatch") + + +def dry_run(args: argparse.Namespace, manifest: Mapping[str, Any]) -> dict[str, Any]: + plan = action_controller.dry_run_plan(args, manifest) + return { + "schema": "active-intervention-prospective-dry-run-v0", + "status": "PASS", + "manifest": str(args.manifest), + "policy": str(args.policy), + "source_first": manifest["source_config_id"], + "post_source_order": "selected by telemetry policy; all remaining cells then annotated", + "candidate_actions": manifest["actions"], + "projected_h20_hours": plan["projected_h20_hours"], + "hard_cap_h20_hours": plan["hard_cap_h20_hours"], + "sessions": plan["sessions"], + } + + +def load_or_build_decision( + *, args: argparse.Namespace, run_root: Path +) -> dict[str, Any]: + path = run_root / "active-decision.json" + if path.exists(): + decision = json.loads(path.read_text(encoding="utf-8")) + if decision.get("manifest_sha256") != prospective_decision.sha256_file( + args.manifest + ): + raise RuntimeError("existing active decision has a different manifest") + if decision.get("policy_sha256") != prospective_decision.sha256_file(args.policy): + raise RuntimeError("existing active decision has a different policy") + return decision + decision = prospective_decision.build_decision( + manifest_path=args.manifest, + policy_path=args.policy, + run_root=run_root, + ) + prospective_decision.atomic_json(path, decision) + return decision + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--manifest", type=Path, required=True) + result.add_argument("--policy", type=Path, required=True) + result.add_argument("--run-root", type=Path, required=True) + result.add_argument("--aituner-root", type=Path, required=True) + result.add_argument("--vllm-source", type=Path, required=True) + result.add_argument("--venv", type=Path, required=True) + result.add_argument("--model", type=Path, required=True) + result.add_argument("--client", type=Path, required=True) + result.add_argument("--dry-run", action="store_true") + return result + + +def main() -> None: + args = parser().parse_args() + manifest = json.loads(args.manifest.read_text(encoding="utf-8")) + validate_inputs(args, manifest) + action_controller.configure(args, manifest) + action_controller.base.MARKER = "active-intervention-prospective-v0" + if args.dry_run: + print(json.dumps(dry_run(args, manifest), indent=2, sort_keys=True)) + return + + args.run_root.mkdir(parents=True, exist_ok=True) + copied_manifest = args.run_root / "prospective-manifest.json" + if not copied_manifest.exists(): + action_controller.atomic_json(copied_manifest, manifest) + state_path = args.run_root / "controller-state.json" + state = action_controller.load_state( + state_path, float(manifest["budget"]["hard_cap_h20_hours"]) + ) + state["schema"] = SCHEMA + state["status"] = "running" + action_controller.atomic_json(state_path, state) + + configs = {str(item["id"]): dict(item) for item in manifest["configs"]} + config_indexes = { + str(item["id"]): index for index, item in enumerate(manifest["configs"]) + } + source_id = str(manifest["source_config_id"]) + action_controller.execute_session( + args=args, + manifest=manifest, + config=configs[source_id], + index=config_indexes[source_id], + state=state, + state_path=state_path, + ) + decision = load_or_build_decision(args=args, run_root=args.run_root) + state["active_decision"] = { + "path": str(args.run_root / "active-decision.json"), + "status": decision["status"], + "outcome_only": { + key: decision["decisions"]["outcome_only"][key] + for key in ("selected_cutoff_s", "decision_kind", "selected_action") + }, + "telemetry": { + key: decision["decisions"]["telemetry"][key] + for key in ("selected_cutoff_s", "decision_kind", "selected_action") + }, + } + action_controller.atomic_json(state_path, state) + if decision["status"] != "SELECTED": + state["status"] = decision["status"].lower() + state["completed_at"] = action_controller.time.time() + action_controller.atomic_json(state_path, state) + action_controller.wait_all_idle() + print(json.dumps({"status": state["status"], "decision": decision["status"]})) + return + + action_order = decision["decisions"]["telemetry"]["intervention_order"] + execution_order = [source_id] + for action_id in action_order: + target_id = str(manifest["actions"][action_id]) + if target_id not in execution_order: + execution_order.append(target_id) + for config_id in configs: + if config_id not in execution_order: + execution_order.append(config_id) + state["execution_order"] = execution_order + action_controller.atomic_json(state_path, state) + for config_id in execution_order[1:]: + action_controller.execute_session( + args=args, + manifest=manifest, + config=configs[config_id], + index=config_indexes[config_id], + state=state, + state_path=state_path, + ) + state["status"] = "complete" + state["completed_at"] = action_controller.time.time() + action_controller.atomic_json(state_path, state) + action_controller.wait_all_idle() + print( + json.dumps( + { + "status": state["status"], + "completed_sessions": state["completed_sessions"], + "gpu_hours_total": state["gpu_hours_total"], + "execution_order": execution_order, + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/prospective_decision.py b/runs/active-intervention-v0/prospective_decision.py new file mode 100644 index 0000000..7b489d7 --- /dev/null +++ b/runs/active-intervention-v0/prospective_decision.py @@ -0,0 +1,441 @@ +#!/usr/bin/env python3 +"""Choose measurement horizon and next intervention from a completed source run.""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +import os +import statistics +import sys +from pathlib import Path +from typing import Any, Mapping, Sequence + +import numpy as np + + +HERE = Path(__file__).resolve().parent +COMMON_STATE = HERE.parent / "telemetry-residual" +sys.path.insert(0, str(COMMON_STATE)) + +from common_state import summarize_engine # noqa: E402 + + +SCHEMA = "active-intervention-prospective-decision-v0" + + +def load_module(name: str, path: Path): + spec = importlib.util.spec_from_file_location(name, path) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +MODEL = load_module("active_intervention_prospective_model", HERE / "model.py") +EXTRACT = load_module( + "active_intervention_prospective_extract", HERE / "extract_training.py" +) + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def numeric(values: Sequence[float]) -> dict[str, Any]: + finite = [float(value) for value in values] + if not finite or any(not math.isfinite(value) for value in finite): + raise ValueError("numeric summary requires finite values") + return { + "n": len(finite), + "min": min(finite), + "max": max(finite), + "distinct_n": len(set(finite)), + } + + +def load_engine_records(source_root: Path) -> tuple[list[dict[str, Any]], Path]: + streams = sorted((source_root / "opprof").glob("*.jsonl")) + if len(streams) != 1: + raise ValueError(f"expected one source engine stream, found {len(streams)}") + records = [ + row for row in EXTRACT.load_jsonl(streams[0]) if "step_index" in row + ] + if not records: + raise ValueError("source engine stream has no Layer-1 records") + return records, streams[0] + + +def candidate_example( + *, + source_config: Mapping[str, Any], + target_config: Mapping[str, Any], + action_id: str, + offered_rate_per_gpu: float, + outcome: Mapping[str, float], + telemetry: Mapping[str, float], +) -> dict[str, Any]: + return { + "source": { + "mns": int(source_config["mns"]), + "mbbt": int(source_config["mbbt"]), + "offered_rate_per_gpu": float(offered_rate_per_gpu), + "outcome": dict(outcome), + "telemetry": dict(telemetry), + }, + "action": { + "id": action_id, + "target_mns": int(target_config["mns"]), + "target_mbbt": int(target_config["mbbt"]), + }, + } + + +def aggregate_checkpoint( + *, + models: Sequence[Any], + examples_by_action: Mapping[str, Sequence[Mapping[str, Any]]], + include_telemetry: bool, + confidence_z: float, + minimum_margin: float, +) -> dict[str, Any]: + rows = [] + for action_id, examples in sorted(examples_by_action.items()): + raw = [] + for example in examples: + source = example["source"] + action = example["action"] + noop = ( + int(source["mns"]) == int(action["target_mns"]) + and int(source["mbbt"]) == int(action["target_mbbt"]) + ) + if noop: + raw.extend(0.0 for _model in models) + continue + names, values = MODEL.feature_vector( + example, include_telemetry=include_telemetry + ) + if any(model.feature_names != tuple(names) for model in models): + raise ValueError("prospective feature schema does not match frozen model") + raw.extend(model.predict(values) for model in models) + clipped = np.clip(np.asarray(raw, dtype=np.float64), -1.0, 1.0) + prediction = { + "mean": float(clipped.mean()), + "std": float(clipped.std(ddof=0)), + "min": float(clipped.min()), + "max": float(clipped.max()), + "distinct_n": len(set(float(value) for value in clipped)), + "sample_n": int(clipped.size), + } + rows.append( + { + "action_id": action_id, + "prediction": prediction, + "lower": prediction["mean"] - confidence_z * prediction["std"], + "upper": prediction["mean"] + confidence_z * prediction["std"], + } + ) + rows.sort(key=lambda row: (-row["prediction"]["mean"], row["action_id"])) + best, second = rows[:2] + margin = float(best["prediction"]["mean"] - second["prediction"]["mean"]) + confident = bool( + margin >= minimum_margin and best["lower"] > second["upper"] + ) + return { + "selected_action": best["action_id"], + "confident": confident, + "predicted_margin": margin, + "candidates": rows, + } + + +def apply_measurement_and_acquisition(checkpoints: list[dict[str, Any]]) -> dict[str, Any]: + selected = checkpoints[-1] + stop_reason = "full_measurement_fallback" + for previous, current in zip(checkpoints, checkpoints[1:], strict=False): + if ( + previous["confident"] + and current["confident"] + and previous["selected_action"] == current["selected_action"] + ): + selected = current + stop_reason = "two_consecutive_confident_checkpoints" + break + candidates = selected["candidates"] + mean_best = candidates[0] + non_noop = [row for row in candidates if row["action_id"] != "noop"] + if selected["confident"]: + chosen = mean_best + decision_kind = "exploit" + else: + positive_ucb = [row for row in non_noop if float(row["upper"]) > 0.0] + if positive_ucb: + chosen = max( + positive_ucb, + key=lambda row: (float(row["upper"]), row["action_id"]), + ) + decision_kind = "diagnostic_ucb" + else: + chosen = next(row for row in candidates if row["action_id"] == "noop") + decision_kind = "abstain_no_positive_ucb" + remaining = [row for row in candidates if row["action_id"] != chosen["action_id"]] + remaining.sort(key=lambda row: (-float(row["upper"]), row["action_id"])) + order = [chosen["action_id"], *(row["action_id"] for row in remaining)] + return { + "selected_phase": selected["phase"], + "selected_cutoff_s": selected["cutoff_s"], + "measurement_stop_reason": stop_reason, + "decision_kind": decision_kind, + "selected_action": chosen["action_id"], + "intervention_order": order, + "selected_checkpoint": selected, + "checkpoints": checkpoints, + } + + +def build_decision( + *, manifest_path: Path, policy_path: Path, run_root: Path +) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + policy = json.loads(policy_path.read_text(encoding="utf-8")) + if manifest.get("schema") != "active-intervention-prospective-manifest-v0": + raise ValueError("unexpected prospective manifest schema") + if policy.get("schema") != "active-intervention-policy-v0": + raise ValueError("unexpected frozen policy schema") + if sha256_file(policy_path) != manifest["policy"]["sha256"]: + raise ValueError("frozen policy hash changed after manifest preparation") + configs = {str(item["id"]): item for item in manifest["configs"]} + source_id = str(manifest["source_config_id"]) + source_config = configs[source_id] + source_root = run_root / "sessions" / source_id + engine_records, stream_path = load_engine_records(source_root) + phases = [f"{fraction:.2f}" for fraction in manifest["checkpoints"]["fractions"]] + confidence_z = float(policy["measurement_policy"]["confidence_z"]) + minimum_margin = float(policy["measurement_policy"]["minimum_margin"]) + + examples: dict[str, dict[str, dict[str, Mapping[str, Any]]]] = {} + source_measurements: dict[str, dict[str, Any]] = {} + source_normalized = [] + telemetry_values = [] + for repetition in sorted(int(key) for key in manifest["repetitions"]): + item = manifest["repetitions"][str(repetition)] + result_root = source_root / f"rep{repetition}" + result = json.loads((result_root / "result.json").read_text(encoding="utf-8")) + if result["selection"]["request_id_order_sha256"] != item["selection"][ + "request_id_order_sha256" + ]: + raise ValueError(f"source request hash mismatch: rep{repetition}") + requests = EXTRACT.load_jsonl(result_root / "requests.jsonl") + offered_rate = float(item["selection"]["offered_req_s_per_gpu"]) + offered_total = offered_rate * int(manifest["engine"]["tp"]) + source_normalized.append( + float(result["slo_pass_count"]) + / float(manifest["engine"]["duration_s"]) + / offered_total + ) + start_ns = int(result["interval"]["start_mono_ns"]) + examples[str(repetition)] = {} + source_measurements[str(repetition)] = { + "result": str(result_root / "result.json"), + "result_sha256": sha256_file(result_root / "result.json"), + "request_sha256": sha256_file(result_root / "requests.jsonl"), + "phases": {}, + } + for phase, cutoff_s in zip( + phases, manifest["checkpoints"]["seconds"], strict=True + ): + outcome = EXTRACT.prefix_outcome( + requests, cutoff_s=float(cutoff_s), offered_total=offered_total + ) + admitted_count = sum( + float(request["arrival_s"]) <= float(cutoff_s) + for request in requests + ) + state = summarize_engine( + engine_records, + start_ns=start_ns, + end_ns=start_ns + round(float(cutoff_s) * 1e9), + request_count=admitted_count, + ) + if not all(state["sanity"]["invariants"].values()): + raise ValueError( + f"source engine state invariant failed: rep{repetition} {phase}" + ) + telemetry = EXTRACT.telemetry_record(state) + telemetry_values.extend(float(value) for value in telemetry.values()) + source_measurements[str(repetition)]["phases"][phase] = { + "cutoff_s": float(cutoff_s), + "outcome": outcome, + "telemetry": telemetry, + "engine_sanity": state["sanity"], + } + examples[str(repetition)][phase] = { + action_id: candidate_example( + source_config=source_config, + target_config=configs[str(target_id)], + action_id=action_id, + offered_rate_per_gpu=offered_rate, + outcome=outcome, + telemetry=telemetry, + ) + for action_id, target_id in manifest["actions"].items() + } + + decisions = {} + for mode, include_telemetry in (("outcome_only", False), ("telemetry", True)): + checkpoints = [] + for phase, cutoff_s in zip( + phases, manifest["checkpoints"]["seconds"], strict=True + ): + models = MODEL.models_from_json(policy["phases"][phase][mode]["models"]) + examples_by_action = { + action_id: [ + examples[str(repetition)][phase][action_id] + for repetition in sorted(int(key) for key in manifest["repetitions"]) + ] + for action_id in manifest["actions"] + } + checkpoint = aggregate_checkpoint( + models=models, + examples_by_action=examples_by_action, + include_telemetry=include_telemetry, + confidence_z=confidence_z, + minimum_margin=minimum_margin, + ) + checkpoints.append( + {"phase": phase, "cutoff_s": float(cutoff_s), **checkpoint} + ) + decisions[mode] = apply_measurement_and_acquisition(checkpoints) + + ceiling = float(manifest["gates"]["source_ceiling_normalized_goodput"]) + source_median = float(statistics.median(source_normalized)) + status = "STOP_SOURCE_CEILING" if source_median >= ceiling else "SELECTED" + phase_admission_monotonic = all( + all( + left <= right + 1e-12 + for left, right in zip(values, values[1:], strict=False) + ) + for repetition in source_measurements.values() + for values in ( + [ + float(repetition["phases"][phase]["outcome"]["admitted_fraction"]) + for phase in phases + ], + ) + ) + telemetry_ratio_keys = { + "prefill_token_fraction", + "kv_usage_mean", + "kv_usage_max", + "graph_none_share", + "graph_full_share", + "graph_padding_fraction", + } + telemetry_records = [ + measurement["telemetry"] + for repetition in source_measurements.values() + for measurement in repetition["phases"].values() + ] + invariants = { + "three_source_repetitions": len(source_normalized) == 3, + "source_goodput_nonnegative": all(value >= 0.0 for value in source_normalized), + "source_goodput_bounded": all( + value <= 1.0 + 1e-12 for value in source_normalized + ), + "four_actions": set(manifest["actions"]) == {"noop", "mns", "mbbt", "joint"}, + "four_checkpoints": len(phases) == 4, + "finite_telemetry": all(math.isfinite(value) for value in telemetry_values), + "nonnegative_telemetry": all( + float(value) >= 0.0 + for record in telemetry_records + for key, value in record.items() + if key != "kv_usage_end_minus_start" + ), + "telemetry_ratios_bounded": all( + 0.0 <= float(record[key]) <= 1.0 + 1e-12 + for record in telemetry_records + for key in telemetry_ratio_keys + ), + "telemetry_not_all_identical": len(set(telemetry_values)) > 1, + "phase_admission_monotonic": phase_admission_monotonic, + "orders_are_permutations": all( + set(decisions[mode]["intervention_order"]) == set(manifest["actions"]) + for mode in decisions + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + if red_flags: + status = "STOP_SANITY" + return { + "schema": SCHEMA, + "status": status, + "manifest": str(manifest_path), + "manifest_sha256": sha256_file(manifest_path), + "policy": str(policy_path), + "policy_sha256": sha256_file(policy_path), + "source_stream": str(stream_path), + "source_stream_sha256": sha256_file(stream_path), + "source_measurements": source_measurements, + "source_normalized_goodput": { + "values": source_normalized, + "median": source_median, + **numeric(source_normalized), + }, + "decisions": decisions, + "sanity": { + "invariants": invariants, + "red_flags": red_flags, + "telemetry_values": numeric(telemetry_values), + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--policy", type=Path, required=True) + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + decision = build_decision( + manifest_path=args.manifest, policy_path=args.policy, run_root=args.run_root + ) + atomic_json(args.output, decision) + print( + json.dumps( + { + "status": decision["status"], + "source_normalized_goodput": decision["source_normalized_goodput"], + "outcome_only": { + key: decision["decisions"]["outcome_only"][key] + for key in ("selected_cutoff_s", "decision_kind", "selected_action") + }, + "telemetry": { + key: decision["decisions"]["telemetry"][key] + for key in ("selected_cutoff_s", "decision_kind", "selected_action") + }, + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/test_model.py b/runs/active-intervention-v0/test_model.py new file mode 100644 index 0000000..c49401a --- /dev/null +++ b/runs/active-intervention-v0/test_model.py @@ -0,0 +1,91 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import sys +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load_model(): + spec = importlib.util.spec_from_file_location( + "active_intervention_model", HERE / "model.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def example(model, decision: str, action: str, pressure: float, target: float): + outcome = { + name: 0.5 for name in model.PREFIX_FEATURES + } + telemetry = {name: 0.0 for name in model.TELEMETRY_FEATURES} + telemetry["queue_waiting_mean"] = pressure + telemetry["batch_size_mean"] = pressure + return { + "decision_id": decision, + "source": { + "mns": 16, + "mbbt": 8192, + "offered_rate_per_gpu": 2.0, + "outcome": outcome, + "telemetry": telemetry, + }, + "action": { + "id": action, + "target_mns": 64 if action == "mns" else 16, + "target_mbbt": 8192 if action == "mns" else 16384, + }, + "target_normalized_goodput": target, + "target_delta_normalized_goodput": target - 0.5, + } + + +def main() -> None: + model = load_model() + examples = [] + for index, pressure in enumerate((0.2, 0.5, 0.8), 1): + examples.extend( + ( + example(model, f"d{index}", "mns", pressure, 0.5 + pressure / 2), + example(model, f"d{index}", "mbbt", pressure, 0.6 - pressure / 4), + ) + ) + fitted = model.fit_ridge( + examples, include_telemetry=True, regularization=1.0 + ) + encoded = fitted.to_json() + restored = model.RidgeModel.from_json(encoded) + names, values = model.feature_vector(examples[-2], include_telemetry=True) + assert tuple(names) == restored.feature_names + assert abs(fitted.predict(values) - restored.predict(values)) < 1e-12 + ensemble = model.fit_jackknife_ensemble( + examples, include_telemetry=True, regularization=1.0 + ) + decision = model.select_action( + ensemble, examples[-2:], include_telemetry=True, minimum_margin=0.0 + ) + assert decision["selected_action"] == "mns" + assert all(-1.0 <= row["prediction"]["mean"] <= 1.0 for row in decision["candidates"]) + noop = example(model, "noop", "noop", 0.8, 0.5) + noop["action"]["target_mbbt"] = 8192 + prediction = model.ensemble_predict( + ensemble, noop, include_telemetry=True + ) + assert prediction == { + "mean": 0.0, + "std": 0.0, + "min": 0.0, + "max": 0.0, + "distinct_n": 1, + } + print("active intervention model: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/test_pipeline.py b/runs/active-intervention-v0/test_pipeline.py new file mode 100644 index 0000000..a6597eb --- /dev/null +++ b/runs/active-intervention-v0/test_pipeline.py @@ -0,0 +1,199 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import json +import sys +import tempfile +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load(name: str, path: Path): + spec = importlib.util.spec_from_file_location(name, path) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def write_json(path: Path, payload) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload) + "\n", encoding="utf-8") + + +def write_jsonl(path: Path, rows) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text( + "".join(json.dumps(row) + "\n" for row in rows), encoding="utf-8" + ) + + +def engine_record(index: int, timestamp_ns: int) -> dict: + alternate = index % 2 + return { + "step_index": index, + "submit_mono_ns": timestamp_ns, + "model_executed": True, + "scheduled_requests": 1 + alternate, + "decode_batch_size": alternate, + "prefill_tokens": 8 + alternate, + "decode_tokens": alternate, + "preemptions": 0, + "queues": {"waiting": alternate, "running": 1 + alternate}, + "kv": {"usage": 0.1 + 0.01 * alternate}, + "cudagraph": { + "runtime_mode": "FULL" if alternate else "NONE", + "bucket_tokens": 16, + "padding_tokens": alternate, + }, + "dropped_records_before": 0, + } + + +def main() -> None: + extractor = load("active_intervention_extract_test", HERE / "extract_training.py") + trainer = load("active_intervention_train_test", HERE / "train_policy.py") + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + run_root = root / "runs" + configs = [ + {"id": "a_base", "mns": 16, "mbbt": 8192}, + {"id": "a_mns", "mns": 64, "mbbt": 8192}, + {"id": "a_mbbt", "mns": 16, "mbbt": 16384}, + {"id": "b_base", "mns": 64, "mbbt": 2048}, + {"id": "b_mns", "mns": 128, "mbbt": 2048}, + {"id": "b_mbbt", "mns": 64, "mbbt": 8192}, + ] + manifest = { + "engine": {"duration_s": 300.0, "tp": 4}, + "configs": configs, + "repetitions": { + str(rep): {"selection": {"offered_req_s_per_gpu": 0.01}} + for rep in (1, 2, 3) + }, + "regimes": { + "A": { + "source": "a_base", + "actions": {"mns": "a_mns", "mbbt": "a_mbbt"}, + }, + "B": { + "source": "b_base", + "actions": {"mns": "b_mns", "mbbt": "b_mbbt"}, + }, + }, + } + manifest_path = root / "manifest.json" + write_json(manifest_path, manifest) + + streams = [] + source_starts: dict[tuple[str, int], int] = {} + for source_index, source_id in enumerate(("a_base", "b_base")): + rows = [] + index = 0 + for repetition in (1, 2, 3): + start_ns = int((source_index * 2000 + repetition * 400) * 1e9) + source_starts[(source_id, repetition)] = start_ns + for second in (1, 30, 76, 105, 151, 180, 226, 255): + rows.append(engine_record(index, start_ns + int(second * 1e9))) + index += 1 + stream_path = root / f"{source_id}-stream.jsonl" + write_jsonl(stream_path, rows) + streams.append( + { + "config_id": source_id, + "stream": str(stream_path), + "stream_sha256": extractor.sha256_file(stream_path), + } + ) + + request_rows = [ + { + "request_id": f"r{index}", + "arrival_s": arrival, + "completed_elapsed_s": arrival + 10, + "slo_pass": index != 3, + "ttft_ms": 1000 + index * 100, + "tpot_ms": 20 + index, + "raw_input_tokens": 1000 + index * 100, + } + for index, arrival in enumerate((5.0, 80.0, 155.0, 230.0), 1) + ] + for source_id in ("a_base", "b_base"): + for repetition in (1, 2, 3): + write_jsonl( + run_root + / "sessions" + / source_id + / f"rep{repetition}" + / "requests.jsonl", + request_rows, + ) + + goodput = { + "a_base": 0.020, + "a_mns": 0.036, + "a_mbbt": 0.028, + "b_base": 0.032, + "b_mns": 0.030, + "b_mbbt": 0.038, + } + runs = [] + for config in configs: + for repetition in (1, 2, 3): + item = { + "config_id": config["id"], + "repetition": repetition, + "outcome": { + "slo_goodput_req_s": goodput[config["id"]] + + repetition * 0.0001 + }, + } + if config["id"] in ("a_base", "b_base"): + start_ns = source_starts[(config["id"], repetition)] + item["state"] = { + "interval": { + "start_ns": start_ns, + "end_ns": start_ns + int(300 * 1e9), + } + } + runs.append(item) + audit = { + "schema": "action-aware-constraint-pilot-audit-v0", + "sanity": {"red_flags": []}, + "streams": streams, + "runs": runs, + } + audit_path = root / "audit.json" + write_json(audit_path, audit) + + dataset = extractor.build_dataset( + audit_path=audit_path, manifest_path=manifest_path, run_root=run_root + ) + assert dataset["status"] == "VALID" + assert len(dataset["examples"]) == 72 + assert not dataset["sanity"]["red_flags"] + assert all( + "exclusive" not in feature + for example in dataset["examples"] + for feature in example["source"]["telemetry"] + ) + dataset_path = root / "dataset.json" + write_json(dataset_path, dataset) + policy = trainer.build_policy(dataset_path) + assert policy["status"] in { + "RETROSPECTIVE_GPU_COST_SIGNAL", + "NO_RETROSPECTIVE_GPU_COST_SIGNAL", + } + assert policy["training"]["acceptable_regret"] == 0.02 + assert policy["sequential_replay"]["outcome_only"]["decision_n"] == 6 + assert policy["sequential_replay"]["telemetry"]["decision_n"] == 6 + assert not policy["sanity"]["red_flags"] + print("active intervention pipeline: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/test_prospective.py b/runs/active-intervention-v0/test_prospective.py new file mode 100644 index 0000000..7c95331 --- /dev/null +++ b/runs/active-intervention-v0/test_prospective.py @@ -0,0 +1,190 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import json +import sys +import tempfile +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load(name: str, path: Path): + spec = importlib.util.spec_from_file_location(name, path) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def write_json(path: Path, payload) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload) + "\n", encoding="utf-8") + + +def main() -> None: + prepare = load("active_intervention_prepare_test", HERE / "prepare_prospective.py") + decision_module = load( + "active_intervention_decision_test", HERE / "prospective_decision.py" + ) + analyzer = load("active_intervention_audit_test", HERE / "analyze_prospective.py") + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + source = root / "source.jsonl" + source.write_text( + "".join( + json.dumps( + { + "request_id": f"request-{index}", + "timestamp": float(index), + "sampling_u": index / 100.0, + } + ) + + "\n" + for index in range(60) + ), + encoding="utf-8", + ) + partition = prepare.partition_trace(source, root / "partitions") + assert sum(item["rows"] for item in partition["partitions"].values()) == 60 + ids = [] + for item in partition["partitions"].values(): + assert item["rows"] > 0 + ids.extend( + json.loads(line)["request_id"] + for line in Path(item["path"]).read_text(encoding="utf-8").splitlines() + ) + assert len(ids) == len(set(ids)) == 60 + + checkpoints = [ + { + "phase": "0.25", + "cutoff_s": 75.0, + "selected_action": "joint", + "confident": True, + "candidates": [ + {"action_id": "joint", "upper": 0.5, "prediction": {"mean": 0.4}}, + {"action_id": "mns", "upper": 0.2, "prediction": {"mean": 0.1}}, + {"action_id": "mbbt", "upper": 0.1, "prediction": {"mean": 0.05}}, + {"action_id": "noop", "upper": 0.0, "prediction": {"mean": 0.0}}, + ], + }, + { + "phase": "0.50", + "cutoff_s": 150.0, + "selected_action": "joint", + "confident": True, + "candidates": [ + {"action_id": "joint", "upper": 0.45, "prediction": {"mean": 0.4}}, + {"action_id": "mns", "upper": 0.2, "prediction": {"mean": 0.1}}, + {"action_id": "mbbt", "upper": 0.1, "prediction": {"mean": 0.05}}, + {"action_id": "noop", "upper": 0.0, "prediction": {"mean": 0.0}}, + ], + }, + ] + selected = decision_module.apply_measurement_and_acquisition(checkpoints) + assert selected["selected_cutoff_s"] == 150.0 + assert selected["selected_action"] == "joint" + + configs = prepare.configs() + repetitions = { + str(rep): { + "selection": { + "offered_req_s_per_gpu": 0.25, + "request_id_order_sha256": f"hash-{rep}", + } + } + for rep in (1, 2, 3) + } + manifest = { + "schema": "active-intervention-prospective-manifest-v0", + "engine": {"duration_s": 300.0, "tp": 4}, + "repetitions": repetitions, + "configs": configs, + "source_config_id": "source_mns32_mbbt4096", + "actions": { + "noop": "source_mns32_mbbt4096", + "mns": "mns64_mbbt4096", + "mbbt": "mns32_mbbt8192", + "joint": "joint_mns64_mbbt8192", + }, + "gates": { + "acceptable_regret": 0.02, + "confirmation_trigger_gpu_cost_reduction": 0.10, + "contribution_gpu_cost_reduction": 0.20, + }, + } + manifest_path = root / "manifest.json" + write_json(manifest_path, manifest) + run_root = root / "run" + scores = { + "source_mns32_mbbt4096": 0.5, + "mns64_mbbt4096": 0.8, + "mns32_mbbt8192": 0.7, + "joint_mns64_mbbt8192": 1.0, + } + sessions = {} + for config in configs: + config_id = config["id"] + sessions[config_id] = {"status": "complete", "gpu_hours": 1.2} + for repetition in (1, 2, 3): + result = { + "selection": { + "request_id_order_sha256": f"hash-{repetition}" + }, + "slo_pass_count": round(scores[config_id] * 300), + "pass_rate": scores[config_id], + "interval": {"elapsed_s": 300.0}, + } + write_json( + run_root + / "sessions" + / config_id + / f"rep{repetition}" + / "result.json", + result, + ) + state = { + "status": "complete", + "gpu_hours_total": 4.8, + "sessions": sessions, + } + write_json(run_root / "controller-state.json", state) + mode_base = { + "selected_cutoff_s": 300.0, + "selected_action": "mns", + "decision_kind": "exploit", + "intervention_order": ["mns", "mbbt", "joint", "noop"], + } + mode_telemetry = { + "selected_cutoff_s": 150.0, + "selected_action": "joint", + "decision_kind": "exploit", + "intervention_order": ["joint", "mns", "mbbt", "noop"], + } + decision = { + "schema": "active-intervention-prospective-decision-v0", + "manifest_sha256": analyzer.sha256_file(manifest_path), + "decisions": { + "outcome_only": mode_base, + "telemetry": mode_telemetry, + }, + } + decision_path = root / "decision.json" + write_json(decision_path, decision) + audit = analyzer.build_audit( + manifest_path=manifest_path, + decision_path=decision_path, + run_root=run_root, + ) + assert audit["status"] == "TRIGGER_ACTUAL_EARLY_STOP_CONFIRMATION" + assert audit["comparison"]["telemetry_gpu_cost_reduction_fraction"] > 0.10 + assert not audit["sanity"]["red_flags"] + print("active intervention prospective pipeline: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/active-intervention-v0/trace13-results.json b/runs/active-intervention-v0/trace13-results.json new file mode 100644 index 0000000..b5c3e01 --- /dev/null +++ b/runs/active-intervention-v0/trace13-results.json @@ -0,0 +1,151 @@ +{ + "schema": "active-intervention-trace13-result-summary-v0", + "status": "STOP_NO_PROSPECTIVE_GPU_COST_SIGNAL", + "provenance": { + "aituner_commit": "39b767e384fc49da53b99ad06a3e2ca1b6ac37d6", + "vllm_commit": "4b253fd8619764b6971a7f2e3a3aa7545f6ace05", + "manifest_sha256": "3bd25ae0ca040729a6351635f14447b3c789d4d86fb0fe4d65940735ad225a78", + "policy_sha256": "4f096d3a5f8c38771e956dfd576dd6cd5d5691286ab258028dbddbe07b20078c", + "decision_sha256": "86b7089151c480e18b5ae6ed65c4e4a3e11159dd0c16d5851312d4d9196d5ca2", + "controller_state_sha256": "d10d3cbe5ce20eacc1392f52e79d16ae23b5e15301123d5e85e48d53ae676cda", + "audit_sha256": "bd95b45e5d4cb93b5ad2f722b7dc96553d64eb8a5d5aae3a82f29c2d015fe3f6", + "remote_root": "/home/admin/cpfs/wjh/active-intervention-prospective-20260715" + }, + "cost": { + "annotation_campaign_h20_hours": 5.0379046784506905, + "hard_cap_h20_hours": 6.0, + "outcome_only_cost_to_acceptable_h20_hours_lower_bound": 2.4284364508172893, + "telemetry_cost_to_acceptable_h20_hours_lower_bound": 2.4284364508172893, + "telemetry_gpu_cost_reduction_fraction": 0.0, + "session_h20_hours": { + "source_mns32_mbbt4096": 1.3566088432735868, + "mns64_mbbt4096": 1.256969277858734, + "mns32_mbbt8192": 1.254111782974667, + "joint_mns64_mbbt8192": 1.1702147743437026 + } + }, + "policy_comparison": { + "outcome_only": { + "measurement_cutoff_s": 300.0, + "selected_action": "joint", + "intervention_order": ["joint", "mns", "mbbt", "noop"] + }, + "telemetry": { + "measurement_cutoff_s": 300.0, + "selected_action": "joint", + "intervention_order": ["joint", "mns", "mbbt", "noop"] + }, + "action_changed": false, + "measurement_changed": false, + "confirmation_trigger": false, + "contribution_gate": false + }, + "surface": { + "source_mns32_mbbt4096": { + "normalized_slo_goodput": [ + 0.40090909090909094, + 0.3978787878787879, + 0.42060606060606065 + ], + "median": 0.40090909090909094 + }, + "mns64_mbbt4096": { + "normalized_slo_goodput": [1.0, 0.9996969696969698, 1.0], + "median": 1.0 + }, + "mns32_mbbt8192": { + "normalized_slo_goodput": [ + 0.44393939393939397, + 0.41515151515151516, + 0.4260606060606061 + ], + "median": 0.42606060606060603 + }, + "joint_mns64_mbbt8192": { + "normalized_slo_goodput": [1.0, 1.0, 1.0], + "median": 1.0 + } + }, + "selected_checkpoint_prediction": { + "actual_median_effect": { + "noop": 0.0, + "mns": 0.5990909090909091, + "mbbt": 0.02515151515151509, + "joint": 0.5990909090909091 + }, + "outcome_only_predicted_effect": { + "noop": 0.0, + "mns": 0.2886250281729182, + "mbbt": 0.17933598309437812, + "joint": 0.3205015384324615 + }, + "telemetry_predicted_effect": { + "noop": 0.0, + "mns": 0.26117798146236215, + "mbbt": 0.09686132563074483, + "joint": 0.35190199346536294 + }, + "actual_joint_minus_mns": 0.0, + "outcome_only_joint_minus_mns": 0.0318765102595433, + "telemetry_joint_minus_mns": 0.09072401200300079 + }, + "engine_mechanism": { + "source_mns32_mbbt4096": { + "scheduler_records": 41086, + "waiting_fraction": 0.9312174463320839, + "mns_exclusive_fraction": 0.8536484447256973, + "mbbt_exclusive_fraction": 0.01114734946210388, + "both_fraction": 0.06642165214428272, + "running_utilization_mean": 0.9738878510928297, + "token_utilization_mean": 0.15694342925509905, + "kv_usage_mean": 0.027507593814715858, + "preemptions": 0 + }, + "mns64_mbbt4096": { + "scheduler_records": 37001, + "waiting_fraction": 0.053809356503878275, + "mns_exclusive_fraction": 0.0, + "mbbt_exclusive_fraction": 0.053809356503878275, + "running_utilization_mean": 0.5410364753655307, + "token_utilization_mean": 0.17425695631536986, + "kv_usage_mean": 0.030549112146415616, + "preemptions": 0 + }, + "mns32_mbbt8192": { + "scheduler_records": 41348, + "waiting_fraction": 0.9119425365192996, + "mns_exclusive_fraction": 0.9108542130211861, + "mbbt_exclusive_fraction": 0.0003627744993711909, + "running_utilization_mean": 0.9652567838831383, + "token_utilization_mean": 0.0779606414231039, + "kv_usage_mean": 0.027355900366122385, + "preemptions": 0 + }, + "joint_mns64_mbbt8192": { + "scheduler_records": 40416, + "waiting_fraction": 0.0088826207442597, + "mns_exclusive_fraction": 0.0, + "mbbt_exclusive_fraction": 0.0088826207442597, + "running_utilization_mean": 0.49403392220902614, + "token_utilization_mean": 0.07978070546782215, + "kv_usage_mean": 0.028003770774946098, + "preemptions": 0 + } + }, + "sanity": { + "surface_outcomes": {"n": 12, "min": 0.3978787878787879, "max": 1.0, "distinct_n": 8}, + "session_h20_hours": {"n": 4, "min": 1.1702147743437026, "max": 1.3566088432735868, "distinct_n": 4}, + "scheduler_records": {"n": 4, "min": 37001, "max": 41348, "distinct_n": 4}, + "invariants": { + "controller_complete": true, + "four_sessions_complete": true, + "twelve_surface_outcomes": true, + "ratios_bounded": true, + "nonnegative_counts_and_costs": true, + "surface_not_all_identical": true, + "request_hashes_match": true, + "no_censored_runs": true + }, + "red_flags": [] + } +} diff --git a/runs/active-intervention-v0/train_policy.py b/runs/active-intervention-v0/train_policy.py new file mode 100644 index 0000000..bd8359c --- /dev/null +++ b/runs/active-intervention-v0/train_policy.py @@ -0,0 +1,519 @@ +#!/usr/bin/env python3 +"""Train and audit outcome-only versus telemetry action-response policies.""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +import os +import sys +from collections import defaultdict +from pathlib import Path +from typing import Any, Mapping, Sequence + + +HERE = Path(__file__).resolve().parent + + +def _load_model(): + spec = importlib.util.spec_from_file_location( + "active_intervention_model", HERE / "model.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +MODEL = _load_model() +REGULARIZATION = 10.0 +MINIMUM_MARGIN = 0.02 +CONFIDENCE_Z = 1.0 +ACCEPTABLE_REGRET = 0.02 + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8" + ) + os.replace(temporary, path) + + +def grouped( + examples: Sequence[Mapping[str, Any]], key: str +) -> dict[str, list[Mapping[str, Any]]]: + result: dict[str, list[Mapping[str, Any]]] = defaultdict(list) + for example in examples: + result[str(example[key])].append(example) + return dict(result) + + +def evaluate_grouped_cv( + examples: Sequence[Mapping[str, Any]], + *, + include_telemetry: bool, + holdout_key: str, +) -> dict[str, Any]: + holdouts = grouped(examples, holdout_key) + decision_rows = [] + for held_out, test_examples in sorted(holdouts.items()): + training = [example for example in examples if str(example[holdout_key]) != held_out] + if len({str(example["decision_id"]) for example in training}) < 2: + continue + model = MODEL.fit_ridge( + training, + include_telemetry=include_telemetry, + regularization=REGULARIZATION, + ) + for decision_id, candidates in sorted(grouped(test_examples, "decision_id").items()): + predictions = [] + for candidate in candidates: + source = candidate["source"] + action = candidate["action"] + noop = ( + int(action["target_mns"]) == int(source["mns"]) + and int(action["target_mbbt"]) == int(source["mbbt"]) + ) + if noop: + prediction = 0.0 + else: + names, vector = MODEL.feature_vector( + candidate, include_telemetry=include_telemetry + ) + if tuple(names) != model.feature_names: + raise ValueError("cross-validation feature schema mismatch") + prediction = max(-1.0, min(1.0, model.predict(vector))) + predictions.append( + { + "action_id": str(candidate["action"]["id"]), + "prediction": prediction, + "real": float(candidate["target_normalized_goodput"]), + } + ) + predictions.sort(key=lambda row: (-row["prediction"], row["action_id"])) + selected = predictions[0] + oracle = max(row["real"] for row in predictions) + regret = 1.0 - selected["real"] / oracle if oracle > 0 else 0.0 + best_actions = { + row["action_id"] for row in predictions if math.isclose(row["real"], oracle) + } + acceptable_actions = { + row["action_id"] + for row in predictions + if oracle <= 0 + or 1.0 - float(row["real"]) / oracle <= ACCEPTABLE_REGRET + 1e-12 + } + decision_rows.append( + { + "holdout": held_out, + "decision_id": decision_id, + "selected_action": selected["action_id"], + "best_actions": sorted(best_actions), + "acceptable_actions": sorted(acceptable_actions), + "correct": regret <= ACCEPTABLE_REGRET + 1e-12, + "selected_real": selected["real"], + "oracle_real": oracle, + "regret": regret, + "predictions": predictions, + } + ) + if not decision_rows: + return {"status": "INSUFFICIENT_GROUPS", "decisions": []} + regrets = [float(row["regret"]) for row in decision_rows] + return { + "status": "VALID", + "holdout_key": holdout_key, + "acceptable_regret": ACCEPTABLE_REGRET, + "decision_n": len(decision_rows), + "correct_n": sum(bool(row["correct"]) for row in decision_rows), + "accuracy": sum(bool(row["correct"]) for row in decision_rows) / len(decision_rows), + "mean_regret": sum(regrets) / len(regrets), + "max_regret": max(regrets), + "decisions": decision_rows, + } + + +def paired_delta(outcome: Mapping[str, Any], telemetry: Mapping[str, Any]) -> dict[str, Any]: + if outcome.get("status") != "VALID" or telemetry.get("status") != "VALID": + return {"status": "INSUFFICIENT_GROUPS"} + outcome_by_id = {row["decision_id"]: row for row in outcome["decisions"]} + telemetry_by_id = {row["decision_id"]: row for row in telemetry["decisions"]} + common = sorted(set(outcome_by_id) & set(telemetry_by_id)) + rows = [] + for decision_id in common: + before = outcome_by_id[decision_id] + after = telemetry_by_id[decision_id] + rows.append( + { + "decision_id": decision_id, + "outcome_action": before["selected_action"], + "telemetry_action": after["selected_action"], + "action_changed": before["selected_action"] != after["selected_action"], + "regret_delta": float(after["regret"]) - float(before["regret"]), + "telemetry_corrected": (not before["correct"]) and bool(after["correct"]), + "telemetry_harmed": bool(before["correct"]) and (not after["correct"]), + } + ) + return { + "status": "VALID", + "decision_n": len(rows), + "action_changed_n": sum(row["action_changed"] for row in rows), + "corrected_n": sum(row["telemetry_corrected"] for row in rows), + "harmed_n": sum(row["telemetry_harmed"] for row in rows), + "mean_regret_delta": ( + sum(float(row["regret_delta"]) for row in rows) / len(rows) if rows else 0.0 + ), + "rows": rows, + } + + +def evaluate_sequential_measurement_cv( + examples: Sequence[Mapping[str, Any]], + *, + include_telemetry: bool, + holdout_key: str, +) -> dict[str, Any]: + """Replay a two-consecutive-confident-checkpoint measurement policy.""" + + phases = sorted({str(example["phase"]) for example in examples}, key=float) + holdouts = grouped(examples, holdout_key) + rows = [] + full_duration_s = max(float(example["cutoff_s"]) for example in examples) + for held_out, test_examples in sorted(holdouts.items()): + training = [ + example for example in examples if str(example[holdout_key]) != held_out + ] + if len({str(example["decision_id"]) for example in training}) < 3: + continue + phase_models = {} + for phase in phases: + phase_training = [ + example for example in training if str(example["phase"]) == phase + ] + phase_models[phase] = MODEL.fit_jackknife_ensemble( + phase_training, + include_telemetry=include_telemetry, + regularization=REGULARIZATION, + ) + for decision_id, decision_examples in sorted( + grouped(test_examples, "decision_id").items() + ): + checkpoints = [] + by_phase = grouped(decision_examples, "phase") + for phase in phases: + candidates = by_phase[phase] + decision = MODEL.select_action( + phase_models[phase], + candidates, + include_telemetry=include_telemetry, + confidence_z=CONFIDENCE_Z, + minimum_margin=MINIMUM_MARGIN, + ) + checkpoints.append( + { + "phase": phase, + "cutoff_s": float(candidates[0]["cutoff_s"]), + **decision, + } + ) + selected_checkpoint = checkpoints[-1] + stop_reason = "full_measurement_fallback" + for previous, current in zip(checkpoints, checkpoints[1:], strict=False): + if ( + previous["confident"] + and current["confident"] + and previous["selected_action"] == current["selected_action"] + ): + selected_checkpoint = current + stop_reason = "two_consecutive_confident_checkpoints" + break + candidates = by_phase[str(selected_checkpoint["phase"])] + real_by_action = { + str(candidate["action"]["id"]): float( + candidate["target_normalized_goodput"] + ) + for candidate in candidates + } + target_by_action = { + str(candidate["action"]["id"]): str( + candidate["action"]["target_config_id"] + ) + for candidate in candidates + } + selected_action = str(selected_checkpoint["selected_action"]) + oracle = max(real_by_action.values()) + selected_real = real_by_action[selected_action] + regret = 1.0 - selected_real / oracle if oracle > 0 else 0.0 + source_tp = 4 + target_s = 0.0 if selected_action == "noop" else full_duration_s + replay_gpu_seconds = source_tp * ( + float(selected_checkpoint["cutoff_s"]) + target_s + ) + rows.append( + { + "holdout": held_out, + "decision_id": decision_id, + "selected_phase": str(selected_checkpoint["phase"]), + "selected_cutoff_s": float(selected_checkpoint["cutoff_s"]), + "stop_reason": stop_reason, + "selected_action": selected_action, + "selected_target_config_id": target_by_action[selected_action], + "selected_real": selected_real, + "oracle_real": oracle, + "regret": regret, + "acceptable": regret <= ACCEPTABLE_REGRET + 1e-12, + "replay_gpu_seconds_lower_bound": replay_gpu_seconds, + "checkpoints": checkpoints, + } + ) + if not rows: + return {"status": "INSUFFICIENT_GROUPS", "decisions": []} + regrets = [float(row["regret"]) for row in rows] + cutoffs = [float(row["selected_cutoff_s"]) for row in rows] + costs = [float(row["replay_gpu_seconds_lower_bound"]) for row in rows] + return { + "status": "VALID", + "holdout_key": holdout_key, + "measurement_rule": "earliest two consecutive confident checkpoints; otherwise full", + "acceptable_regret": ACCEPTABLE_REGRET, + "decision_n": len(rows), + "acceptable_n": sum(bool(row["acceptable"]) for row in rows), + "mean_regret": sum(regrets) / len(regrets), + "max_regret": max(regrets), + "mean_cutoff_s": sum(cutoffs) / len(cutoffs), + "total_replay_gpu_seconds_lower_bound": sum(costs), + "total_replay_h20_hours_lower_bound": sum(costs) / 3600.0, + "decisions": rows, + } + + +def paired_sequential_delta( + outcome: Mapping[str, Any], telemetry: Mapping[str, Any] +) -> dict[str, Any]: + if outcome.get("status") != "VALID" or telemetry.get("status") != "VALID": + return {"status": "INSUFFICIENT_GROUPS"} + before_by_id = {row["decision_id"]: row for row in outcome["decisions"]} + after_by_id = {row["decision_id"]: row for row in telemetry["decisions"]} + rows = [] + for decision_id in sorted(set(before_by_id) & set(after_by_id)): + before = before_by_id[decision_id] + after = after_by_id[decision_id] + rows.append( + { + "decision_id": decision_id, + "outcome_action": before["selected_action"], + "telemetry_action": after["selected_action"], + "outcome_cutoff_s": before["selected_cutoff_s"], + "telemetry_cutoff_s": after["selected_cutoff_s"], + "outcome_regret": before["regret"], + "telemetry_regret": after["regret"], + "regret_delta": float(after["regret"]) - float(before["regret"]), + "gpu_seconds_delta": float( + after["replay_gpu_seconds_lower_bound"] + ) + - float(before["replay_gpu_seconds_lower_bound"]), + "telemetry_corrected": (not before["acceptable"]) + and bool(after["acceptable"]), + "telemetry_harmed": bool(before["acceptable"]) + and (not after["acceptable"]), + } + ) + outcome_cost = float(outcome["total_replay_gpu_seconds_lower_bound"]) + telemetry_cost = float(telemetry["total_replay_gpu_seconds_lower_bound"]) + return { + "status": "VALID", + "decision_n": len(rows), + "corrected_n": sum(row["telemetry_corrected"] for row in rows), + "harmed_n": sum(row["telemetry_harmed"] for row in rows), + "outcome_replay_gpu_seconds_lower_bound": outcome_cost, + "telemetry_replay_gpu_seconds_lower_bound": telemetry_cost, + "gpu_cost_reduction_fraction": ( + 1.0 - telemetry_cost / outcome_cost if outcome_cost > 0 else 0.0 + ), + "rows": rows, + } + + +def build_policy(dataset_path: Path) -> dict[str, Any]: + dataset = json.loads(dataset_path.read_text(encoding="utf-8")) + if dataset.get("status") != "VALID" or dataset["sanity"]["red_flags"]: + raise ValueError("training dataset is not valid") + examples = dataset["examples"] + phases = sorted({str(example["phase"]) for example in examples}, key=float) + phase_results = {} + incremental_candidates = [] + for phase in phases: + selected = [example for example in examples if str(example["phase"]) == phase] + outcome_cv = evaluate_grouped_cv( + selected, include_telemetry=False, holdout_key="repetition" + ) + telemetry_cv = evaluate_grouped_cv( + selected, include_telemetry=True, holdout_key="repetition" + ) + outcome_regime = evaluate_grouped_cv( + selected, include_telemetry=False, holdout_key="regime" + ) + telemetry_regime = evaluate_grouped_cv( + selected, include_telemetry=True, holdout_key="regime" + ) + delta = paired_delta(outcome_cv, telemetry_cv) + outcome_models = MODEL.fit_jackknife_ensemble( + selected, + include_telemetry=False, + regularization=REGULARIZATION, + ) + telemetry_models = MODEL.fit_jackknife_ensemble( + selected, + include_telemetry=True, + regularization=REGULARIZATION, + ) + incremental = bool( + delta.get("status") == "VALID" + and int(delta["corrected_n"]) >= 1 + and int(delta["harmed_n"]) == 0 + and float(delta["mean_regret_delta"]) < -1e-12 + and float(telemetry_cv["max_regret"]) <= 0.05 + and telemetry_regime.get("status") == "VALID" + and float(telemetry_regime["mean_regret"]) + <= float(outcome_regime["mean_regret"]) + 1e-12 + and float(telemetry_regime["max_regret"]) <= 0.05 + ) + if incremental: + incremental_candidates.append(phase) + phase_results[phase] = { + "cutoff_s": float(selected[0]["cutoff_s"]), + "outcome_only": { + "leave_repetition_out": outcome_cv, + "leave_regime_out": outcome_regime, + "models": MODEL.models_to_json(outcome_models), + }, + "telemetry": { + "leave_repetition_out": telemetry_cv, + "leave_regime_out": telemetry_regime, + "models": MODEL.models_to_json(telemetry_models), + }, + "paired_incremental": delta, + "incremental_gate": incremental, + } + outcome_sequential = evaluate_sequential_measurement_cv( + examples, include_telemetry=False, holdout_key="repetition" + ) + telemetry_sequential = evaluate_sequential_measurement_cv( + examples, include_telemetry=True, holdout_key="repetition" + ) + sequential_delta = paired_sequential_delta( + outcome_sequential, telemetry_sequential + ) + retrospective_cost_gate = bool( + sequential_delta.get("status") == "VALID" + and int(sequential_delta["harmed_n"]) == 0 + and int(telemetry_sequential["acceptable_n"]) + >= int(outcome_sequential["acceptable_n"]) + and float(telemetry_sequential["max_regret"]) <= 0.05 + and float(sequential_delta["gpu_cost_reduction_fraction"]) >= 0.10 + ) + status = ( + "RETROSPECTIVE_GPU_COST_SIGNAL" + if retrospective_cost_gate + else "NO_RETROSPECTIVE_GPU_COST_SIGNAL" + ) + target_values = [float(example["target_normalized_goodput"]) for example in examples] + effect_values = [ + float(example["target_delta_normalized_goodput"]) for example in examples + ] + invariants = { + "four_phases": len(phases) == 4, + "targets_bounded": all(0.0 <= value <= 1.0 for value in target_values), + "targets_not_all_identical": len(set(target_values)) > 1, + "effects_bounded": all(-1.0 <= value <= 1.0 for value in effect_values), + "effects_not_all_identical": len(set(effect_values)) > 1, + "models_present_every_phase": all( + phase_results[phase][mode]["models"] + for phase in phases + for mode in ("outcome_only", "telemetry") + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + if red_flags: + raise RuntimeError(f"policy sanity failed: {red_flags}") + return { + "schema": "active-intervention-policy-v0", + "status": status, + "training": { + "dataset": str(dataset_path), + "dataset_sha256": sha256_file(dataset_path), + "examples": len(examples), + "decisions": len({example["decision_id"] for example in examples}), + "regularization": REGULARIZATION, + "confidence_z": CONFIDENCE_Z, + "minimum_margin": MINIMUM_MARGIN, + "acceptable_regret": ACCEPTABLE_REGRET, + }, + "measurement_policy": { + "rule": "earliest two consecutive confident checkpoints; otherwise full", + "checkpoints": [phase_results[phase]["cutoff_s"] for phase in phases], + "confidence_z": CONFIDENCE_Z, + "minimum_margin": MINIMUM_MARGIN, + }, + "sequential_replay": { + "outcome_only": outcome_sequential, + "telemetry": telemetry_sequential, + "paired_delta": sequential_delta, + "retrospective_gpu_cost_gate": retrospective_cost_gate, + "minimum_cost_reduction_fraction": 0.10, + }, + "phases": phase_results, + "sanity": { + "invariants": invariants, + "red_flags": red_flags, + "target_normalized_goodput": { + "n": len(target_values), + "min": min(target_values), + "max": max(target_values), + "distinct_n": len(set(target_values)), + }, + "target_delta_normalized_goodput": { + "n": len(effect_values), + "min": min(effect_values), + "max": max(effect_values), + "distinct_n": len(set(effect_values)), + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--dataset", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + policy = build_policy(args.dataset) + atomic_json(args.output, policy) + print( + json.dumps( + { + "status": policy["status"], + "measurement_policy": policy["measurement_policy"], + "sanity": policy["sanity"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_existing.py b/runs/fidelity-headroom/analyze_existing.py new file mode 100644 index 0000000..7fcc74f --- /dev/null +++ b/runs/fidelity-headroom/analyze_existing.py @@ -0,0 +1,508 @@ +#!/usr/bin/env python3 +"""Retrospective headroom audit for a fidelity-aware tuning harness. + +This analysis intentionally separates two questions: + +1. How many real cell evaluations does a simulator top-k shortlist already + need to recover the real optimum on the frozen SimFid surface? +2. On the P6 anchor ladder, do Layer-1 engine features predict the next + anchor's feasibility better than outcome-only features from the same + current anchor? + +The second question is diagnostic rather than decision-bearing: it uses a +small, already-observed single-workload surface and full current-anchor +summaries. It is a premise check for a future prospective early-probe study. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from dataclasses import dataclass +from pathlib import Path +from typing import Any, Iterable + +import numpy as np + + +SCHEMA = "fidelity-headroom-v1" +DEFAULT_REGULARIZATION = 1.0 +REGULARIZATION_SENSITIVITY = (0.1, 1.0, 10.0) +BOOTSTRAP_SEED = 20260714 +BOOTSTRAP_REPLICATES = 10_000 + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def numeric(values: Iterable[float | int]) -> dict[str, Any]: + array = [float(value) for value in values] + return { + "n": len(array), + "min": min(array) if array else None, + "max": max(array) if array else None, + "distinct_n": len(set(array)), + } + + +def score_buckets(scores: dict[str, float], tolerance: float) -> dict[str, int]: + if tolerance <= 0: + raise ValueError("score tolerance must be positive") + return {cell: math.floor(float(score) / tolerance) for cell, score in scores.items()} + + +def topk_curve( + real_scores: dict[str, float], + simulated_scores: dict[str, float], + tolerance: float, +) -> dict[str, Any]: + if set(real_scores) != set(simulated_scores): + raise ValueError("real and simulator score cells differ") + buckets = score_buckets(simulated_scores, tolerance) + ordered = sorted( + simulated_scores, + key=lambda cell: (-buckets[cell], -float(simulated_scores[cell]), cell), + ) + real_best = max(float(value) for value in real_scores.values()) + points = [] + for nominal_k in range(1, len(ordered) + 1): + cutoff_bucket = buckets[ordered[nominal_k - 1]] + candidates = [cell for cell in ordered if buckets[cell] >= cutoff_bucket] + selected = max(candidates, key=lambda cell: (float(real_scores[cell]), cell)) + selected_score = float(real_scores[selected]) + points.append( + { + "nominal_k": nominal_k, + "expanded_k": len(candidates), + "candidates": candidates, + "selected_cell_after_real_final": selected, + "selected_real_score": selected_score, + "real_regret": 1.0 - selected_score / real_best, + } + ) + + minimum_k = {} + for name, threshold in (("zero", 1e-15), ("one_percent", 0.01), ("five_percent", 0.05)): + eligible = [point for point in points if point["real_regret"] <= threshold] + minimum_k[name] = ( + { + "nominal_k": eligible[0]["nominal_k"], + "expanded_k": eligible[0]["expanded_k"], + } + if eligible + else None + ) + return { + "real_best": real_best, + "minimum_k": minimum_k, + "points": points, + } + + +@dataclass(frozen=True) +class Transition: + cell: str + current_anchor: float + next_anchor: float + external: tuple[float, ...] + instrumentation: tuple[float, ...] + next_feasible: int + + +EXTERNAL_FEATURES = ( + "log_current_rate_per_gpu", + "log_next_over_current_rate", + "log2_tp", + "log2_mns", + "current_pass_rate", + "ttft_max_over_6s", + "tpot_max_over_50ms", + "exact_output_fraction", + "early_stopped", +) + +INSTRUMENTATION_FEATURES = ( + "waiting_mean", + "waiting_max", + "decode_batch_mean", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "model_steps_per_second", +) + + +def _finite(value: float | int | None) -> float: + if value is None: + return 0.0 + result = float(value) + if not math.isfinite(result): + raise ValueError(f"non-finite feature: {value}") + return result + + +def build_transitions(phase6: dict[str, Any]) -> list[Transition]: + transitions = [] + for cell, cell_result in sorted(phase6["cells"].items()): + anchors = sorted(cell_result["anchors"], key=lambda item: float(item["anchor"])) + for current, following in zip(anchors, anchors[1:]): + if following["accepted_feasible"] is None: + continue + primary = current["primary"] + next_primary = following["primary"] + layer = current["layer1"] + rate = float(primary["selection"]["offered_req_s_per_gpu"]) + next_rate = float(next_primary["selection"]["offered_req_s_per_gpu"]) + selected_count = int(primary["selection"]["count"]) + if rate <= 0 or next_rate <= 0 or selected_count <= 0: + raise ValueError("rates and selected counts must be positive") + external = ( + math.log(rate), + math.log(next_rate / rate), + math.log2(float(cell_result["tp"])), + math.log2(float(cell_result["mns"])), + float(primary["pass_rate"]), + _finite(primary["ttft_ms"]["max"]) / 6000.0, + _finite(primary["tpot_ms"]["max"]) / 50.0, + float(primary["exact_output_count"]) / selected_count, + float(bool(primary["early_stopped"])), + ) + graph_shares = layer.get("graph_mode_shares", {}) + prefill_tokens = _finite(layer["prefill_tokens"]) + decode_tokens = _finite(layer["decode_tokens"]) + instrumentation = ( + _finite(layer["waiting_mean"]), + _finite(layer["waiting_max"]), + _finite(layer["decode_B_mean"]), + _finite(layer["decode_B_cv"]), + _finite(layer["kv_usage_mean"]), + _finite(layer["kv_usage_max"]), + float(graph_shares.get("NONE", 0.0)), + float(graph_shares.get("FULL", 0.0)), + _finite(layer["padding_fraction"]), + prefill_tokens / max(1.0, prefill_tokens + decode_tokens), + _finite(layer["model_steps"]) / float(primary["interval"]["elapsed_s"]), + ) + transitions.append( + Transition( + cell=cell, + current_anchor=float(current["anchor"]), + next_anchor=float(following["anchor"]), + external=external, + instrumentation=instrumentation, + next_feasible=int(bool(following["accepted_feasible"])), + ) + ) + return transitions + + +def _sigmoid(values: np.ndarray) -> np.ndarray: + clipped = np.clip(values, -30.0, 30.0) + return 1.0 / (1.0 + np.exp(-clipped)) + + +def _fit_logistic(x: np.ndarray, y: np.ndarray, regularization: float) -> np.ndarray: + weights = np.zeros(x.shape[1], dtype=np.float64) + penalty = np.eye(x.shape[1], dtype=np.float64) + penalty[0, 0] = 0.0 + for _ in range(100): + probability = _sigmoid(x @ weights) + gradient = x.T @ (probability - y) / len(y) + gradient += regularization * penalty @ weights / len(y) + curvature = probability * (1.0 - probability) + hessian = (x.T * curvature) @ x / len(y) + hessian += regularization * penalty / len(y) + step = np.linalg.lstsq(hessian, gradient, rcond=None)[0] + weights -= step + if float(np.max(np.abs(step))) < 1e-9: + break + return weights + + +def _classification_metrics(y: np.ndarray, probability: np.ndarray) -> dict[str, Any]: + if np.any(probability < 0.0) or np.any(probability > 1.0): + raise ValueError("classification probabilities must be in [0, 1]") + prediction = probability >= 0.5 + true_positive = int(np.sum(prediction & (y == 1))) + true_negative = int(np.sum(~prediction & (y == 0))) + false_positive = int(np.sum(prediction & (y == 0))) + false_negative = int(np.sum(~prediction & (y == 1))) + positive_total = true_positive + false_negative + negative_total = true_negative + false_positive + balanced = 0.5 * ( + true_positive / positive_total + true_negative / negative_total + ) + clipped = np.clip(probability, 1e-12, 1.0 - 1e-12) + return { + "accuracy": float(np.mean(prediction == y)), + "balanced_accuracy": float(balanced), + "brier": float(np.mean((probability - y) ** 2)), + "log_loss": float(np.mean(-(y * np.log(clipped) + (1 - y) * np.log(1 - clipped)))), + "confusion": { + "true_positive": true_positive, + "true_negative": true_negative, + "false_positive": false_positive, + "false_negative": false_negative, + }, + } + + +def _mcnemar_exact_p(outcome_only_correct: int, instrumentation_only_correct: int) -> float: + discordant = outcome_only_correct + instrumentation_only_correct + if discordant == 0: + return 1.0 + tail = sum( + math.comb(discordant, value) + for value in range(min(outcome_only_correct, instrumentation_only_correct) + 1) + ) / (2**discordant) + return min(1.0, 2.0 * tail) + + +def grouped_predictions( + transitions: list[Transition], + *, + instrumentation_aware: bool, + regularization: float, +) -> tuple[np.ndarray, np.ndarray, list[str]]: + probabilities = [] + labels = [] + test_cells = [] + for held_out in sorted({transition.cell for transition in transitions}): + train = [transition for transition in transitions if transition.cell != held_out] + test = [transition for transition in transitions if transition.cell == held_out] + + def row(transition: Transition) -> np.ndarray: + values = transition.external + if instrumentation_aware: + values += transition.instrumentation + return np.asarray((1.0, *values), dtype=np.float64) + + x_train = np.stack([row(transition) for transition in train]) + x_test = np.stack([row(transition) for transition in test]) + y_train = np.asarray([transition.next_feasible for transition in train], dtype=np.float64) + mean = x_train[:, 1:].mean(axis=0) + standard_deviation = x_train[:, 1:].std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + x_train[:, 1:] = (x_train[:, 1:] - mean) / standard_deviation + x_test[:, 1:] = (x_test[:, 1:] - mean) / standard_deviation + weights = _fit_logistic(x_train, y_train, regularization) + probabilities.extend(_sigmoid(x_test @ weights).tolist()) + labels.extend(transition.next_feasible for transition in test) + test_cells.extend(held_out for _ in test) + return ( + np.asarray(labels, dtype=np.int64), + np.asarray(probabilities, dtype=np.float64), + test_cells, + ) + + +def _group_bootstrap_delta( + y: np.ndarray, + outcome_probability: np.ndarray, + instrumentation_probability: np.ndarray, + cells: list[str], +) -> dict[str, Any]: + groups = sorted(set(cells)) + indices = {group: np.asarray([i for i, cell in enumerate(cells) if cell == group]) for group in groups} + random = np.random.default_rng(BOOTSTRAP_SEED) + accuracy_deltas = [] + brier_deltas = [] + for _ in range(BOOTSTRAP_REPLICATES): + sampled = random.choice(groups, size=len(groups), replace=True) + selected = np.concatenate([indices[group] for group in sampled]) + selected_y = y[selected] + outcome = outcome_probability[selected] + instrumentation = instrumentation_probability[selected] + accuracy_deltas.append( + float(np.mean((instrumentation >= 0.5) == selected_y)) + - float(np.mean((outcome >= 0.5) == selected_y)) + ) + brier_deltas.append( + float(np.mean((instrumentation - selected_y) ** 2)) + - float(np.mean((outcome - selected_y) ** 2)) + ) + return { + "semantics": "group bootstrap over cells; diagnostic confidence interval", + "replicates": BOOTSTRAP_REPLICATES, + "seed": BOOTSTRAP_SEED, + "accuracy_delta_instrumentation_minus_outcome": { + "point": float(np.mean((instrumentation_probability >= 0.5) == y)) + - float(np.mean((outcome_probability >= 0.5) == y)), + "ci95": [float(x) for x in np.percentile(accuracy_deltas, [2.5, 97.5])], + }, + "brier_delta_instrumentation_minus_outcome": { + "point": float(np.mean((instrumentation_probability - y) ** 2)) + - float(np.mean((outcome_probability - y) ** 2)), + "ci95": [float(x) for x in np.percentile(brier_deltas, [2.5, 97.5])], + }, + } + + +def transition_analysis(transitions: list[Transition]) -> dict[str, Any]: + sensitivity = {} + headline_payload = None + for regularization in REGULARIZATION_SENSITIVITY: + y, outcome_probability, cells = grouped_predictions( + transitions, + instrumentation_aware=False, + regularization=regularization, + ) + instrumentation_y, instrumentation_probability, instrumentation_cells = grouped_predictions( + transitions, + instrumentation_aware=True, + regularization=regularization, + ) + if not np.array_equal(y, instrumentation_y) or cells != instrumentation_cells: + raise AssertionError("model folds or labels differ") + outcome_correct = (outcome_probability >= 0.5) == y + instrumentation_correct = (instrumentation_probability >= 0.5) == y + payload = { + "outcome_only": _classification_metrics(y, outcome_probability), + "instrumentation_aware": _classification_metrics(y, instrumentation_probability), + "paired_correctness": { + "both_correct": int(np.sum(outcome_correct & instrumentation_correct)), + "outcome_only_correct": int(np.sum(outcome_correct & ~instrumentation_correct)), + "instrumentation_only_correct": int(np.sum(~outcome_correct & instrumentation_correct)), + "both_wrong": int(np.sum(~outcome_correct & ~instrumentation_correct)), + }, + "bootstrap": _group_bootstrap_delta( + y, + outcome_probability, + instrumentation_probability, + cells, + ), + } + payload["paired_correctness"]["mcnemar_exact_two_sided_p"] = _mcnemar_exact_p( + payload["paired_correctness"]["outcome_only_correct"], + payload["paired_correctness"]["instrumentation_only_correct"], + ) + sensitivity[str(regularization)] = payload + if regularization == DEFAULT_REGULARIZATION: + headline_payload = payload + assert headline_payload is not None + labels = [transition.next_feasible for transition in transitions] + accuracy_deltas = [ + value["instrumentation_aware"]["accuracy"] - value["outcome_only"]["accuracy"] + for value in sensitivity.values() + ] + brier_deltas = [ + value["instrumentation_aware"]["brier"] - value["outcome_only"]["brier"] + for value in sensitivity.values() + ] + return { + "status": "RETROSPECTIVE_DIAGNOSTIC_ONLY", + "estimand": "next-anchor feasibility from the full current-anchor summary", + "split": "leave-one-cell-out", + "model": "L2 logistic regression with train-fold standardization", + "external_features": list(EXTERNAL_FEATURES), + "instrumentation_features": list(INSTRUMENTATION_FEATURES), + "headline_regularization": DEFAULT_REGULARIZATION, + "headline": headline_payload, + "regularization_sensitivity": sensitivity, + "sensitivity_summary": { + "accuracy_delta_min_max": [min(accuracy_deltas), max(accuracy_deltas)], + "brier_delta_min_max": [min(brier_deltas), max(brier_deltas)], + "incremental_signal_verdict": "NEEDS_PROSPECTIVE_EVIDENCE", + }, + "label_sanity": { + **numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + }, + } + + +def analyze(simfid_path: Path, phase6_path: Path) -> dict[str, Any]: + simfid = json.loads(simfid_path.read_text()) + phase6 = json.loads(phase6_path.read_text()) + real_scores = {cell: float(score) for cell, score in simfid["real_scores"].items()} + topk = {} + for reading, payload in sorted(simfid["analyses"].items()): + tie = payload["metrics"]["tie_buckets"]["simulator"] + topk[reading] = topk_curve( + real_scores, + {cell: float(score) for cell, score in payload["simulated_scores"].items()}, + float(tie["tolerance"]), + ) + transitions = build_transitions(phase6) + transition_result = transition_analysis(transitions) + red_flags = [] + if len(real_scores) != 12: + red_flags.append("unexpected_simfid_cell_count") + if len(transitions) == 0 or len(set(x.next_feasible for x in transitions)) != 2: + red_flags.append("transition_labels_missing_or_single_class") + if any(not math.isfinite(value) or value < 0 for value in real_scores.values()): + red_flags.append("invalid_real_score") + return { + "schema": SCHEMA, + "status": "PASS" if not red_flags else "STOP", + "scope": "retrospective single-workload premise audit; not prospective contribution evidence", + "provenance": { + "simfid_metrics": str(simfid_path.resolve()), + "simfid_sha256": sha256_file(simfid_path), + "phase6_metrics": str(phase6_path.resolve()), + "phase6_sha256": sha256_file(phase6_path), + }, + "topk_headroom": topk, + "next_anchor_prediction": transition_result, + "decision": { + "current_surface_can_show_selection_contribution": False, + "reason": ( + "The strongest frozen-calibrated SLO reading reaches zero real regret " + "after real evaluation of its first two-cell tie bucket. A method that " + "requires one calibration probe and one final verification cannot use " + "this single task to demonstrate fewer real cell evaluations." + ), + "prospective_target": ( + "Test whether internal features from a short, shared real probe reduce " + "the number or duration of full frontier evaluations relative to an " + "outcome-only model given the same probe." + ), + }, + "sanity": { + "real_scores": numeric(real_scores.values()), + "simulator_readings": len(topk), + "transitions": len(transitions), + "transition_cells": len({transition.cell for transition in transitions}), + "red_flags": red_flags, + "invariants": { + "same_cells_all_readings": all( + set(payload["simulated_scores"]) == set(real_scores) + for payload in simfid["analyses"].values() + ), + "scores_nonnegative": all(value >= 0 for value in real_scores.values()), + "transition_features_finite": all( + all(math.isfinite(value) for value in (*item.external, *item.instrumentation)) + for item in transitions + ), + "probabilities_bounded": True, + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--simfid-metrics", type=Path, required=True) + parser.add_argument("--phase6-metrics", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze(args.simfid_metrics, args.phase6_metrics) + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + print(json.dumps({"status": result["status"], "output": str(args.output)}, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_pilot.py b/runs/fidelity-headroom/analyze_pilot.py new file mode 100644 index 0000000..b1293b3 --- /dev/null +++ b/runs/fidelity-headroom/analyze_pilot.py @@ -0,0 +1,336 @@ +#!/usr/bin/env python3 +"""Evaluate frozen outcome-only and instrumentation-aware policies on P1.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path +from typing import Any + +import numpy as np + +from analyze_existing import _classification_metrics, _mcnemar_exact_p +from analyze_prefixes import ( + PrefixExample, + _load_jsonl, + _prefix_features, + numeric, + policy_metrics, + predict_frozen_model, + sha256_file, +) + + +def result_path(run_root: Path, cell: str, level: str, replicate: int) -> Path: + return run_root / "cells" / cell / f"{level}-rep{replicate}" / "result.json" + + +def requests_path(run_root: Path, cell: str, level: str, replicate: int) -> Path: + return run_root / "cells" / cell / f"{level}-rep{replicate}" / "requests.jsonl" + + +def selection_for( + manifest: dict[str, Any], cell: str, level: str, replicate: int +) -> dict[str, Any]: + role = f"{level}{replicate}" + return manifest["cells"][cell]["targets"][level]["selections"][role] + + +def campaign_gpu_accounting( + primary_state_path: Path, prior_state_paths: tuple[Path, ...] = () +) -> dict[str, Any]: + attempts = [] + for role, path in ( + [("prior_failure", path) for path in prior_state_paths] + + [("primary", primary_state_path)] + ): + state = json.loads(path.read_text(encoding="utf-8")) + gpu_hours = float(state["gpu_hours_total"]) + attempts.append( + { + "role": role, + "path": str(path.resolve()), + "sha256": sha256_file(path), + "status": state["status"], + "h20_hours": gpu_hours, + } + ) + total = sum(attempt["h20_hours"] for attempt in attempts) + primary = json.loads(primary_state_path.read_text(encoding="utf-8")) + hard_cap = float(primary["hard_cap_h20_hours"]) + return { + "attempts": attempts, + "aggregate_h20_hours": total, + "hard_cap_h20_hours": hard_cap, + "invariants": { + "costs_nonnegative": all( + attempt["h20_hours"] >= 0.0 for attempt in attempts + ), + "aggregate_below_cap": 0.0 <= total < hard_cap, + }, + } + + +def build_pilot_examples( + manifest: dict[str, Any], run_root: Path, cutoff_s: float +) -> tuple[list[PrefixExample], list[dict[str, Any]], list[str]]: + examples = [] + details = [] + red_flags = [] + for cell, config in sorted(manifest["cells"].items()): + stream_path = next((run_root / "cells" / cell / "opprof").glob("*.jsonl")) + stream = _load_jsonl(stream_path, require_key="submit_mono_ns") + for level in ("low", "high"): + results = [ + json.loads(result_path(run_root, cell, level, replicate).read_text()) + for replicate in (1, 2, 3) + ] + votes = [bool(result["feasible"]) for result in results] + adjudicated = sum(votes) >= 2 + primary = results[0] + requests = _load_jsonl(requests_path(run_root, cell, level, 1)) + exact_timestamps = sum( + request.get("completed_elapsed_s") is not None for request in requests + ) + actual_outcomes = sum( + request.get("completed_mono_ns") is not None for request in requests + ) + if exact_timestamps != actual_outcomes: + red_flags.append(f"timestamp_count_mismatch_{cell}_{level}") + expected = selection_for(manifest, cell, level, 1) + if int(primary["selection"]["count"]) != int(expected["selected_count"]): + red_flags.append(f"selection_count_mismatch_{cell}_{level}") + for result_key, manifest_key in ( + ("request_id_order_sha256", "request_id_order_sha256"), + ("arrival_order_sha256", "arrival_order_sha256"), + ("raw_length_order_sha256", "input_length_order_sha256"), + ): + if primary["selection"][result_key] != expected[manifest_key]: + red_flags.append(f"selection_hash_mismatch_{cell}_{level}_{result_key}") + start_ns = int(primary["interval"]["start_mono_ns"]) + end_ns = start_ns + int(cutoff_s * 1e9) + records = [ + record + for record in stream + if record.get("model_executed") + and start_ns <= int(record["submit_mono_ns"]) <= end_ns + ] + outcome, instrumentation, completion_source = _prefix_features( + primary=primary, + tp=int(config["tp"]), + max_num_seqs=int(config["mns"]), + requests=requests, + records=records, + cutoff_s=cutoff_s, + ) + example = PrefixExample( + cell=cell, + anchor=float(primary["anchor"]), + cutoff_s=cutoff_s, + tp=int(config["tp"]), + full_elapsed_s=float(primary["interval"]["elapsed_s"]), + feasible=int(adjudicated), + primary_feasible=int(bool(primary["feasible"])), + outcome=outcome, + instrumentation=instrumentation, + completion_time_source=completion_source, + ) + examples.append(example) + details.append( + { + "cell": cell, + "level": level, + "anchor_rep1": primary["anchor"], + "selected_count_rep1": primary["selection"]["count"], + "votes": votes, + "pass_rates": [result["pass_rate"] for result in results], + "adjudicated_feasible": adjudicated, + "primary_feasible": bool(primary["feasible"]), + "actual_timestamped_outcomes": actual_outcomes, + "selected_outcomes": len(requests), + "prefix_layer1_records": len(records), + "completion_time_source": completion_source, + } + ) + return examples, details, red_flags + + +def analyze( + manifest_path: Path, + model_path: Path, + run_root: Path, + prior_state_paths: tuple[Path, ...] = (), +) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + models = json.loads(model_path.read_text(encoding="utf-8")) + state_path = run_root / "controller-state.json" + state = json.loads(state_path.read_text(encoding="utf-8")) + gpu_accounting = campaign_gpu_accounting(state_path, prior_state_paths) + cutoff_s = float(models["cutoff_s"]) + threshold = float(models["accept_probability"]) + examples, details, red_flags = build_pilot_examples(manifest, run_root, cutoff_s) + labels = np.asarray([example.feasible for example in examples], dtype=np.int64) + outcome_probability = predict_frozen_model(models["models"]["outcome_only"], examples) + instrumentation_probability = predict_frozen_model( + models["models"]["instrumentation_aware"], examples + ) + outcome_policy = policy_metrics( + examples, labels, outcome_probability, threshold + ) + instrumentation_policy = policy_metrics( + examples, labels, instrumentation_probability, threshold + ) + outcome_correct = (outcome_probability >= 0.5) == labels + instrumentation_correct = (instrumentation_probability >= 0.5) == labels + paired = { + "both_correct": int(np.sum(outcome_correct & instrumentation_correct)), + "outcome_only_correct": int(np.sum(outcome_correct & ~instrumentation_correct)), + "instrumentation_only_correct": int(np.sum(~outcome_correct & instrumentation_correct)), + "both_wrong": int(np.sum(~outcome_correct & ~instrumentation_correct)), + } + paired["mcnemar_exact_two_sided_p"] = _mcnemar_exact_p( + paired["outcome_only_correct"], paired["instrumentation_only_correct"] + ) + for detail, outcome_p, instrumentation_p in zip( + details, outcome_probability, instrumentation_probability + ): + detail["outcome_probability_feasible"] = float(outcome_p) + detail["instrumentation_probability_feasible"] = float(instrumentation_p) + + positive = int(np.sum(labels)) + negative = len(labels) - positive + if state["status"] != "complete" or int(state["completed_cells"]) != 6: + red_flags.append("campaign_incomplete") + if positive < 3 or negative < 3: + red_flags.append("insufficient_label_balance") + if any( + detail["actual_timestamped_outcomes"] == 0 for detail in details + ): + red_flags.append("no_exact_request_timestamps") + if not all(gpu_accounting["invariants"].values()): + red_flags.append("hard_cap_exceeded") + + outcome_errors = outcome_policy["false_accept"] + outcome_policy["false_reject"] + instrumentation_errors = ( + instrumentation_policy["false_accept"] + + instrumentation_policy["false_reject"] + ) + outcome_decisions = outcome_policy["early_accept"] + outcome_policy["early_reject"] + instrumentation_decisions = ( + instrumentation_policy["early_accept"] + + instrumentation_policy["early_reject"] + ) + outcome_reduction = outcome_policy["valid_cost_reduction_fraction"] + instrumentation_reduction = instrumentation_policy["valid_cost_reduction_fraction"] + cost_delta = ( + instrumentation_reduction - outcome_reduction + if outcome_reduction is not None and instrumentation_reduction is not None + else None + ) + data_valid = not red_flags + safety_gate = instrumentation_errors == 0 and instrumentation_errors <= outcome_errors + incremental_gate = ( + instrumentation_decisions - outcome_decisions >= 3 + or (cost_delta is not None and cost_delta >= 0.15) + ) + pilot_pass = data_valid and safety_gate and incremental_gate + + return { + "schema": "fidelity-prefix-pilot-result-v1", + "status": "PILOT_PASS" if pilot_pass else "PILOT_FAIL", + "scope": "held-out single-task gate; not paper-facing contribution evidence", + "provenance": { + "manifest": str(manifest_path.resolve()), + "manifest_sha256": sha256_file(manifest_path), + "frozen_models": str(model_path.resolve()), + "frozen_models_sha256": sha256_file(model_path), + "controller_state": str(state_path.resolve()), + "controller_state_sha256": sha256_file(state_path), + }, + "cutoff_s": cutoff_s, + "threshold": threshold, + "examples": details, + "outcome_only": { + "classification": _classification_metrics(labels, outcome_probability), + "policy": outcome_policy, + }, + "instrumentation_aware": { + "classification": _classification_metrics(labels, instrumentation_probability), + "policy": instrumentation_policy, + }, + "paired_correctness": paired, + "gate": { + "data_valid": data_valid, + "safety_gate": safety_gate, + "incremental_gate": incremental_gate, + "additional_early_decisions": instrumentation_decisions - outcome_decisions, + "valid_cost_reduction_fraction_delta": cost_delta, + "opens_expanded_p2": pilot_pass, + }, + "gpu": { + "primary_attempt_h20_hours": state["gpu_hours_total"], + **gpu_accounting, + }, + "sanity": { + "red_flags": red_flags, + "labels": { + **numeric(labels.tolist()), + "positive": positive, + "negative": negative, + }, + "full_elapsed_s": numeric(example.full_elapsed_s for example in examples), + "remaining_h20_hours": numeric( + example.remaining_h20_hours for example in examples + ), + "outcome_probability": numeric(outcome_probability.tolist()), + "instrumentation_probability": numeric( + instrumentation_probability.tolist() + ), + "invariants": { + "examples_12": len(examples) == 12, + "cells_6": len({example.cell for example in examples}) == 6, + "ratios_bounded": bool( + np.all((outcome_probability >= 0) & (outcome_probability <= 1)) + and np.all( + (instrumentation_probability >= 0) + & (instrumentation_probability <= 1) + ) + ), + "costs_nonnegative": all( + example.remaining_h20_hours >= 0 for example in examples + ), + "all_cell_validations": all( + all(cell["validation"]["invariants"].values()) + for cell in state["cells"].values() + ), + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--frozen-models", type=Path, required=True) + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--prior-state", type=Path, action="append", default=[]) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze( + args.manifest, + args.frozen_models, + args.run_root, + tuple(args.prior_state), + ) + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + print(json.dumps({ + "status": result["status"], + "gate": result["gate"], + "sanity_red_flags": result["sanity"]["red_flags"], + }, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_pilot_e2e.py b/runs/fidelity-headroom/analyze_pilot_e2e.py new file mode 100644 index 0000000..09a6581 --- /dev/null +++ b/runs/fidelity-headroom/analyze_pilot_e2e.py @@ -0,0 +1,429 @@ +#!/usr/bin/env python3 +"""Replay the P1 simulator shortlist under full and prefix policies.""" + +from __future__ import annotations + +import argparse +import json +import math +import subprocess +from pathlib import Path +from typing import Any + +from analyze_prefixes import numeric, sha256_file + + +AITUNER_ROOT = Path(__file__).resolve().parents[2] +FROZEN_K = 2 +CUTOFF_S = 5.0 +THRESHOLD = 0.95 + + +def git_capture(*arguments: str) -> str: + return subprocess.run( + ["git", "-C", str(AITUNER_ROOT), *arguments], + check=True, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + ).stdout + + +def setup_costs(state: dict[str, Any]) -> dict[str, float]: + result = {} + for cell, payload in state["cells"].items(): + tp = int(payload["tp"]) + annotation_intervals = sum( + float(run["elapsed_s"]) * tp / 3600.0 + for run in payload["runs"] + if run["role"] not in {"low1", "high1"} + ) + primary_intervals = sum( + float(run["elapsed_s"]) * tp / 3600.0 + for run in payload["runs"] + if run["role"] in {"low1", "high1"} + ) + setup = float(payload["gpu_hours"]) - annotation_intervals - primary_intervals + if setup < -1e-12: + raise ValueError(f"negative inferred setup cost: {cell}={setup}") + result[cell] = max(0.0, setup) + return result + + +def build_candidates( + manifest: dict[str, Any], + state: dict[str, Any], + strong: dict[str, Any], +) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: + baseline_probability = strong["headline"]["sim_plus_outcome"]["probability"] + instrument_probability = strong["headline"][ + "sim_plus_outcome_plus_instrumentation" + ]["probability"] + setup = setup_costs(state) + anchors = [] + for detail, baseline_p, instrument_p in zip( + strong["pilot_examples"], baseline_probability, instrument_probability + ): + cell = str(detail["cell"]) + level = str(detail["level"]) + role = f"{level}1" + selection = manifest["cells"][cell]["targets"][level]["selections"][role] + run = next( + item for item in state["cells"][cell]["runs"] if item["role"] == role + ) + tp = int(state["cells"][cell]["tp"]) + full_cost = float(run["elapsed_s"]) * tp / 3600.0 + prefix_cost = min(CUTOFF_S, float(run["elapsed_s"])) * tp / 3600.0 + anchors.append( + { + "cell": cell, + "level": level, + "role": role, + "tp": tp, + "real_feasible": bool(detail["adjudicated_feasible"]), + "real_goodput_req_s_per_gpu": float( + selection["offered_req_s_per_gpu"] + ), + "sim_feasible": bool(detail["sim_slo_feasible"]), + "sim_pass_rate": float(detail["sim_slo_pass_rate"]), + "sim_throughput_req_s_per_gpu": float( + detail["sim_completed_throughput_per_gpu"] + ), + "baseline_probability": float(baseline_p), + "instrument_probability": float(instrument_p), + "setup_h20_hours": setup[cell], + "full_trial_h20_hours": full_cost, + "prefix_h20_hours": prefix_cost, + } + ) + + candidates = [] + for cell in sorted(manifest["cells"]): + feasible = [ + anchor for anchor in anchors if anchor["cell"] == cell and anchor["sim_feasible"] + ] + if not feasible: + continue + candidates.append( + max(feasible, key=lambda anchor: anchor["sim_throughput_req_s_per_gpu"]) + ) + candidates.sort( + key=lambda anchor: ( + -anchor["sim_throughput_req_s_per_gpu"], + anchor["cell"], + ) + ) + return anchors, candidates + + +def expanded_top_k(candidates: list[dict[str, Any]], k: int) -> list[dict[str, Any]]: + if not candidates or k <= 0: + return [] + boundary = candidates[min(k, len(candidates)) - 1][ + "sim_throughput_req_s_per_gpu" + ] + return [ + candidate + for candidate in candidates + if candidate["sim_throughput_req_s_per_gpu"] >= boundary - 1e-12 + ] + + +def selected_result( + evaluated: list[dict[str, Any]], feasible_key: str +) -> tuple[str | None, float | None]: + feasible = [candidate for candidate in evaluated if candidate[feasible_key]] + if not feasible: + return None, None + best = max(feasible, key=lambda candidate: candidate["real_goodput_req_s_per_gpu"]) + return str(best["cell"]), float(best["real_goodput_req_s_per_gpu"]) + + +def replay( + shortlist: list[dict[str, Any]], + *, + probability_key: str | None, + oracle_goodput: float, + common_failure_h20_hours: float, +) -> dict[str, Any]: + evaluated = [] + online_cost = 0.0 + early_accept = 0 + early_reject = 0 + false_accept = 0 + false_reject = 0 + for candidate in shortlist: + current = dict(candidate) + online_cost += current["setup_h20_hours"] + if probability_key is None: + predicted_feasible = current["real_feasible"] + online_cost += current["full_trial_h20_hours"] + action = "full" + else: + probability = float(current[probability_key]) + if probability >= THRESHOLD: + predicted_feasible = True + early_accept += 1 + online_cost += current["prefix_h20_hours"] + action = "early_accept" + false_accept += int(not current["real_feasible"]) + elif probability <= 1.0 - THRESHOLD: + predicted_feasible = False + early_reject += 1 + online_cost += current["prefix_h20_hours"] + action = "early_reject" + false_reject += int(current["real_feasible"]) + else: + predicted_feasible = current["real_feasible"] + online_cost += current["full_trial_h20_hours"] + action = "continue_full" + current["policy_feasible"] = predicted_feasible + current["action"] = action + evaluated.append(current) + selected_cell, selected_goodput = selected_result(evaluated, "policy_feasible") + regret = ( + 1.0 - selected_goodput / oracle_goodput + if selected_goodput is not None and oracle_goodput > 0 + else None + ) + return { + "selected_cell": selected_cell, + "selected_real_goodput_req_s_per_gpu": selected_goodput, + "real_regret": regret, + "online_h20_hours": online_cost, + "conservative_h20_hours_with_prior_failure": ( + online_cost + common_failure_h20_hours + ), + "early_accept": early_accept, + "early_reject": early_reject, + "false_accept": false_accept, + "false_reject": false_reject, + "evaluated": [ + { + "cell": item["cell"], + "level": item["level"], + "action": item["action"], + "real_feasible": item["real_feasible"], + "policy_feasible": item["policy_feasible"], + } + for item in evaluated + ], + } + + +def analyze( + manifest_path: Path, + state_path: Path, + prior_state_path: Path, + strong_path: Path, +) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + state = json.loads(state_path.read_text(encoding="utf-8")) + prior = json.loads(prior_state_path.read_text(encoding="utf-8")) + strong = json.loads(strong_path.read_text(encoding="utf-8")) + anchors, candidates = build_candidates(manifest, state, strong) + oracle_anchor = max( + (anchor for anchor in anchors if anchor["real_feasible"]), + key=lambda anchor: anchor["real_goodput_req_s_per_gpu"], + ) + oracle_goodput = float(oracle_anchor["real_goodput_req_s_per_gpu"]) + common_failure = float(prior["gpu_hours_total"]) + by_k = {} + for k in (1, 2, 3, 6): + shortlist = expanded_top_k(candidates, k) + full = replay( + shortlist, + probability_key=None, + oracle_goodput=oracle_goodput, + common_failure_h20_hours=common_failure, + ) + baseline = replay( + shortlist, + probability_key="baseline_probability", + oracle_goodput=oracle_goodput, + common_failure_h20_hours=common_failure, + ) + instrument = replay( + shortlist, + probability_key="instrument_probability", + oracle_goodput=oracle_goodput, + common_failure_h20_hours=common_failure, + ) + for result in (baseline, instrument): + result["online_cost_reduction_vs_full"] = ( + 1.0 - result["online_h20_hours"] / full["online_h20_hours"] + ) + result["conservative_cost_reduction_vs_full"] = 1.0 - ( + result["conservative_h20_hours_with_prior_failure"] + / full["conservative_h20_hours_with_prior_failure"] + ) + by_k[str(k)] = { + "actual_shortlist_size": len(shortlist), + "shortlist": [candidate["cell"] for candidate in shortlist], + "sim_top_k_plus_real_final": full, + "sim_plus_outcome": baseline, + "sim_plus_outcome_plus_instrumentation": instrument, + } + + frozen = by_k[str(FROZEN_K)] + full = frozen["sim_top_k_plus_real_final"] + baseline = frozen["sim_plus_outcome"] + instrument = frozen["sim_plus_outcome_plus_instrumentation"] + baseline_safe = baseline["false_accept"] == 0 and baseline["false_reject"] == 0 + instrument_safe = ( + instrument["false_accept"] == 0 and instrument["false_reject"] == 0 + ) + incremental_reduction = ( + 1.0 - instrument["online_h20_hours"] / baseline["online_h20_hours"] + if baseline_safe and instrument_safe and baseline["online_h20_hours"] > 0 + else None + ) + contribution_gate = { + "frozen_k": FROZEN_K, + "instrument_safe": instrument_safe, + "outcome_baseline_safe": baseline_safe, + "instrument_regret_at_most_5pct": ( + instrument["real_regret"] is not None + and instrument["real_regret"] <= 0.05 + ), + "instrument_cost_reduction_vs_full_at_least_30pct": ( + instrument["online_cost_reduction_vs_full"] >= 0.30 + ), + "instrument_cost_reduction_vs_outcome_at_least_20pct": ( + incremental_reduction is not None and incremental_reduction >= 0.20 + ), + "incremental_reduction_vs_outcome": incremental_reduction, + } + contribution_gate["passes"] = all( + contribution_gate[key] + for key in ( + "instrument_safe", + "outcome_baseline_safe", + "instrument_regret_at_most_5pct", + "instrument_cost_reduction_vs_full_at_least_30pct", + "instrument_cost_reduction_vs_outcome_at_least_20pct", + ) + ) + + red_flags = [] + if state["status"] != "complete" or int(state["completed_cells"]) != 6: + red_flags.append("pilot_incomplete") + if strong["status"] != "PASS" or strong["sanity"]["red_flags"]: + red_flags.append("strong_input_invalid") + if len(anchors) != 12 or len(candidates) != 6: + red_flags.append("unexpected_surface_size") + probabilities = [ + value + for anchor in anchors + for value in (anchor["baseline_probability"], anchor["instrument_probability"]) + ] + costs = [ + value + for anchor in anchors + for value in ( + anchor["setup_h20_hours"], + anchor["full_trial_h20_hours"], + anchor["prefix_h20_hours"], + ) + ] + if not all(0.0 <= value <= 1.0 for value in probabilities): + red_flags.append("probability_out_of_range") + if not all(value >= 0.0 and math.isfinite(value) for value in costs): + red_flags.append("invalid_cost") + + return { + "schema": "fidelity-pilot-e2e-v1", + "status": "PASS" if not red_flags else "STOP", + "scope": "held-out P1 replay; gate diagnostic, not paper-facing evidence", + "ranking": [ + { + "rank": rank, + "cell": candidate["cell"], + "level": candidate["level"], + "sim_throughput_req_s_per_gpu": candidate[ + "sim_throughput_req_s_per_gpu" + ], + "real_feasible": candidate["real_feasible"], + "real_goodput_req_s_per_gpu": candidate[ + "real_goodput_req_s_per_gpu" + ], + } + for rank, candidate in enumerate(candidates, start=1) + ], + "real_oracle": { + "cell": oracle_anchor["cell"], + "level": oracle_anchor["level"], + "goodput_req_s_per_gpu": oracle_goodput, + }, + "by_k": by_k, + "contribution_gate": contribution_gate, + "analysis": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": git_capture("rev-parse", "HEAD").strip(), + "aituner_git_status_short": git_capture("status", "--short"), + }, + "provenance": { + "manifest": str(manifest_path.resolve()), + "manifest_sha256": sha256_file(manifest_path), + "controller_state": str(state_path.resolve()), + "controller_state_sha256": sha256_file(state_path), + "prior_state": str(prior_state_path.resolve()), + "prior_state_sha256": sha256_file(prior_state_path), + "strong_metrics": str(strong_path.resolve()), + "strong_metrics_sha256": sha256_file(strong_path), + }, + "sanity": { + "red_flags": red_flags, + "anchors": numeric([1 for _ in anchors]), + "candidates": numeric([1 for _ in candidates]), + "probabilities": numeric(probabilities), + "costs_h20_hours": numeric(costs), + "invariants": { + "anchors_12": len(anchors) == 12, + "candidates_6": len(candidates) == 6, + "probabilities_bounded": all( + 0.0 <= value <= 1.0 for value in probabilities + ), + "costs_nonnegative": all(value >= 0.0 for value in costs), + "per_config_not_all_identical": len( + {candidate["sim_throughput_req_s_per_gpu"] for candidate in candidates} + ) + > 1, + "tie_expansion_applied": True, + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--controller-state", type=Path, required=True) + parser.add_argument("--prior-state", type=Path, required=True) + parser.add_argument("--strong-metrics", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze( + args.manifest, + args.controller_state, + args.prior_state, + args.strong_metrics, + ) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + print( + json.dumps( + { + "status": result["status"], + "red_flags": result["sanity"]["red_flags"], + "contribution_gate": result["contribution_gate"], + }, + sort_keys=True, + ) + ) + if result["status"] != "PASS": + raise RuntimeError(result["sanity"]["red_flags"]) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_prefixes.py b/runs/fidelity-headroom/analyze_prefixes.py new file mode 100644 index 0000000..8fb480b --- /dev/null +++ b/runs/fidelity-headroom/analyze_prefixes.py @@ -0,0 +1,629 @@ +#!/usr/bin/env python3 +"""Retrospective, leakage-bounded audit of short real-probe prefixes. + +The outcome-only and instrumentation-aware models receive the same trial +prefix. The latter differs only by Layer-1 engine state. Existing Phase-6 +request artifacts predate exact completion timestamps, so their completion +time is reconstructed from arrival + TTFT + token intervals and is explicitly +marked approximate. New artifacts use ``completed_elapsed_s`` directly. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from dataclasses import dataclass +from pathlib import Path +from typing import Any, Iterable + +import numpy as np + +from analyze_existing import ( + DEFAULT_REGULARIZATION, + REGULARIZATION_SENSITIVITY, + _classification_metrics, + _fit_logistic, + _group_bootstrap_delta, + _mcnemar_exact_p, + _sigmoid, +) + + +SCHEMA = "fidelity-prefix-v1" +DEFAULT_CUTOFFS = (5.0, 10.0, 15.0, 20.0) +POLICY_THRESHOLDS = (0.8, 0.9, 0.95) + +OUTCOME_FEATURES = ( + "log_offered_rate_per_gpu", + "log2_tp", + "log2_max_num_seqs", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit", +) + +INSTRUMENTATION_FEATURES = ( + "model_steps_per_second", + "waiting_mean", + "waiting_max", + "waiting_nonzero_share", + "running_mean", + "running_max", + "decode_batch_mean", + "decode_batch_max", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "preemptions", +) + + +@dataclass(frozen=True) +class PrefixExample: + cell: str + anchor: float + cutoff_s: float + tp: int + full_elapsed_s: float + feasible: int + primary_feasible: int + outcome: tuple[float, ...] + instrumentation: tuple[float, ...] + completion_time_source: str + + @property + def remaining_h20_hours(self) -> float: + return self.tp * max(0.0, self.full_elapsed_s - self.cutoff_s) / 3600.0 + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def numeric(values: Iterable[float | int]) -> dict[str, Any]: + array = [float(value) for value in values] + return { + "n": len(array), + "min": min(array) if array else None, + "max": max(array) if array else None, + "distinct_n": len(set(array)), + } + + +def _cv(values: list[float]) -> float: + if not values: + return 0.0 + array = np.asarray(values, dtype=np.float64) + mean = float(array.mean()) + return float(array.std(ddof=0) / mean) if mean else 0.0 + + +def completion_elapsed_s(request: dict[str, Any]) -> tuple[float | None, str]: + exact = request.get("completed_elapsed_s") + if exact is not None: + value = float(exact) + if value < 0 or not math.isfinite(value): + raise ValueError(f"invalid completed_elapsed_s={exact}") + return value, "exact_monotonic" + if not request.get("success"): + return None, "unobserved_failure" + required = ( + request.get("arrival_s"), + request.get("ttft_ms"), + request.get("tpot_ms"), + request.get("completion_tokens"), + ) + if any(value is None for value in required): + return None, "unobserved_failure" + arrival_s, ttft_ms, tpot_ms, completion_tokens = required + value = float(arrival_s) + ( + float(ttft_ms) + max(int(completion_tokens) - 1, 0) * float(tpot_ms) + ) / 1000.0 + if value < 0 or not math.isfinite(value): + raise ValueError(f"invalid reconstructed completion time={value}") + return value, "reconstructed_from_latency" + + +def _load_jsonl(path: Path, *, require_key: str | None = None) -> list[dict[str, Any]]: + records = [] + with path.open(encoding="utf-8") as source: + for line in source: + item = json.loads(line) + if require_key is None or require_key in item: + records.append(item) + return records + + +def _anchor_directory(cell_root: Path, anchor: float) -> Path: + matches = [] + for result_path in cell_root.glob("anchor-*/result.json"): + payload = json.loads(result_path.read_text(encoding="utf-8")) + if math.isclose(float(payload["anchor"]), anchor, rel_tol=0.0, abs_tol=1e-15): + matches.append(result_path.parent) + if len(matches) != 1: + raise ValueError(f"expected one primary directory for anchor {anchor}: {matches}") + return matches[0] + + +def _prefix_features( + *, + primary: dict[str, Any], + tp: int, + max_num_seqs: int, + requests: list[dict[str, Any]], + records: list[dict[str, Any]], + cutoff_s: float, +) -> tuple[tuple[float, ...], tuple[float, ...], str]: + admitted = [request for request in requests if float(request["arrival_s"]) <= cutoff_s] + completed = [] + sources = set() + for request in requests: + completed_s, source = completion_elapsed_s(request) + if completed_s is None or completed_s > cutoff_s: + continue + completed.append(request) + sources.add(source) + if not admitted or not records: + raise ValueError("prefix has no admitted requests or Layer-1 records") + if any(request not in admitted for request in completed): + raise ValueError("completed request was not admitted inside prefix") + + total = len(requests) + passed = sum(bool(request["slo_pass"]) for request in completed) + ttft = [float(request["ttft_ms"]) for request in completed if request["ttft_ms"] is not None] + tpot = [float(request["tpot_ms"]) for request in completed if request["tpot_ms"] is not None] + offered_rate = float(primary["selection"]["offered_req_s_per_gpu"]) + if offered_rate <= 0 or total <= 0: + raise ValueError("offered rate and selected request count must be positive") + + outcome = ( + math.log(offered_rate), + math.log2(float(tp)), + math.log2(float(max_num_seqs)), + len(admitted) / total, + len(completed) / len(admitted), + passed / max(1, len(completed)), + (len(completed) - passed) / total, + (len(admitted) - len(completed)) / len(admitted), + max(ttft, default=0.0) / 6000.0, + float(np.mean(ttft)) / 6000.0 if ttft else 0.0, + max(tpot, default=0.0) / 50.0, + float(np.mean(tpot)) / 50.0 if tpot else 0.0, + float(np.mean([float(request["raw_input_tokens"]) for request in admitted])) / 8192.0, + ) + + waiting = [float(record["queues"]["waiting"]) for record in records] + running = [float(record["queues"]["running"]) for record in records] + decode_batch = [float(record["decode_batch_size"]) for record in records] + kv_usage = [float(record["kv"]["usage"]) for record in records] + graph_modes = [str(record["cudagraph"]["runtime_mode"]) for record in records] + bucket_tokens = sum(int(record["cudagraph"]["bucket_tokens"]) for record in records) + padding_tokens = sum(int(record["cudagraph"]["padding_tokens"]) for record in records) + prefill_tokens = sum(int(record["prefill_tokens"]) for record in records) + decode_tokens = sum(int(record["decode_tokens"]) for record in records) + instrumentation = ( + len(records) / cutoff_s, + float(np.mean(waiting)), + max(waiting), + sum(value > 0 for value in waiting) / len(waiting), + float(np.mean(running)), + max(running), + float(np.mean(decode_batch)), + max(decode_batch), + _cv(decode_batch), + float(np.mean(kv_usage)), + max(kv_usage), + kv_usage[-1] - kv_usage[0], + graph_modes.count("NONE") / len(graph_modes), + graph_modes.count("FULL") / len(graph_modes), + padding_tokens / max(1, bucket_tokens), + prefill_tokens / max(1, prefill_tokens + decode_tokens), + float(sum(int(record["preemptions"]) for record in records)), + ) + completion_source = "+".join(sorted(sources)) if sources else "none_completed" + return outcome, instrumentation, completion_source + + +def build_examples( + phase6: dict[str, Any], + raw_root: Path, + cutoff_s: float, +) -> list[PrefixExample]: + examples = [] + for cell, cell_result in sorted(phase6["cells"].items()): + cell_root = raw_root / cell + stream_path = next((cell_root / "opprof").glob("*.jsonl")) + stream = _load_jsonl(stream_path, require_key="submit_mono_ns") + for anchor in cell_result["anchors"]: + primary = anchor["primary"] + full_elapsed_s = float(primary["interval"]["elapsed_s"]) + if full_elapsed_s + 1e-9 < cutoff_s: + continue + anchor_value = float(primary["anchor"]) + anchor_root = _anchor_directory(cell_root, anchor_value) + requests = _load_jsonl(anchor_root / "requests.jsonl") + start_ns = int(primary["interval"]["start_mono_ns"]) + end_ns = start_ns + int(cutoff_s * 1e9) + records = [ + record + for record in stream + if record.get("model_executed") + and start_ns <= int(record["submit_mono_ns"]) <= end_ns + ] + outcome, instrumentation, source = _prefix_features( + primary=primary, + tp=int(cell_result["tp"]), + max_num_seqs=int(cell_result["mns"]), + requests=requests, + records=records, + cutoff_s=cutoff_s, + ) + examples.append( + PrefixExample( + cell=cell, + anchor=anchor_value, + cutoff_s=cutoff_s, + tp=int(cell_result["tp"]), + full_elapsed_s=full_elapsed_s, + feasible=int(bool(anchor["accepted_feasible"])), + primary_feasible=int(bool(primary["feasible"])), + outcome=outcome, + instrumentation=instrumentation, + completion_time_source=source, + ) + ) + return examples + + +def grouped_predictions( + examples: list[PrefixExample], + *, + instrumentation_aware: bool, + regularization: float, +) -> tuple[np.ndarray, np.ndarray, list[str]]: + probabilities = [] + labels = [] + groups = [] + for held_out in sorted({example.cell for example in examples}): + train = [example for example in examples if example.cell != held_out] + test = [example for example in examples if example.cell == held_out] + + def row(example: PrefixExample) -> np.ndarray: + values = example.outcome + if instrumentation_aware: + values += example.instrumentation + return np.asarray((1.0, *values), dtype=np.float64) + + x_train = np.stack([row(example) for example in train]) + x_test = np.stack([row(example) for example in test]) + y_train = np.asarray([example.feasible for example in train], dtype=np.float64) + if len(set(y_train.tolist())) != 2: + raise ValueError(f"training fold for {held_out} has a single label") + mean = x_train[:, 1:].mean(axis=0) + standard_deviation = x_train[:, 1:].std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + x_train[:, 1:] = (x_train[:, 1:] - mean) / standard_deviation + x_test[:, 1:] = (x_test[:, 1:] - mean) / standard_deviation + weights = _fit_logistic(x_train, y_train, regularization) + probabilities.extend(_sigmoid(x_test @ weights).tolist()) + labels.extend(example.feasible for example in test) + groups.extend(held_out for _ in test) + return ( + np.asarray(labels, dtype=np.int64), + np.asarray(probabilities, dtype=np.float64), + groups, + ) + + +def fit_frozen_model( + examples: list[PrefixExample], + *, + instrumentation_aware: bool, + regularization: float, +) -> dict[str, Any]: + def row(example: PrefixExample) -> np.ndarray: + values = example.outcome + if instrumentation_aware: + values += example.instrumentation + return np.asarray((1.0, *values), dtype=np.float64) + + matrix = np.stack([row(example) for example in examples]) + labels = np.asarray([example.feasible for example in examples], dtype=np.float64) + if len(set(labels.tolist())) != 2: + raise ValueError("frozen model requires both feasibility labels") + mean = matrix[:, 1:].mean(axis=0) + standard_deviation = matrix[:, 1:].std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + standardized = matrix.copy() + standardized[:, 1:] = (standardized[:, 1:] - mean) / standard_deviation + weights = _fit_logistic(standardized, labels, regularization) + probabilities = _sigmoid(standardized @ weights) + names = list(OUTCOME_FEATURES) + if instrumentation_aware: + names.extend(INSTRUMENTATION_FEATURES) + return { + "instrumentation_aware": instrumentation_aware, + "regularization": regularization, + "feature_names": names, + "feature_mean": mean.tolist(), + "feature_standard_deviation": standard_deviation.tolist(), + "weights_with_intercept_first": weights.tolist(), + "training_classification": _classification_metrics(labels, probabilities), + } + + +def predict_frozen_model( + model: dict[str, Any], + examples: list[PrefixExample], +) -> np.ndarray: + instrumentation_aware = bool(model["instrumentation_aware"]) + rows = [] + for example in examples: + values = example.outcome + if instrumentation_aware: + values += example.instrumentation + rows.append((1.0, *values)) + matrix = np.asarray(rows, dtype=np.float64) + mean = np.asarray(model["feature_mean"], dtype=np.float64) + standard_deviation = np.asarray( + model["feature_standard_deviation"], dtype=np.float64 + ) + weights = np.asarray(model["weights_with_intercept_first"], dtype=np.float64) + if matrix.shape[1] != len(weights) or matrix.shape[1] - 1 != len(mean): + raise ValueError("frozen model feature dimensions do not match examples") + matrix[:, 1:] = (matrix[:, 1:] - mean) / standard_deviation + return _sigmoid(matrix @ weights) + + +def policy_metrics( + examples: list[PrefixExample], + labels: np.ndarray, + probabilities: np.ndarray, + threshold: float, +) -> dict[str, Any]: + accept = probabilities >= threshold + reject = probabilities <= 1.0 - threshold + decide = accept | reject + prediction = accept.astype(np.int64) + correct = prediction == labels + remaining = np.asarray( + [example.remaining_h20_hours for example in examples], dtype=np.float64 + ) + full_cost = sum(example.tp * example.full_elapsed_s / 3600.0 for example in examples) + saved = float(np.sum(remaining[decide])) + correct_saved = float(np.sum(remaining[decide & correct])) + invalid_saved = float(np.sum(remaining[decide & ~correct])) + + def describe(mask: np.ndarray) -> list[dict[str, Any]]: + return [ + { + "cell": example.cell, + "anchor": example.anchor, + "label_feasible": bool(label), + "probability_feasible": float(probability), + "remaining_h20_hours": example.remaining_h20_hours, + } + for example, label, probability, selected in zip( + examples, labels, probabilities, mask + ) + if selected + ] + + return { + "threshold": threshold, + "early_accept": int(np.sum(accept)), + "early_reject": int(np.sum(reject)), + "abstain_continue_full": int(np.sum(~decide)), + "false_accept": int(np.sum(accept & (labels == 0))), + "false_reject": int(np.sum(reject & (labels == 1))), + "false_accept_examples": describe(accept & (labels == 0)), + "false_reject_examples": describe(reject & (labels == 1)), + "decision_coverage": float(np.mean(decide)), + "full_trial_h20_hours": float(full_cost), + "remaining_h20_hours_at_cutoff": float(np.sum(remaining)), + "saved_h20_hours_if_decisions_used": saved, + "correctly_saved_h20_hours": correct_saved, + "invalidly_saved_h20_hours": invalid_saved, + "valid_zero_error_policy": bool(np.all(correct[decide])), + "valid_cost_reduction_fraction": ( + correct_saved / full_cost if invalid_saved == 0.0 and full_cost else None + ), + } + + +def analyze_cutoff(examples: list[PrefixExample]) -> dict[str, Any]: + sensitivity = {} + headline = None + for regularization in REGULARIZATION_SENSITIVITY: + labels, outcome_probability, groups = grouped_predictions( + examples, + instrumentation_aware=False, + regularization=regularization, + ) + instrument_labels, instrument_probability, instrument_groups = grouped_predictions( + examples, + instrumentation_aware=True, + regularization=regularization, + ) + if not np.array_equal(labels, instrument_labels) or groups != instrument_groups: + raise AssertionError("paired folds or labels differ") + if groups != [example.cell for example in examples]: + raise AssertionError("prediction order differs from example order") + outcome_correct = (outcome_probability >= 0.5) == labels + instrument_correct = (instrument_probability >= 0.5) == labels + result = { + "outcome_only": { + "classification": _classification_metrics(labels, outcome_probability), + "policies": [ + policy_metrics(examples, labels, outcome_probability, threshold) + for threshold in POLICY_THRESHOLDS + ], + }, + "instrumentation_aware": { + "classification": _classification_metrics(labels, instrument_probability), + "policies": [ + policy_metrics(examples, labels, instrument_probability, threshold) + for threshold in POLICY_THRESHOLDS + ], + }, + "paired_correctness": { + "both_correct": int(np.sum(outcome_correct & instrument_correct)), + "outcome_only_correct": int(np.sum(outcome_correct & ~instrument_correct)), + "instrumentation_only_correct": int(np.sum(~outcome_correct & instrument_correct)), + "both_wrong": int(np.sum(~outcome_correct & ~instrument_correct)), + }, + "bootstrap": _group_bootstrap_delta( + labels, + outcome_probability, + instrument_probability, + groups, + ), + } + paired = result["paired_correctness"] + paired["mcnemar_exact_two_sided_p"] = _mcnemar_exact_p( + paired["outcome_only_correct"], paired["instrumentation_only_correct"] + ) + sensitivity[str(regularization)] = result + if regularization == DEFAULT_REGULARIZATION: + headline = result + assert headline is not None + labels = [example.feasible for example in examples] + return { + "examples": len(examples), + "cells": len({example.cell for example in examples}), + "label_sanity": { + **numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + "primary_adjudicated_disagreements": sum( + example.feasible != example.primary_feasible for example in examples + ), + }, + "completion_time_sources": { + source: sum(example.completion_time_source == source for example in examples) + for source in sorted({example.completion_time_source for example in examples}) + }, + "headline_regularization": DEFAULT_REGULARIZATION, + "headline": headline, + "regularization_sensitivity": sensitivity, + "remaining_h20_hours": numeric( + example.remaining_h20_hours for example in examples + ), + } + + +def analyze( + phase6_path: Path, + raw_root: Path, + cutoffs: tuple[float, ...], +) -> dict[str, Any]: + phase6 = json.loads(phase6_path.read_text(encoding="utf-8")) + by_cutoff = {} + red_flags = [] + for cutoff in cutoffs: + examples = build_examples(phase6, raw_root, cutoff) + if len({example.feasible for example in examples}) != 2: + red_flags.append(f"single_label_at_{cutoff:g}s") + continue + by_cutoff[f"{cutoff:g}"] = analyze_cutoff(examples) + if len({example.cell for example in examples}) != 12: + red_flags.append(f"incomplete_cells_at_{cutoff:g}s") + if not all( + math.isfinite(value) + for example in examples + for value in (*example.outcome, *example.instrumentation) + ): + red_flags.append(f"nonfinite_features_at_{cutoff:g}s") + + headline_deltas = { + cutoff: { + "accuracy": ( + result["headline"]["instrumentation_aware"]["classification"]["accuracy"] + - result["headline"]["outcome_only"]["classification"]["accuracy"] + ), + "brier": ( + result["headline"]["instrumentation_aware"]["classification"]["brier"] + - result["headline"]["outcome_only"]["classification"]["brier"] + ), + } + for cutoff, result in by_cutoff.items() + } + return { + "schema": SCHEMA, + "status": "PASS" if not red_flags else "STOP", + "scope": ( + "retrospective single-workload prefix diagnostic; model selection, " + "threshold choice, and contribution claims require held-out prospective tasks" + ), + "estimand": ( + "2-of-3 adjudicated anchor feasibility from the first primary trial's " + "identical short real prefix" + ), + "split": "leave-one-configuration-cell-out", + "model": "same L2 logistic model and folds; instrumentation model appends Layer-1 features", + "outcome_features": list(OUTCOME_FEATURES), + "instrumentation_features": list(INSTRUMENTATION_FEATURES), + "provenance": { + "phase6_metrics": str(phase6_path.resolve()), + "phase6_metrics_sha256": sha256_file(phase6_path), + "raw_root": str(raw_root.resolve()), + }, + "cutoffs_s": list(cutoffs), + "cutoffs": by_cutoff, + "headline_incremental_deltas": headline_deltas, + "decision": { + "contribution_established": False, + "reason": ( + "This dataset contains one workload and reconstructed rather than exact request " + "completion times. Three TP4 primary trials also disagree with their 2-of-3 " + "labels. It can reject a missing-signal premise but cannot establish " + "generalization or a paper-facing cost reduction." + ), + }, + "sanity": { + "red_flags": red_flags, + "cutoff_count": len(by_cutoff), + "invariants": { + "cutoffs_positive": all(cutoff > 0 for cutoff in cutoffs), + "paired_same_model_family": True, + "probabilities_checked_in_unit_interval": True, + "full_trial_label_not_used_as_feature": True, + "records_strictly_prefix_sliced": True, + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--phase6-metrics", type=Path, required=True) + parser.add_argument("--raw-root", type=Path, required=True) + parser.add_argument("--cutoffs", type=float, nargs="+", default=DEFAULT_CUTOFFS) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze(args.phase6_metrics, args.raw_root, tuple(args.cutoffs)) + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n", encoding="utf-8") + print(json.dumps({"status": result["status"], "output": str(args.output)}, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_strong_baseline.py b/runs/fidelity-headroom/analyze_strong_baseline.py new file mode 100644 index 0000000..4880afd --- /dev/null +++ b/runs/fidelity-headroom/analyze_strong_baseline.py @@ -0,0 +1,298 @@ +#!/usr/bin/env python3 +"""Audit telemetry against a simulator-aware outcome calibration baseline. + +This is a retrospective headroom check. It strengthens the earlier +outcome-only baseline by giving both nested models the same per-anchor +Frontier throughput and SLO predictions. The only additional inputs to the +larger model are real engine Layer-1 features. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from pathlib import Path +from typing import Any + +import numpy as np + +from analyze_existing import ( + DEFAULT_REGULARIZATION, + REGULARIZATION_SENSITIVITY, + _classification_metrics, + _fit_logistic, + _group_bootstrap_delta, + _mcnemar_exact_p, + _sigmoid, +) +from analyze_prefixes import ( + INSTRUMENTATION_FEATURES, + OUTCOME_FEATURES, + PrefixExample, + build_examples, + numeric, + policy_metrics, + sha256_file, +) + + +SIMULATOR_FEATURES = ( + "log_sim_completed_throughput_per_gpu", + "sim_slo_pass_rate", + "sim_slo_feasible", +) + + +def load_simulator_features(raw_root: Path) -> tuple[dict[tuple[str, float], tuple[float, ...]], str]: + features: dict[tuple[str, float], tuple[float, ...]] = {} + digest = hashlib.sha256() + paths = sorted(raw_root.glob("*/trial-0001/run_manifest.json")) + for manifest_path in paths: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + run = manifest["run"] + if run["mode"] != "frozen-calibrated": + continue + scorer_path = manifest_path.parent / "scorer_output.json" + scorer = json.loads(scorer_path.read_text(encoding="utf-8")) + key = (str(run["cell_id"]), float(run["sampling_u"])) + if key in features: + raise ValueError(f"duplicate frozen simulator run: {key}") + throughput = float(scorer["throughput_requests_per_second_per_gpu"]) + pass_rate = float(scorer["slo"]["pass_rate"]) + if throughput <= 0 or not 0.0 <= pass_rate <= 1.0: + raise ValueError(f"invalid simulator output: {key}") + features[key] = ( + math.log(throughput), + pass_rate, + float(bool(scorer["slo"]["feasible"])), + ) + for path in (manifest_path, scorer_path): + digest.update(str(path.relative_to(raw_root)).encode()) + digest.update(path.read_bytes()) + return features, digest.hexdigest() + + +def simulator_row( + example: PrefixExample, + features: dict[tuple[str, float], tuple[float, ...]], +) -> tuple[float, ...]: + matches = [ + values + for (cell, anchor), values in features.items() + if cell == example.cell + and math.isclose(anchor, example.anchor, rel_tol=0.0, abs_tol=1e-12) + ] + if len(matches) != 1: + raise ValueError( + f"expected one simulator match for {example.cell}/{example.anchor}: {len(matches)}" + ) + return matches[0] + + +def grouped_predictions( + examples: list[PrefixExample], + simulator: dict[tuple[str, float], tuple[float, ...]], + *, + instrumentation_aware: bool, + regularization: float, +) -> tuple[np.ndarray, np.ndarray, list[str]]: + probabilities: list[float] = [] + labels: list[int] = [] + groups: list[str] = [] + for held_out in sorted({example.cell for example in examples}): + train = [example for example in examples if example.cell != held_out] + test = [example for example in examples if example.cell == held_out] + + def row(example: PrefixExample) -> np.ndarray: + values = example.outcome + simulator_row(example, simulator) + if instrumentation_aware: + values += example.instrumentation + return np.asarray((1.0, *values), dtype=np.float64) + + x_train = np.stack([row(example) for example in train]) + x_test = np.stack([row(example) for example in test]) + y_train = np.asarray([example.feasible for example in train], dtype=np.float64) + mean = x_train[:, 1:].mean(axis=0) + standard_deviation = x_train[:, 1:].std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + x_train[:, 1:] = (x_train[:, 1:] - mean) / standard_deviation + x_test[:, 1:] = (x_test[:, 1:] - mean) / standard_deviation + weights = _fit_logistic(x_train, y_train, regularization) + probabilities.extend(_sigmoid(x_test @ weights).tolist()) + labels.extend(example.feasible for example in test) + groups.extend(held_out for _ in test) + return ( + np.asarray(labels, dtype=np.int64), + np.asarray(probabilities, dtype=np.float64), + groups, + ) + + +def analyze( + phase6_path: Path, + phase6_raw_root: Path, + simulator_raw_root: Path, + simulator_metrics_path: Path, +) -> dict[str, Any]: + phase6 = json.loads(phase6_path.read_text(encoding="utf-8")) + examples = build_examples(phase6, phase6_raw_root, 5.0) + simulator, simulator_raw_sha256 = load_simulator_features(simulator_raw_root) + red_flags = [] + try: + matched = [simulator_row(example, simulator) for example in examples] + except ValueError as error: + matched = [] + red_flags.append(str(error)) + + sensitivity = {} + if matched: + for regularization in REGULARIZATION_SENSITIVITY: + labels, baseline_probability, groups = grouped_predictions( + examples, + simulator, + instrumentation_aware=False, + regularization=regularization, + ) + instrument_labels, instrument_probability, instrument_groups = grouped_predictions( + examples, + simulator, + instrumentation_aware=True, + regularization=regularization, + ) + if not np.array_equal(labels, instrument_labels) or groups != instrument_groups: + raise AssertionError("nested baseline folds differ") + baseline_correct = (baseline_probability >= 0.5) == labels + instrument_correct = (instrument_probability >= 0.5) == labels + paired = { + "both_correct": int(np.sum(baseline_correct & instrument_correct)), + "sim_outcome_only_correct": int( + np.sum(baseline_correct & ~instrument_correct) + ), + "instrumentation_only_correct": int( + np.sum(~baseline_correct & instrument_correct) + ), + "both_wrong": int(np.sum(~baseline_correct & ~instrument_correct)), + } + paired["mcnemar_exact_two_sided_p"] = _mcnemar_exact_p( + paired["sim_outcome_only_correct"], + paired["instrumentation_only_correct"], + ) + sensitivity[str(regularization)] = { + "sim_plus_outcome": { + "classification": _classification_metrics(labels, baseline_probability), + "policy_0p95": policy_metrics( + examples, labels, baseline_probability, 0.95 + ), + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": _classification_metrics(labels, instrument_probability), + "policy_0p95": policy_metrics( + examples, labels, instrument_probability, 0.95 + ), + }, + "paired_correctness": paired, + "group_bootstrap": _group_bootstrap_delta( + labels, + baseline_probability, + instrument_probability, + groups, + ), + } + + headline = sensitivity.get(str(DEFAULT_REGULARIZATION)) + simulator_pass_rates = [row[1] for row in matched] + labels = [example.feasible for example in examples] + if len(examples) != 37: + red_flags.append("examples_not_37") + if len(simulator) != 92: + red_flags.append("frozen_simulator_runs_not_92") + if len(set(labels)) != 2: + red_flags.append("single_label") + if matched and not all(0.0 <= value <= 1.0 for value in simulator_pass_rates): + red_flags.append("simulator_pass_rate_out_of_range") + + return { + "schema": "fidelity-strong-baseline-v1", + "status": "PASS" if not red_flags else "STOP", + "scope": "retrospective one-task headroom audit; not contribution evidence", + "comparison": ( + "same 5-second prefix, folds, logistic family, regularization, and frozen " + "Frontier outputs; the only nested difference is real Layer-1 engine state" + ), + "features": { + "shared_outcome": list(OUTCOME_FEATURES), + "shared_simulator": list(SIMULATOR_FEATURES), + "instrumentation_only": list(INSTRUMENTATION_FEATURES), + }, + "headline_regularization": DEFAULT_REGULARIZATION, + "headline": headline, + "regularization_sensitivity": sensitivity, + "provenance": { + "phase6_metrics": str(phase6_path.resolve()), + "phase6_metrics_sha256": sha256_file(phase6_path), + "phase6_raw_root": str(phase6_raw_root.resolve()), + "simulator_metrics": str(simulator_metrics_path.resolve()), + "simulator_metrics_sha256": sha256_file(simulator_metrics_path), + "simulator_raw_root": str(simulator_raw_root.resolve()), + "frozen_simulator_manifest_scorer_set_sha256": simulator_raw_sha256, + }, + "decision": { + "contribution_established": False, + "prospective_requirement": ( + "repeat sim+outcome versus sim+outcome+instrumentation on complete held-out tasks" + ), + }, + "sanity": { + "red_flags": red_flags, + "examples": numeric([1 for _ in examples]), + "labels": { + **numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + }, + "matched_simulator_pass_rate": numeric(simulator_pass_rates), + "frozen_simulator_runs": len(simulator), + "invariants": { + "all_examples_matched_once": len(matched) == len(examples), + "same_nested_folds": True, + "simulator_ratios_bounded": all( + 0.0 <= value <= 1.0 for value in simulator_pass_rates + ), + "labels_not_identical": len(set(labels)) == 2, + "per_config_results_not_all_identical": len(set(simulator_pass_rates)) > 1, + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--phase6-metrics", type=Path, required=True) + parser.add_argument("--phase6-raw-root", type=Path, required=True) + parser.add_argument("--simulator-raw-root", type=Path, required=True) + parser.add_argument("--simulator-metrics", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze( + args.phase6_metrics, + args.phase6_raw_root, + args.simulator_raw_root, + args.simulator_metrics, + ) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + print( + json.dumps( + { + "status": result["status"], + "output": str(args.output), + "red_flags": result["sanity"]["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/analyze_strong_pilot.py b/runs/fidelity-headroom/analyze_strong_pilot.py new file mode 100644 index 0000000..05cebca --- /dev/null +++ b/runs/fidelity-headroom/analyze_strong_pilot.py @@ -0,0 +1,507 @@ +#!/usr/bin/env python3 +"""Exploratory P1 audit against the strengthened simulator-aware baseline. + +P1 was already running when the strong baseline was added, so this script is +not paper-facing prospective evidence. It trains only on the historical +Phase-6 task and evaluates the exact P1 primary probes. Both nested models +receive identical Frontier predictions; engine telemetry is the sole feature +difference. +""" + +from __future__ import annotations + +import argparse +import json +import math +import subprocess +from pathlib import Path +from typing import Any + +import numpy as np + +from analyze_existing import ( + DEFAULT_REGULARIZATION, + REGULARIZATION_SENSITIVITY, + _classification_metrics, + _fit_logistic, + _mcnemar_exact_p, + _sigmoid, +) +from analyze_pilot import build_pilot_examples, campaign_gpu_accounting +from analyze_prefixes import ( + INSTRUMENTATION_FEATURES, + OUTCOME_FEATURES, + PrefixExample, + build_examples, + numeric, + policy_metrics, + sha256_file, +) +from analyze_strong_baseline import ( + SIMULATOR_FEATURES, + load_simulator_features, + simulator_row, +) + + +AITUNER_ROOT = Path(__file__).resolve().parents[2] + + +def git_capture(*arguments: str) -> str: + return subprocess.run( + ["git", "-C", str(AITUNER_ROOT), *arguments], + check=True, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + ).stdout + + +def load_pilot_simulator( + path: Path, +) -> tuple[dict[tuple[str, str], tuple[float, ...]], list[str]]: + payload = json.loads(path.read_text(encoding="utf-8")) + red_flags = [] + if payload.get("status") != "PASS": + red_flags.append("pilot_simulator_not_pass") + features: dict[tuple[str, str], tuple[float, ...]] = {} + for item in payload.get("results", []): + key = (str(item["cell"]), str(item["role"])) + if key in features: + red_flags.append(f"duplicate_pilot_simulator_{key[0]}_{key[1]}") + continue + scorer = item["scorer"] + throughput = float(scorer["throughput_requests_per_second_per_gpu"]) + pass_rate = float(scorer["slo"]["pass_rate"]) + if throughput <= 0: + red_flags.append(f"nonpositive_pilot_simulator_throughput_{key[0]}_{key[1]}") + if not 0.0 <= pass_rate <= 1.0: + red_flags.append(f"pilot_simulator_ratio_out_of_range_{key[0]}_{key[1]}") + features[key] = ( + math.log(throughput), + pass_rate, + float(bool(scorer["slo"]["feasible"])), + ) + if len(features) != 12: + red_flags.append("pilot_simulator_entries_not_12") + return features, red_flags + + +def fit_model( + examples: list[PrefixExample], + simulator: list[tuple[float, ...]], + *, + instrumentation_aware: bool, + regularization: float, +) -> dict[str, Any]: + rows = [] + for example, simulator_features in zip(examples, simulator): + values = example.outcome + simulator_features + if instrumentation_aware: + values += example.instrumentation + rows.append((1.0, *values)) + matrix = np.asarray(rows, dtype=np.float64) + labels = np.asarray([example.feasible for example in examples], dtype=np.float64) + mean = matrix[:, 1:].mean(axis=0) + standard_deviation = matrix[:, 1:].std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + standardized = matrix.copy() + standardized[:, 1:] = (standardized[:, 1:] - mean) / standard_deviation + weights = _fit_logistic(standardized, labels, regularization) + return { + "instrumentation_aware": instrumentation_aware, + "regularization": regularization, + "feature_mean": mean, + "feature_standard_deviation": standard_deviation, + "weights": weights, + } + + +def predict_model( + model: dict[str, Any], + examples: list[PrefixExample], + simulator: list[tuple[float, ...]], +) -> np.ndarray: + rows = [] + for example, simulator_features in zip(examples, simulator): + values = example.outcome + simulator_features + if model["instrumentation_aware"]: + values += example.instrumentation + rows.append((1.0, *values)) + matrix = np.asarray(rows, dtype=np.float64) + matrix[:, 1:] = ( + matrix[:, 1:] - model["feature_mean"] + ) / model["feature_standard_deviation"] + return _sigmoid(matrix @ model["weights"]) + + +def covariate_shift( + training_examples: list[PrefixExample], + training_simulator: list[tuple[float, ...]], + pilot_examples: list[PrefixExample], + pilot_simulator: list[tuple[float, ...]], + *, + instrumentation_aware: bool, +) -> dict[str, Any]: + def matrix( + examples: list[PrefixExample], simulator: list[tuple[float, ...]] + ) -> np.ndarray: + rows = [] + for example, simulator_features in zip(examples, simulator): + values = example.outcome + simulator_features + if instrumentation_aware: + values += example.instrumentation + rows.append(values) + return np.asarray(rows, dtype=np.float64) + + training = matrix(training_examples, training_simulator) + pilot = matrix(pilot_examples, pilot_simulator) + mean = training.mean(axis=0) + standard_deviation = training.std(axis=0) + standard_deviation[standard_deviation < 1e-8] = 1.0 + absolute_z = np.abs((pilot - mean) / standard_deviation) + names = [*OUTCOME_FEATURES, *SIMULATOR_FEATURES] + if instrumentation_aware: + names.extend(INSTRUMENTATION_FEATURES) + return { + "values": numeric(absolute_z.ravel().tolist()), + "count_gt_3": int(np.sum(absolute_z > 3.0)), + "count_gt_5": int(np.sum(absolute_z > 5.0)), + "total_feature_values": int(absolute_z.size), + "per_feature_max_abs_z": { + name: float(value) for name, value in zip(names, absolute_z.max(axis=0)) + }, + } + + +def comparison( + training_examples: list[PrefixExample], + training_simulator: list[tuple[float, ...]], + pilot_examples: list[PrefixExample], + pilot_simulator: list[tuple[float, ...]], + regularization: float, +) -> dict[str, Any]: + labels = np.asarray([example.feasible for example in pilot_examples], dtype=np.int64) + baseline_model = fit_model( + training_examples, + training_simulator, + instrumentation_aware=False, + regularization=regularization, + ) + instrument_model = fit_model( + training_examples, + training_simulator, + instrumentation_aware=True, + regularization=regularization, + ) + baseline_probability = predict_model( + baseline_model, pilot_examples, pilot_simulator + ) + instrument_probability = predict_model( + instrument_model, pilot_examples, pilot_simulator + ) + baseline_correct = (baseline_probability >= 0.5) == labels + instrument_correct = (instrument_probability >= 0.5) == labels + paired = { + "both_correct": int(np.sum(baseline_correct & instrument_correct)), + "sim_outcome_only_correct": int( + np.sum(baseline_correct & ~instrument_correct) + ), + "instrumentation_only_correct": int( + np.sum(~baseline_correct & instrument_correct) + ), + "both_wrong": int(np.sum(~baseline_correct & ~instrument_correct)), + } + paired["mcnemar_exact_two_sided_p"] = _mcnemar_exact_p( + paired["sim_outcome_only_correct"], paired["instrumentation_only_correct"] + ) + return { + "sim_plus_outcome": { + "classification": _classification_metrics(labels, baseline_probability), + "policy_0p95": policy_metrics( + pilot_examples, labels, baseline_probability, 0.95 + ), + "probability": baseline_probability.tolist(), + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": _classification_metrics(labels, instrument_probability), + "policy_0p95": policy_metrics( + pilot_examples, labels, instrument_probability, 0.95 + ), + "probability": instrument_probability.tolist(), + }, + "paired_correctness": paired, + } + + +def analyze( + phase6_path: Path, + phase6_raw_root: Path, + training_simulator_root: Path, + pilot_manifest_path: Path, + pilot_run_root: Path, + pilot_simulator_path: Path, + prior_state_paths: tuple[Path, ...] = (), +) -> dict[str, Any]: + phase6 = json.loads(phase6_path.read_text(encoding="utf-8")) + pilot_manifest = json.loads(pilot_manifest_path.read_text(encoding="utf-8")) + pilot_state_path = pilot_run_root / "controller-state.json" + pilot_state = json.loads(pilot_state_path.read_text(encoding="utf-8")) + gpu_accounting = campaign_gpu_accounting( + pilot_state_path, prior_state_paths + ) + training_examples = build_examples(phase6, phase6_raw_root, 5.0) + training_simulator_map, training_simulator_sha256 = load_simulator_features( + training_simulator_root + ) + training_simulator = [ + simulator_row(example, training_simulator_map) + for example in training_examples + ] + pilot_examples, pilot_details, red_flags = build_pilot_examples( + pilot_manifest, pilot_run_root, 5.0 + ) + pilot_simulator_map, simulator_red_flags = load_pilot_simulator( + pilot_simulator_path + ) + red_flags.extend(simulator_red_flags) + pilot_simulator = [] + for example, detail in zip(pilot_examples, pilot_details): + role = f"{detail['level']}1" + key = (example.cell, role) + if key not in pilot_simulator_map: + red_flags.append(f"missing_pilot_simulator_{example.cell}_{role}") + pilot_simulator.append((0.0, 0.0, 0.0)) + else: + pilot_simulator.append(pilot_simulator_map[key]) + + sensitivity = {} + if not red_flags: + for regularization in REGULARIZATION_SENSITIVITY: + sensitivity[str(regularization)] = comparison( + training_examples, + training_simulator, + pilot_examples, + pilot_simulator, + regularization, + ) + headline = sensitivity.get(str(DEFAULT_REGULARIZATION)) + labels = [example.feasible for example in pilot_examples] + simulator_pass_rates = [row[1] for row in pilot_simulator] + simulator_labels = [int(row[2]) for row in pilot_simulator] + if len(training_examples) != 37: + red_flags.append("training_examples_not_37") + if len(pilot_examples) != 12: + red_flags.append("pilot_examples_not_12") + if len(set(labels)) != 2: + red_flags.append("pilot_single_label") + if len(set(simulator_pass_rates)) <= 1: + red_flags.append("pilot_simulator_results_identical") + if pilot_state.get("status") != "complete" or int( + pilot_state.get("completed_cells", 0) + ) != 6: + red_flags.append("pilot_campaign_incomplete") + if any(detail["actual_timestamped_outcomes"] == 0 for detail in pilot_details): + red_flags.append("pilot_no_exact_request_timestamps") + all_cell_validations = all( + cell.get("validation") is not None + and all(cell["validation"]["invariants"].values()) + for cell in pilot_state.get("cells", {}).values() + ) + if not all_cell_validations: + red_flags.append("pilot_cell_validation_failed") + if not all(gpu_accounting["invariants"].values()): + red_flags.append("pilot_hard_cap_exceeded") + covariate_diagnostics = { + "sim_plus_outcome": covariate_shift( + training_examples, + training_simulator, + pilot_examples, + pilot_simulator, + instrumentation_aware=False, + ), + "sim_plus_outcome_plus_instrumentation": covariate_shift( + training_examples, + training_simulator, + pilot_examples, + pilot_simulator, + instrumentation_aware=True, + ), + } + + if headline is None: + decision = { + "strong_incremental_gate": False, + "reason": "analysis red flag prevented nested comparison", + } + else: + baseline_policy = headline["sim_plus_outcome"]["policy_0p95"] + instrument_policy = headline[ + "sim_plus_outcome_plus_instrumentation" + ]["policy_0p95"] + baseline_errors = baseline_policy["false_accept"] + baseline_policy["false_reject"] + instrument_errors = ( + instrument_policy["false_accept"] + instrument_policy["false_reject"] + ) + baseline_reduction = baseline_policy["valid_cost_reduction_fraction"] + instrument_reduction = instrument_policy["valid_cost_reduction_fraction"] + reduction_delta = ( + instrument_reduction - baseline_reduction + if baseline_reduction is not None and instrument_reduction is not None + else None + ) + per_lambda_safe_and_better = [] + for item in sensitivity.values(): + baseline = item["sim_plus_outcome"]["policy_0p95"] + instrument = item["sim_plus_outcome_plus_instrumentation"]["policy_0p95"] + base_errors = baseline["false_accept"] + baseline["false_reject"] + inst_errors = instrument["false_accept"] + instrument["false_reject"] + base_reduction = baseline["valid_cost_reduction_fraction"] + inst_reduction = instrument["valid_cost_reduction_fraction"] + per_lambda_safe_and_better.append( + inst_errors == 0 + and inst_errors <= base_errors + and base_reduction is not None + and inst_reduction is not None + and inst_reduction > base_reduction + ) + decision = { + "strong_incremental_gate": bool( + not red_flags + and instrument_errors == 0 + and instrument_errors <= baseline_errors + and reduction_delta is not None + and reduction_delta >= 0.15 + ), + "regularization_robust": all(per_lambda_safe_and_better), + "valid_cost_reduction_fraction_delta": reduction_delta, + "scope": "exploratory task; may choose P2 design but cannot establish contribution", + } + + return { + "schema": "fidelity-strong-pilot-v1", + "status": "PASS" if not red_flags else "STOP", + "scope": ( + "post-amendment exploratory P1 audit; strong model was not frozen before " + "partial P1 outcomes, so this is not prospective contribution evidence" + ), + "features": { + "shared_outcome": list(OUTCOME_FEATURES), + "shared_simulator": list(SIMULATOR_FEATURES), + "instrumentation_only": list(INSTRUMENTATION_FEATURES), + }, + "headline_regularization": DEFAULT_REGULARIZATION, + "headline": headline, + "regularization_sensitivity": sensitivity, + "simulator_only": { + "classification": _classification_metrics( + np.asarray(labels, dtype=np.int64), + np.asarray(simulator_labels, dtype=np.float64), + ) + if labels + else None, + "predicted_feasible": simulator_labels, + }, + "pilot_examples": [ + { + **detail, + "sim_completed_throughput_per_gpu": math.exp(simulator[0]), + "sim_slo_pass_rate": simulator[1], + "sim_slo_feasible": bool(simulator[2]), + } + for detail, simulator in zip(pilot_details, pilot_simulator) + ], + "covariate_shift_diagnostic": covariate_diagnostics, + "decision": decision, + "gpu": { + "primary_attempt_h20_hours": pilot_state["gpu_hours_total"], + **gpu_accounting, + }, + "analysis": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": git_capture("rev-parse", "HEAD").strip(), + "aituner_git_status_short": git_capture("status", "--short"), + }, + "provenance": { + "phase6_metrics": str(phase6_path.resolve()), + "phase6_metrics_sha256": sha256_file(phase6_path), + "phase6_raw_root": str(phase6_raw_root.resolve()), + "training_simulator_root": str(training_simulator_root.resolve()), + "training_simulator_manifest_scorer_set_sha256": training_simulator_sha256, + "pilot_manifest": str(pilot_manifest_path.resolve()), + "pilot_manifest_sha256": sha256_file(pilot_manifest_path), + "pilot_run_root": str(pilot_run_root.resolve()), + "pilot_controller_state": str(pilot_state_path.resolve()), + "pilot_controller_state_sha256": sha256_file(pilot_state_path), + "pilot_simulator": str(pilot_simulator_path.resolve()), + "pilot_simulator_sha256": sha256_file(pilot_simulator_path), + }, + "sanity": { + "red_flags": red_flags, + "training_examples": numeric([1 for _ in training_examples]), + "pilot_labels": { + **numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + }, + "pilot_simulator_pass_rate": numeric(simulator_pass_rates), + "invariants": { + "training_examples_37": len(training_examples) == 37, + "pilot_examples_12": len(pilot_examples) == 12, + "pilot_cells_6": len({example.cell for example in pilot_examples}) == 6, + "pilot_both_labels": len(set(labels)) == 2, + "simulator_ratios_bounded": all( + 0.0 <= value <= 1.0 for value in simulator_pass_rates + ), + "per_config_not_all_identical": len(set(simulator_pass_rates)) > 1, + "all_prefixes_exact_monotonic": all( + example.completion_time_source in {"exact_monotonic", "none_completed"} + for example in pilot_examples + ), + "all_cell_validations": all_cell_validations, + "gpu_cost_nonnegative_below_cap": ( + all(gpu_accounting["invariants"].values()) + ), + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--phase6-metrics", type=Path, required=True) + parser.add_argument("--phase6-raw-root", type=Path, required=True) + parser.add_argument("--training-simulator-root", type=Path, required=True) + parser.add_argument("--pilot-manifest", type=Path, required=True) + parser.add_argument("--pilot-run-root", type=Path, required=True) + parser.add_argument("--pilot-simulator", type=Path, required=True) + parser.add_argument("--prior-state", type=Path, action="append", default=[]) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + result = analyze( + args.phase6_metrics, + args.phase6_raw_root, + args.training_simulator_root, + args.pilot_manifest, + args.pilot_run_root, + args.pilot_simulator, + tuple(args.prior_state), + ) + args.output.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + print( + json.dumps( + { + "status": result["status"], + "red_flags": result["sanity"]["red_flags"], + "decision": result["decision"], + }, + sort_keys=True, + ) + ) + if result["status"] != "PASS": + raise RuntimeError(result["sanity"]["red_flags"]) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/freeze_models.py b/runs/fidelity-headroom/freeze_models.py new file mode 100644 index 0000000..e452761 --- /dev/null +++ b/runs/fidelity-headroom/freeze_models.py @@ -0,0 +1,93 @@ +#!/usr/bin/env python3 +"""Freeze the training-task prefix models before prospective GPU work.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +from analyze_prefixes import ( + DEFAULT_REGULARIZATION, + POLICY_THRESHOLDS, + build_examples, + fit_frozen_model, + sha256_file, +) + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--phase6-metrics", type=Path, required=True) + parser.add_argument("--prefix-metrics", type=Path, required=True) + parser.add_argument("--raw-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + + cutoff_s = 5.0 + threshold = 0.95 + if threshold not in POLICY_THRESHOLDS: + raise AssertionError("frozen threshold is outside audited policy thresholds") + phase6 = json.loads(args.phase6_metrics.read_text(encoding="utf-8")) + examples = build_examples(phase6, args.raw_root, cutoff_s) + payload = { + "schema": "fidelity-prefix-model-v1", + "status": "FROZEN_BEFORE_PROSPECTIVE_RUN", + "cutoff_s": cutoff_s, + "accept_probability": threshold, + "reject_probability": 1.0 - threshold, + "regularization": DEFAULT_REGULARIZATION, + "label": "same-placement 2-of-3 adjudicated anchor feasibility", + "training_split_role": "historical training only; never headline test", + "training_examples": [ + { + "cell": example.cell, + "anchor": example.anchor, + "label_feasible": bool(example.feasible), + "primary_feasible": bool(example.primary_feasible), + "completion_time_source": example.completion_time_source, + } + for example in examples + ], + "models": { + "outcome_only": fit_frozen_model( + examples, + instrumentation_aware=False, + regularization=DEFAULT_REGULARIZATION, + ), + "instrumentation_aware": fit_frozen_model( + examples, + instrumentation_aware=True, + regularization=DEFAULT_REGULARIZATION, + ), + }, + "provenance": { + "phase6_metrics": str(args.phase6_metrics.resolve()), + "phase6_metrics_sha256": sha256_file(args.phase6_metrics), + "prefix_metrics": str(args.prefix_metrics.resolve()), + "prefix_metrics_sha256": sha256_file(args.prefix_metrics), + "raw_root": str(args.raw_root.resolve()), + }, + "sanity": { + "n": len(examples), + "positive": sum(example.feasible for example in examples), + "negative": sum(not example.feasible for example in examples), + "cells": len({example.cell for example in examples}), + "invariants": { + "n_37": len(examples) == 37, + "cells_12": len({example.cell for example in examples}) == 12, + "both_labels": len({example.feasible for example in examples}) == 2, + "cutoff_5s": cutoff_s == 5.0, + "threshold_0.95": threshold == 0.95, + }, + }, + } + if not all(payload["sanity"]["invariants"].values()): + raise RuntimeError(f"model freeze invariants failed: {payload['sanity']}") + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps({"status": payload["status"], "output": str(args.output)})) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/frozen-models.json b/runs/fidelity-headroom/frozen-models.json new file mode 100644 index 0000000..46fa17c --- /dev/null +++ b/runs/fidelity-headroom/frozen-models.json @@ -0,0 +1,515 @@ +{ + "accept_probability": 0.95, + "cutoff_s": 5.0, + "label": "same-placement 2-of-3 adjudicated anchor feasibility", + "models": { + "instrumentation_aware": { + "feature_mean": [ + 0.8984976998643891, + 0.8378378378378378, + 4.324324324324325, + 0.07552086023066117, + 0.6758807403968693, + 0.9459459459459459, + 0.0, + 0.3241192596031305, + 0.04468545442770093, + 0.025590516908533558, + 0.23873649352596996, + 0.1943716628122394, + 0.4321792125178198, + 112.70270270270272, + 0.21087752102856197, + 0.918918918918919, + 0.055470351361483664, + 4.904239530899751, + 10.162162162162161, + 4.822982150502539, + 10.135135135135135, + 0.43557131397798415, + 0.031387890158936414, + 0.05804311436894179, + 0.03298678556030958, + 0.030437119300455177, + 0.9503065396705037, + 0.07127076398319926, + 0.6234198543231205, + 0.0 + ], + "feature_names": [ + "log_offered_rate_per_gpu", + "log2_tp", + "log2_max_num_seqs", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit", + "model_steps_per_second", + "waiting_mean", + "waiting_max", + "waiting_nonzero_share", + "running_mean", + "running_max", + "decode_batch_mean", + "decode_batch_max", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "preemptions" + ], + "feature_standard_deviation": [ + 0.2953332526155246, + 0.8546696378833459, + 1.1402715194448103, + 0.006588255148989237, + 0.2751217635728275, + 0.22612433149569594, + 1.0, + 0.27512176357282747, + 0.048292427420964075, + 0.02574874589991541, + 0.1635381690436309, + 0.14098674719611365, + 0.02516276437103069, + 61.39272994412853, + 0.7234949448561444, + 2.198013579605131, + 0.18326586413988316, + 2.4542471212960844, + 6.08726412391018, + 2.4074006634033043, + 6.067913672185017, + 0.12556414020947543, + 0.03256962310836033, + 0.054049610008010444, + 0.048321850100969746, + 0.04298231458641556, + 0.041906068064246155, + 0.08212268757576466, + 0.4089385238422411, + 1.0 + ], + "instrumentation_aware": true, + "regularization": 1.0, + "training_classification": { + "accuracy": 0.972972972972973, + "balanced_accuracy": 0.9444444444444444, + "brier": 0.02820726479488704, + "confusion": { + "false_negative": 0, + "false_positive": 1, + "true_negative": 8, + "true_positive": 28 + }, + "log_loss": 0.11247563885308659 + }, + "weights_with_intercept_first": [ + 2.109507425802979, + -0.8372240489271802, + -0.2476229678897366, + 0.18172257646801393, + -0.07076358054975332, + 0.3035586906752765, + 0.08500005412355496, + -7.754818242684634e-26, + -0.3035586906752766, + 0.4014234393196892, + 0.513218716194957, + -0.35161457106287, + 0.10558147889556725, + 0.5674345291616134, + 0.15895995157114373, + -0.4274260624362057, + -0.048791959001756195, + -0.37221380985270663, + -0.35527537277290255, + 0.20582736797173468, + -0.35837576944545413, + 0.2342062515631318, + 0.45071068059490843, + 0.3326948315186803, + 0.2698892549960913, + 0.017868065865726347, + -0.1540209080477302, + 0.3412427440368233, + 0.5831011876762794, + -0.583920360300169, + 0.0 + ] + }, + "outcome_only": { + "feature_mean": [ + 0.8984976998643891, + 0.8378378378378378, + 4.324324324324325, + 0.07552086023066117, + 0.6758807403968693, + 0.9459459459459459, + 0.0, + 0.3241192596031305, + 0.04468545442770093, + 0.025590516908533558, + 0.23873649352596996, + 0.1943716628122394, + 0.4321792125178198 + ], + "feature_names": [ + "log_offered_rate_per_gpu", + "log2_tp", + "log2_max_num_seqs", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit" + ], + "feature_standard_deviation": [ + 0.2953332526155246, + 0.8546696378833459, + 1.1402715194448103, + 0.006588255148989237, + 0.2751217635728275, + 0.22612433149569594, + 1.0, + 0.27512176357282747, + 0.048292427420964075, + 0.02574874589991541, + 0.1635381690436309, + 0.14098674719611365, + 0.02516276437103069 + ], + "instrumentation_aware": false, + "regularization": 1.0, + "training_classification": { + "accuracy": 0.9459459459459459, + "balanced_accuracy": 0.8888888888888888, + "brier": 0.051887373873176545, + "confusion": { + "false_negative": 0, + "false_positive": 2, + "true_negative": 7, + "true_positive": 28 + }, + "log_loss": 0.184988719119571 + }, + "weights_with_intercept_first": [ + 1.8996338126233983, + -1.1536861934230125, + -0.3806404559018098, + 0.5901136731733696, + 0.022432085012851908, + 0.5805554730881304, + 0.25786307099613026, + -8.077935669463161e-27, + -0.5805554730881304, + -0.15413292402348447, + 0.0986842306063204, + -0.5181573573074624, + 0.06283513013708956, + 0.911619634884147 + ] + } + }, + "provenance": { + "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", + "phase6_metrics_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", + "prefix_metrics": "/home/gahow/phd/aituner/runs/fidelity-headroom/prefix-metrics.json", + "prefix_metrics_sha256": "cda821bcde1ae8427507aa4f03a1c116ccc7f7b8b717f73ca587bee3670a0340", + "raw_root": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/solo-authoritative/cells" + }, + "regularization": 1.0, + "reject_probability": 0.050000000000000044, + "sanity": { + "cells": 12, + "invariants": { + "both_labels": true, + "cells_12": true, + "cutoff_5s": true, + "n_37": true, + "threshold_0.95": true + }, + "n": 37, + "negative": 9, + "positive": 28 + }, + "schema": "fidelity-prefix-model-v1", + "status": "FROZEN_BEFORE_PROSPECTIVE_RUN", + "training_examples": [ + { + "anchor": 0.24609375, + "cell": "tp1_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.25, + "cell": "tp1_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.5, + "cell": "tp1_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.2421875, + "cell": "tp1_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.24609375, + "cell": "tp1_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.25, + "cell": "tp1_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.5, + "cell": "tp1_mns32", + "completion_time_source": "none_completed", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.2421875, + "cell": "tp1_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.24609375, + "cell": "tp1_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.25, + "cell": "tp1_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.5, + "cell": "tp1_mns64", + "completion_time_source": "none_completed", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.2265625, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.23046875, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.234375, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.25, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.5, + "cell": "tp1_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.4921875, + "cell": "tp2_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.5, + "cell": "tp2_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.75, + "cell": "tp2_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.75390625, + "cell": "tp2_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.5, + "cell": "tp2_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.4921875, + "cell": "tp2_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.033182214016, + "cell": "tp4_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": false + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns16", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": false + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns32", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": false + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns64", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.016055910008, + "cell": "tp4_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.016591107009, + "cell": "tp4_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.017126304009, + "cell": "tp4_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": true, + "primary_feasible": true + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "completion_time_source": "reconstructed_from_latency", + "label_feasible": false, + "primary_feasible": false + } + ], + "training_split_role": "historical training only; never headline test" +} diff --git a/runs/fidelity-headroom/metrics.json b/runs/fidelity-headroom/metrics.json new file mode 100644 index 0000000..24f0550 --- /dev/null +++ b/runs/fidelity-headroom/metrics.json @@ -0,0 +1,1142 @@ +{ + "decision": { + "current_surface_can_show_selection_contribution": false, + "prospective_target": "Test whether internal features from a short, shared real probe reduce the number or duration of full frontier evaluations relative to an outcome-only model given the same probe.", + "reason": "The strongest frozen-calibrated SLO reading reaches zero real regret after real evaluation of its first two-cell tie bucket. A method that requires one calibration probe and one final verification cannot use this single task to demonstrate fewer real cell evaluations." + }, + "next_anchor_prediction": { + "estimand": "next-anchor feasibility from the full current-anchor summary", + "external_features": [ + "log_current_rate_per_gpu", + "log_next_over_current_rate", + "log2_tp", + "log2_mns", + "current_pass_rate", + "ttft_max_over_6s", + "tpot_max_over_50ms", + "exact_output_fraction", + "early_stopped" + ], + "headline": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.125 + ], + "point": 0.040000000000000036 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.019846177751851126, + 0.02199527924137797 + ], + "point": -0.0019563088303256454 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "accuracy": 0.88, + "balanced_accuracy": 0.8576388888888888, + "brier": 0.1031574576574755, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 7, + "true_positive": 15 + }, + "log_loss": 0.327997342465158 + }, + "outcome_only": { + "accuracy": 0.84, + "balanced_accuracy": 0.8020833333333333, + "brier": 0.10511376648780114, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 15 + }, + "log_loss": 0.3385851118121996 + }, + "paired_correctness": { + "both_correct": 21, + "both_wrong": 3, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 0 + } + }, + "headline_regularization": 1.0, + "instrumentation_features": [ + "waiting_mean", + "waiting_max", + "decode_batch_mean", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "model_steps_per_second" + ], + "label_sanity": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 25, + "negative": 9, + "positive": 16 + }, + "model": "L2 logistic regression with train-fold standardization", + "regularization_sensitivity": { + "0.1": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.13636363636363646 + ], + "point": 0.040000000000000036 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.029611344558941783, + 0.030283692179295475 + ], + "point": -0.0013643768871213907 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "accuracy": 0.92, + "balanced_accuracy": 0.9131944444444444, + "brier": 0.08887235975611406, + "confusion": { + "false_negative": 1, + "false_positive": 1, + "true_negative": 8, + "true_positive": 15 + }, + "log_loss": 0.3015353076332512 + }, + "outcome_only": { + "accuracy": 0.88, + "balanced_accuracy": 0.8819444444444444, + "brier": 0.09023673664323545, + "confusion": { + "false_negative": 2, + "false_positive": 1, + "true_negative": 8, + "true_positive": 14 + }, + "log_loss": 0.2927458553148107 + }, + "paired_correctness": { + "both_correct": 22, + "both_wrong": 2, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 0 + } + }, + "1.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.125 + ], + "point": 0.040000000000000036 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.019846177751851126, + 0.02199527924137797 + ], + "point": -0.0019563088303256454 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "accuracy": 0.88, + "balanced_accuracy": 0.8576388888888888, + "brier": 0.1031574576574755, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 7, + "true_positive": 15 + }, + "log_loss": 0.327997342465158 + }, + "outcome_only": { + "accuracy": 0.84, + "balanced_accuracy": 0.8020833333333333, + "brier": 0.10511376648780114, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 15 + }, + "log_loss": 0.3385851118121996 + }, + "paired_correctness": { + "both_correct": 21, + "both_wrong": 3, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 0 + } + }, + "10.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.13636363636363646, + 0.0 + ], + "point": -0.039999999999999925 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.04125619566309649, + 0.012885694983237113 + ], + "point": -0.013003224609162978 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "accuracy": 0.8, + "balanced_accuracy": 0.7465277777777778, + "brier": 0.16045670460468575, + "confusion": { + "false_negative": 1, + "false_positive": 4, + "true_negative": 5, + "true_positive": 15 + }, + "log_loss": 0.4963310825527336 + }, + "outcome_only": { + "accuracy": 0.84, + "balanced_accuracy": 0.7777777777777778, + "brier": 0.17345992921384873, + "confusion": { + "false_negative": 0, + "false_positive": 4, + "true_negative": 5, + "true_positive": 16 + }, + "log_loss": 0.5254987874287221 + }, + "paired_correctness": { + "both_correct": 20, + "both_wrong": 4, + "instrumentation_only_correct": 0, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 1 + } + } + }, + "sensitivity_summary": { + "accuracy_delta_min_max": [ + -0.039999999999999925, + 0.040000000000000036 + ], + "brier_delta_min_max": [ + -0.013003224609162978, + -0.0013643768871213907 + ], + "incremental_signal_verdict": "NEEDS_PROSPECTIVE_EVIDENCE" + }, + "split": "leave-one-cell-out", + "status": "RETROSPECTIVE_DIAGNOSTIC_ONLY" + }, + "provenance": { + "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", + "phase6_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", + "simfid_metrics": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json", + "simfid_sha256": "55edb37d5692e979ab6f6dc6c65913a9db0aa0a836c350e4c05d9c38eee78206" + }, + "sanity": { + "invariants": { + "probabilities_bounded": true, + "same_cells_all_readings": true, + "scores_nonnegative": true, + "transition_features_finite": true + }, + "real_scores": { + "distinct_n": 8, + "max": 3.283333333333333, + "min": 1.2833333333333334, + "n": 12 + }, + "red_flags": [], + "simulator_readings": 4, + "transition_cells": 12, + "transitions": 25 + }, + "schema": "fidelity-headroom-v1", + "scope": "retrospective single-workload premise audit; not prospective contribution evidence", + "status": "PASS", + "topk_headroom": { + "frozen-calibrated/SLO-gated": { + "minimum_k": { + "five_percent": { + "expanded_k": 2, + "nominal_k": 1 + }, + "one_percent": { + "expanded_k": 2, + "nominal_k": 1 + }, + "zero": { + "expanded_k": 2, + "nominal_k": 1 + } + }, + "points": [ + { + "candidates": [ + "tp2_mns32", + "tp2_mns64" + ], + "expanded_k": 2, + "nominal_k": 1, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64" + ], + "expanded_k": 2, + "nominal_k": 2, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 5, + "nominal_k": 3, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 5, + "nominal_k": 4, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 5, + "nominal_k": 5, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "expanded_k": 8, + "nominal_k": 6, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "expanded_k": 8, + "nominal_k": 7, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "expanded_k": 8, + "nominal_k": 8, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns16" + ], + "expanded_k": 9, + "nominal_k": 9, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns16", + "tp2_mns8" + ], + "expanded_k": 10, + "nominal_k": 10, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns16", + "tp2_mns8", + "tp1_mns8" + ], + "expanded_k": 11, + "nominal_k": 11, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp2_mns32", + "tp2_mns64", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns16", + "tp2_mns8", + "tp1_mns8", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 12, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + } + ], + "real_best": 3.283333333333333 + }, + "frozen-calibrated/throughput-proxy": { + "minimum_k": { + "five_percent": { + "expanded_k": 4, + "nominal_k": 3 + }, + "one_percent": { + "expanded_k": 4, + "nominal_k": 3 + }, + "zero": { + "expanded_k": 4, + "nominal_k": 3 + } + }, + "points": [ + { + "candidates": [ + "tp1_mns64" + ], + "expanded_k": 1, + "nominal_k": 1, + "real_regret": 0.30456852791878175, + "selected_cell_after_real_final": "tp1_mns64", + "selected_real_score": 2.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32" + ], + "expanded_k": 2, + "nominal_k": 2, + "real_regret": 0.30456852791878175, + "selected_cell_after_real_final": "tp1_mns64", + "selected_real_score": 2.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64" + ], + "expanded_k": 4, + "nominal_k": 3, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64" + ], + "expanded_k": 4, + "nominal_k": 4, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16" + ], + "expanded_k": 5, + "nominal_k": 5, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 7, + "nominal_k": 6, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 7, + "nominal_k": 7, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns16" + ], + "expanded_k": 8, + "nominal_k": 8, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns16", + "tp2_mns16" + ], + "expanded_k": 9, + "nominal_k": 9, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns16", + "tp2_mns16", + "tp1_mns8" + ], + "expanded_k": 10, + "nominal_k": 10, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns16", + "tp2_mns16", + "tp1_mns8", + "tp2_mns8" + ], + "expanded_k": 11, + "nominal_k": 11, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64", + "tp1_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns16", + "tp2_mns16", + "tp1_mns8", + "tp2_mns8", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 12, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + } + ], + "real_best": 3.283333333333333 + }, + "uncalibrated/SLO-gated": { + "minimum_k": { + "five_percent": { + "expanded_k": 12, + "nominal_k": 8 + }, + "one_percent": { + "expanded_k": 12, + "nominal_k": 8 + }, + "zero": { + "expanded_k": 12, + "nominal_k": 8 + } + }, + "points": [ + { + "candidates": [ + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 2, + "nominal_k": 1, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64" + ], + "expanded_k": 2, + "nominal_k": 2, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16" + ], + "expanded_k": 3, + "nominal_k": 3, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8" + ], + "expanded_k": 7, + "nominal_k": 4, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8" + ], + "expanded_k": 7, + "nominal_k": 5, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8" + ], + "expanded_k": 7, + "nominal_k": 6, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8" + ], + "expanded_k": 7, + "nominal_k": 7, + "real_regret": 0.25634517766497456, + "selected_cell_after_real_final": "tp4_mns64", + "selected_real_score": 2.441666666666667 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8", + "tp2_mns32", + "tp2_mns64", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 8, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8", + "tp2_mns32", + "tp2_mns64", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 9, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8", + "tp2_mns32", + "tp2_mns64", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 10, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8", + "tp2_mns32", + "tp2_mns64", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 11, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp1_mns8", + "tp2_mns32", + "tp2_mns64", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 12, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + } + ], + "real_best": 3.283333333333333 + }, + "uncalibrated/throughput-proxy": { + "minimum_k": { + "five_percent": { + "expanded_k": 3, + "nominal_k": 3 + }, + "one_percent": { + "expanded_k": 3, + "nominal_k": 3 + }, + "zero": { + "expanded_k": 5, + "nominal_k": 5 + } + }, + "points": [ + { + "candidates": [ + "tp1_mns64" + ], + "expanded_k": 1, + "nominal_k": 1, + "real_regret": 0.30456852791878175, + "selected_cell_after_real_final": "tp1_mns64", + "selected_real_score": 2.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32" + ], + "expanded_k": 2, + "nominal_k": 2, + "real_regret": 0.30456852791878175, + "selected_cell_after_real_final": "tp1_mns64", + "selected_real_score": 2.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64" + ], + "expanded_k": 3, + "nominal_k": 3, + "real_regret": 0.0076142131979695105, + "selected_cell_after_real_final": "tp2_mns64", + "selected_real_score": 3.2583333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16" + ], + "expanded_k": 4, + "nominal_k": 4, + "real_regret": 0.0076142131979695105, + "selected_cell_after_real_final": "tp2_mns64", + "selected_real_score": 3.2583333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32" + ], + "expanded_k": 5, + "nominal_k": 5, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64" + ], + "expanded_k": 6, + "nominal_k": 6, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8" + ], + "expanded_k": 7, + "nominal_k": 7, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8", + "tp4_mns32" + ], + "expanded_k": 8, + "nominal_k": 8, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8", + "tp4_mns32", + "tp2_mns16" + ], + "expanded_k": 9, + "nominal_k": 9, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8", + "tp4_mns32", + "tp2_mns16", + "tp2_mns8" + ], + "expanded_k": 10, + "nominal_k": 10, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8", + "tp4_mns32", + "tp2_mns16", + "tp2_mns8", + "tp4_mns16" + ], + "expanded_k": 11, + "nominal_k": 11, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + }, + { + "candidates": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns64", + "tp1_mns16", + "tp2_mns32", + "tp4_mns64", + "tp1_mns8", + "tp4_mns32", + "tp2_mns16", + "tp2_mns8", + "tp4_mns16", + "tp4_mns8" + ], + "expanded_k": 12, + "nominal_k": 12, + "real_regret": 0.0, + "selected_cell_after_real_final": "tp2_mns32", + "selected_real_score": 3.283333333333333 + } + ], + "real_best": 3.283333333333333 + } + } +} diff --git a/runs/fidelity-headroom/pilot_controller.py b/runs/fidelity-headroom/pilot_controller.py new file mode 100644 index 0000000..6ffb4f1 --- /dev/null +++ b/runs/fidelity-headroom/pilot_controller.py @@ -0,0 +1,461 @@ +#!/usr/bin/env python3 +"""Serialized dash0 controller for the exact-timestamp prefix pilot.""" + +from __future__ import annotations + +import argparse +import json +import os +import shlex +import subprocess +import sys +import time +from pathlib import Path +from typing import Any + + +HERE = Path(__file__).resolve().parent +PHASE6 = HERE.parent / "opprof-phase6" +sys.path.insert(0, str(PHASE6)) + +import opprof_phase6_controller as base # noqa: E402 + + +ORDER = ( + "tp1_mns8", + "tp1_mns64", + "tp2_mns8", + "tp2_mns64", + "tp4_mns16", + "tp4_mns64", +) +CELL_ESTIMATE_H20_HOURS = {1: 0.20, 2: 0.40, 4: 0.80} +SAFETY_H20_HOURS = 0.20 + + +def atomic_json(path: Path, payload: Any) -> None: + base.atomic_json(path, payload) + + +def wait_all_idle(timeout_s: float = 30.0) -> None: + deadline = time.monotonic() + timeout_s + last_error: Exception | None = None + while time.monotonic() < deadline: + try: + base.assert_all_idle() + return + except RuntimeError as error: + last_error = error + time.sleep(1.0) + raise last_error or RuntimeError("GPU idle timeout") + + +def configure_base(args: argparse.Namespace, manifest: dict[str, Any]) -> None: + base.WORKDIR = args.run_root.parent + base.RUN_ROOT = args.run_root + base.STATE = args.run_root / "controller-state.json" + base.SOURCE = args.vllm_source + base.VENV = args.venv + base.AITUNER = args.aituner_root + base.MODEL = args.model + base.CLIENT = args.client + base.GPU_LIMIT = float(manifest["execution"]["hard_cap_h20_hours"]) + base.MARKER = "fidelity-prefix-pilot-20260714" + base.CELLS = { + cell: {"tp": int(config["tp"]), "mns": int(config["mns"])} + for cell, config in manifest["cells"].items() + } + + +def load_state(path: Path, hard_cap: float) -> dict[str, Any]: + if path.exists(): + return json.loads(path.read_text(encoding="utf-8")) + return { + "schema": "fidelity-prefix-pilot-state-v1", + "status": "initialized", + "hard_cap_h20_hours": hard_cap, + "gpu_hours_total": 0.0, + "completed_cells": 0, + "cells": {}, + "failures": [], + "started_at": time.time(), + } + + +def save_state(path: Path, state: dict[str, Any]) -> None: + atomic_json(path, state) + + +def append_echo(run_root: Path, line: str) -> None: + run_root.mkdir(parents=True, exist_ok=True) + with (run_root / "launch-echo.log").open("a", encoding="utf-8") as target: + target.write(line + "\n") + print(line, flush=True) + + +def remaining_projection(manifest: dict[str, Any], index: int) -> float: + return sum( + CELL_ESTIMATE_H20_HOURS[int(manifest["cells"][cell]["tp"])] + for cell in ORDER[index:] + ) + SAFETY_H20_HOURS + + +def start_server( + *, + cell: str, + index: int, + run_root: Path, +) -> dict[str, Any]: + config = base.CELLS[cell] + gpus = tuple(range(int(config["tp"]))) + cell_root = run_root / "cells" / cell + cell_root.mkdir(parents=True, exist_ok=True) + port = 8900 + index + command = base.server_command(cell, gpus, port) + with (cell_root / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"SERVER {shlex.join(command)}\n") + server_log = (cell_root / "server.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update( + { + "CUDA_VISIBLE_DEVICES": ",".join(map(str, gpus)), + "VLLM_OPPROF_DIR": str(cell_root / "opprof"), + "OPPROF_PHASE6_MARKER": base.MARKER, + "AITUNER_ROOT": str(base.AITUNER), + "HF_HUB_OFFLINE": "1", + "TRANSFORMERS_OFFLINE": "1", + "PYTHONUNBUFFERED": "1", + } + ) + server = subprocess.Popen( + command, + cwd=base.SOURCE, + env=environment, + stdout=server_log, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + base.OWNED_PGIDS.add(server.pid) + return { + "cell": cell, + "gpus": gpus, + "port": port, + "dir": cell_root, + "server": server, + "server_handle": server_log, + "spawned_at": time.time(), + "results": [], + } + + +def selection_for( + manifest: dict[str, Any], cell: str, role: str +) -> tuple[str, dict[str, Any]]: + level = "low" if role == "burnin" or role.startswith("low") else "high" + return level, manifest["cells"][cell]["targets"][level]["selections"][role] + + +def client_command( + entry: dict[str, Any], + *, + role: str, + selection: dict[str, Any], + output: Path, + warmup: bool, +) -> list[str]: + config = base.CELLS[entry["cell"]] + return [ + "taskset", + "-c", + base.cpu_mask(entry["gpus"]), + str(base.VENV / "bin/python"), + str(base.CLIENT), + "warmup" if warmup else "run-anchor", + "--study", + str(selection["study"]), + "--cell", + entry["cell"], + "--anchor", + str(selection["anchor"]), + "--tp", + str(config["tp"]), + "--mns", + str(config["mns"]), + "--base-url", + f"http://127.0.0.1:{entry['port']}", + "--result-dir", + str(output), + ] + + +def run_client( + *, + entry: dict[str, Any], + role: str, + selection: dict[str, Any], + output: Path, + state: dict[str, Any], + warmup: bool = False, +) -> dict[str, Any]: + command = client_command( + entry, role=role, selection=selection, output=output, warmup=warmup + ) + with (entry["dir"] / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"CLIENT role={role} {shlex.join(command)}\n") + handle = (output.parent / f"{output.name}.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update({"AITUNER_ROOT": str(base.AITUNER), "PYTHONUNBUFFERED": "1"}) + process = subprocess.Popen( + command, + cwd=base.WORKDIR, + env=environment, + stdout=handle, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + deadline = time.monotonic() + 180.0 + try: + while process.poll() is None: + if time.monotonic() > deadline: + process.terminate() + raise TimeoutError(f"client timeout: {entry['cell']} {role}") + if entry["server"].poll() is not None: + raise RuntimeError(f"server exited during {entry['cell']} {role}") + base.assert_no_other_compute() + if state["gpu_hours_total"] + base.live_gpu_hours([entry]) >= base.GPU_LIMIT: + process.terminate() + raise RuntimeError("pilot H20-hour hard cap reached") + time.sleep(1.0) + finally: + handle.close() + if process.returncode: + raise RuntimeError( + f"client failed: cell={entry['cell']} role={role} rc={process.returncode}" + ) + result = json.loads((output / "result.json").read_text(encoding="utf-8")) + validate_result_selection( + result=result, + selection=selection, + cell=entry["cell"], + role=role, + warmup=warmup, + ) + entry["results"].append( + {"anchor": float(selection["anchor"]), "dir": str(output), "kind": result["kind"]} + ) + return result + + +def validate_result_selection( + *, + result: dict[str, Any], + selection: dict[str, Any], + cell: str, + role: str, + warmup: bool, +) -> None: + if warmup: + if result["kind"] != "warmup" or int(result["selection"]["count"]) != 16: + raise RuntimeError(f"invalid warmup selection: {cell} {role}") + for key in ("warmup_16", "warmup_exact_16", "warmup_long"): + if not result["invariants"].get(key, False): + raise RuntimeError(f"warmup invariant {key} failed: {cell} {role}") + return + + if int(result["selection"]["count"]) != int(selection["selected_count"]): + raise RuntimeError(f"selection count mismatch: {cell} {role}") + for key in ( + "request_id_order_sha256", + "arrival_order_sha256", + "raw_length_order_sha256", + ): + manifest_key = ( + "input_length_order_sha256" if key == "raw_length_order_sha256" else key + ) + if result["selection"][key] != selection[manifest_key]: + raise RuntimeError(f"selection hash mismatch {key}: {cell} {role}") + + +def execute_cell( + *, + index: int, + cell: str, + manifest: dict[str, Any], + run_root: Path, + state_path: Path, + state: dict[str, Any], +) -> None: + if state["cells"].get(cell, {}).get("status") == "complete": + return + projection = remaining_projection(manifest, index) + if state["gpu_hours_total"] + projection > base.GPU_LIMIT: + state["status"] = "budget_projection_stop" + state["budget_stop"] = { + "before_cell": cell, + "spent_h20_hours": state["gpu_hours_total"], + "remaining_projection_h20_hours": projection, + "hard_cap_h20_hours": base.GPU_LIMIT, + } + save_state(state_path, state) + raise RuntimeError(f"projected pilot cost exceeds hard cap before {cell}") + + config = manifest["cells"][cell] + echo = ( + f"PILOT_CELL_ECHO cell={cell} tp={config['tp']} mns={config['mns']} " + f"gpus=0-{int(config['tp']) - 1} workload={manifest['source']['window_id']} " + f"roles=burnin+low1/high1/low2/high2/low3/high3 " + f"spent_h20h={state['gpu_hours_total']:.6f} " + f"remaining_projection_h20h={projection:.3f} cap_h20h={base.GPU_LIMIT:.1f} " + f"manifest={run_root / 'pilot-manifest.json'}" + ) + append_echo(run_root, echo) + wait_all_idle() + cell_state = { + "status": "starting", + "tp": int(config["tp"]), + "mns": int(config["mns"]), + "started_at": time.time(), + "runs": [], + } + state["status"] = "running" + state["cells"][cell] = cell_state + save_state(state_path, state) + entry = start_server(cell=cell, index=index, run_root=run_root) + failure: Exception | None = None + try: + base.wait_ready(entry) + _level, burnin = selection_for(manifest, cell, "burnin") + cell_state["status"] = "warmup" + save_state(state_path, state) + warmup = run_client( + entry=entry, + role="burnin", + selection=burnin, + output=entry["dir"] / "warmup", + state=state, + warmup=True, + ) + cell_state["warmup"] = { + "exact_output_count": warmup["exact_output_count"], + "long_gt4096": warmup["selection"]["long_gt4096"], + } + cell_state["status"] = "burnin" + save_state(state_path, state) + burnin_result = run_client( + entry=entry, + role="burnin", + selection=burnin, + output=entry["dir"] / "burnin", + state=state, + ) + cell_state["burnin"] = { + "pass_rate": burnin_result["pass_rate"], + "feasible": burnin_result["feasible"], + } + role_order = manifest["execution"][ + "even_cell_order" if index % 2 == 0 else "odd_cell_order" + ] + cell_state["status"] = "measured" + cell_state["role_order"] = role_order + save_state(state_path, state) + for role in role_order: + level, selection = selection_for(manifest, cell, role) + result = run_client( + entry=entry, + role=role, + selection=selection, + output=entry["dir"] / f"{level}-rep{role[-1]}", + state=state, + ) + cell_state["runs"].append( + { + "role": role, + "level": level, + "anchor": selection["anchor"], + "selected_count": selection["selected_count"], + "pass_rate": result["pass_rate"], + "feasible": result["feasible"], + "elapsed_s": result["interval"]["elapsed_s"], + } + ) + save_state(state_path, state) + cell_state["status"] = "stopping" + save_state(state_path, state) + except Exception as error: # noqa: BLE001 + failure = error + finally: + try: + base.stop_entry(entry) + except Exception as error: # noqa: BLE001 + failure = failure or error + time.sleep(2.0) + try: + wait_all_idle() + except Exception as error: # noqa: BLE001 + failure = failure or error + + cell_hours = base.live_gpu_hours([entry]) + state["gpu_hours_total"] += cell_hours + cell_state["gpu_hours"] = cell_hours + if failure is not None: + cell_state["status"] = "failed" + cell_state["failure"] = repr(failure) + state["status"] = "failed" + state["failures"].append({"cell": cell, "failure": repr(failure)}) + save_state(state_path, state) + raise failure + validation = base.validate_cell(entry) + cell_state["validation"] = validation + cell_state["status"] = "complete" + cell_state["completed_at"] = time.time() + state["completed_cells"] += 1 + save_state(state_path, state) + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--manifest", type=Path, required=True) + result.add_argument("--run-root", type=Path, required=True) + result.add_argument("--aituner-root", type=Path, required=True) + result.add_argument("--vllm-source", type=Path, required=True) + result.add_argument("--venv", type=Path, required=True) + result.add_argument("--model", type=Path, required=True) + result.add_argument("--client", type=Path, required=True) + return result + + +def main() -> None: + args = parser().parse_args() + manifest = json.loads(args.manifest.read_text(encoding="utf-8")) + if manifest["status"] != "PASS": + raise RuntimeError("pilot manifest did not pass preflight") + args.run_root.mkdir(parents=True, exist_ok=True) + copied_manifest = args.run_root / "pilot-manifest.json" + if not copied_manifest.exists(): + atomic_json(copied_manifest, manifest) + configure_base(args, manifest) + state_path = args.run_root / "controller-state.json" + state = load_state(state_path, base.GPU_LIMIT) + state["status"] = "running" + save_state(state_path, state) + for index, cell in enumerate(ORDER): + execute_cell( + index=index, + cell=cell, + manifest=manifest, + run_root=args.run_root, + state_path=state_path, + state=state, + ) + state["status"] = "complete" + state["completed_at"] = time.time() + save_state(state_path, state) + print(json.dumps({ + "status": state["status"], + "completed_cells": state["completed_cells"], + "gpu_hours_total": state["gpu_hours_total"], + }, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/prefix-metrics.json b/runs/fidelity-headroom/prefix-metrics.json new file mode 100644 index 0000000..eeb6210 --- /dev/null +++ b/runs/fidelity-headroom/prefix-metrics.json @@ -0,0 +1,3696 @@ +{ + "cutoffs": { + "10": { + "cells": 12, + "completion_time_sources": { + "reconstructed_from_latency": 37 + }, + "examples": 37, + "headline": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.1333333333333333, + 0.05555555555555547 + ], + "point": -0.027027027027027084 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.047874993358699436, + 0.04753434529263568 + ], + "point": -0.0026830647286030934 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8531746031746033, + "brier": 0.07268572313801988, + "confusion": { + "false_negative": 2, + "false_positive": 2, + "true_negative": 7, + "true_positive": 26 + }, + "log_loss": 0.22136936001258795 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7938369600363888, + "decision_coverage": 0.918918918918919, + "early_accept": 27, + "early_reject": 7, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8623652659223954, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.11022926282564288, + "remaining_h20_hours": 0.018465968481111112 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.025806101009444443, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.8196430610458332, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.7340601279138887, + "decision_coverage": 0.7837837837837838, + "early_accept": 24, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7340601279138887, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6879924922438714, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 15, + "correctly_saved_h20_hours": 0.620235621837222, + "decision_coverage": 0.5945945945945946, + "early_accept": 17, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.620235621837222, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5813113054633519, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.918918918918919, + "balanced_accuracy": 0.8333333333333333, + "brier": 0.07536878786662297, + "confusion": { + "false_negative": 0, + "false_positive": 3, + "true_negative": 6, + "true_positive": 28 + }, + "log_loss": 0.23513650573912553 + }, + "policies": [ + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.7625537088033332, + "decision_coverage": 0.8648648648648649, + "early_accept": 27, + "early_reject": 5, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8486897200744762, + "remaining_h20_hours": 0.007340132528333333 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.8216385310591849, + "remaining_h20_hours": 0.01680367415111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.024143806679444442, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7866975154827778, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 11, + "correctly_saved_h20_hours": 0.7042463331286111, + "decision_coverage": 0.7027027027027027, + "early_accept": 22, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7042463331286111, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6600497308847137, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 19, + "correctly_saved_h20_hours": 0.5876473660130556, + "decision_coverage": 0.4864864864864865, + "early_accept": 15, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.5876473660130556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5507682007641966, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 32, + "both_wrong": 2, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 2 + } + }, + "headline_regularization": 1.0, + "label_sanity": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 37, + "negative": 9, + "positive": 28, + "primary_adjudicated_disagreements": 3 + }, + "regularization_sensitivity": { + "0.1": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.08571428571428574, + 0.07317073170731703 + ], + "point": 0.0 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.01721313553473963, + 0.0846932121032576 + ], + "point": 0.019153359424323196 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.918918918918919, + "balanced_accuracy": 0.871031746031746, + "brier": 0.06720185291989068, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 7, + "true_positive": 27 + }, + "log_loss": 0.22428253113260074 + }, + "policies": [ + { + "abstain_continue_full": 2, + "correctly_saved_h20_hours": 0.802005166048611, + "decision_coverage": 0.9459459459459459, + "early_accept": 27, + "early_reject": 8, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.905553117931728, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.02745974548325657, + "remaining_h20_hours": 0.018465968481111112 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.025806101009444443, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.8278112670580555, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.7653433791469442, + "decision_coverage": 0.8918918918918919, + "early_accept": 26, + "early_reject": 7, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.905553117931728, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.02745974548325657, + "remaining_h20_hours": 0.018465968481111112 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.025806101009444443, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7911494801563888, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 7, + "correctly_saved_h20_hours": 0.7226741330227777, + "decision_coverage": 0.8108108108108109, + "early_accept": 23, + "early_reject": 7, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.02745974548325657, + "remaining_h20_hours": 0.018465968481111112 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.018465968481111112, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7411401015038888, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.918918918918919, + "balanced_accuracy": 0.8333333333333333, + "brier": 0.04804849349556749, + "confusion": { + "false_negative": 0, + "false_positive": 3, + "true_negative": 6, + "true_positive": 28 + }, + "log_loss": 0.15081259765362642 + }, + "policies": [ + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.7836433176388888, + "decision_coverage": 0.8918918918918919, + "early_accept": 27, + "early_reject": 6, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8878008796435344, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.007340132528333333, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7909834501672222, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.718791861695, + "decision_coverage": 0.7837837837837838, + "early_accept": 24, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.718791861695, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6736824212718535, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 10, + "correctly_saved_h20_hours": 0.7046020469899998, + "decision_coverage": 0.7297297297297297, + "early_accept": 23, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7046020469899998, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6603831211026486, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 33, + "both_wrong": 2, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 1 + } + }, + "1.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.1333333333333333, + 0.05555555555555547 + ], + "point": -0.027027027027027084 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.047874993358699436, + 0.04753434529263568 + ], + "point": -0.0026830647286030934 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8531746031746033, + "brier": 0.07268572313801988, + "confusion": { + "false_negative": 2, + "false_positive": 2, + "true_negative": 7, + "true_positive": 26 + }, + "log_loss": 0.22136936001258795 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7938369600363888, + "decision_coverage": 0.918918918918919, + "early_accept": 27, + "early_reject": 7, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8623652659223954, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.11022926282564288, + "remaining_h20_hours": 0.018465968481111112 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.025806101009444443, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.8196430610458332, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.7340601279138887, + "decision_coverage": 0.7837837837837838, + "early_accept": 24, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7340601279138887, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6879924922438714, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 15, + "correctly_saved_h20_hours": 0.620235621837222, + "decision_coverage": 0.5945945945945946, + "early_accept": 17, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.620235621837222, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5813113054633519, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.918918918918919, + "balanced_accuracy": 0.8333333333333333, + "brier": 0.07536878786662297, + "confusion": { + "false_negative": 0, + "false_positive": 3, + "true_negative": 6, + "true_positive": 28 + }, + "log_loss": 0.23513650573912553 + }, + "policies": [ + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.7625537088033332, + "decision_coverage": 0.8648648648648649, + "early_accept": 27, + "early_reject": 5, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8486897200744762, + "remaining_h20_hours": 0.007340132528333333 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.8216385310591849, + "remaining_h20_hours": 0.01680367415111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.024143806679444442, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7866975154827778, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 11, + "correctly_saved_h20_hours": 0.7042463331286111, + "decision_coverage": 0.7027027027027027, + "early_accept": 22, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7042463331286111, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6600497308847137, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 19, + "correctly_saved_h20_hours": 0.5876473660130556, + "decision_coverage": 0.4864864864864865, + "early_accept": 15, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.5876473660130556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5507682007641966, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 32, + "both_wrong": 2, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 2 + } + }, + "10.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.0 + ], + "point": 0.0 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.026808476987744526, + 0.02087777396654144 + ], + "point": -0.005783107568599871 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7420634920634921, + "brier": 0.08901809268133021, + "confusion": { + "false_negative": 2, + "false_positive": 4, + "true_negative": 5, + "true_positive": 26 + }, + "log_loss": 0.2787387373669301 + }, + "policies": [ + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.7770332858852778, + "decision_coverage": 0.8648648648648649, + "early_accept": 27, + "early_reject": 5, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8449371690787663, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.007340132528333333, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7843734184136111, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 17, + "correctly_saved_h20_hours": 0.6023881611255555, + "decision_coverage": 0.5405405405405406, + "early_accept": 16, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.6023881611255555, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.564583903295168, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 30, + "correctly_saved_h20_hours": 0.28353444813305556, + "decision_coverage": 0.1891891891891892, + "early_accept": 6, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.28353444813305556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.2657405901644815, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7420634920634921, + "brier": 0.09480120024993008, + "confusion": { + "false_negative": 2, + "false_positive": 4, + "true_negative": 5, + "true_positive": 26 + }, + "log_loss": 0.31278388302658006 + }, + "policies": [ + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.7566135917169443, + "decision_coverage": 0.7837837837837838, + "early_accept": 26, + "early_reject": 3, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8320431084673425, + "remaining_h20_hours": 0.007340132528333333 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.007340132528333333, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.7639537242452776, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 24, + "correctly_saved_h20_hours": 0.5226926016833333, + "decision_coverage": 0.35135135135135137, + "early_accept": 12, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.5226926016833333, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.48988982242028906, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 34, + "correctly_saved_h20_hours": 0.16934238506333332, + "decision_coverage": 0.08108108108108109, + "early_accept": 3, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.8475150590230554, + "saved_h20_hours_if_decisions_used": 0.16934238506333332, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.15871491327739184, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 6, + "instrumentation_only_correct": 0, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 0 + } + } + }, + "remaining_h20_hours": { + "distinct_n": 37, + "max": 0.057199826214444444, + "min": 0.0012684118816666664, + "n": 37 + } + }, + "15": { + "cells": 12, + "completion_time_sources": { + "reconstructed_from_latency": 36 + }, + "examples": 36, + "headline": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.06896551724137923, + 0.11428571428571421 + ], + "point": 0.02777777777777779 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.07972316141025643, + -0.004101962239149991 + ], + "point": -0.03744665409603411 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9166666666666666, + "balanced_accuracy": 0.8571428571428572, + "brier": 0.03946379437502599, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.12484297369457635 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7164359347680556, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.7164359347680556, + "threshold": 0.8, + "valid_cost_reduction_fraction": 0.6740304458143536, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 6, + "correctly_saved_h20_hours": 0.6905889710727777, + "decision_coverage": 0.8333333333333334, + "early_accept": 25, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6905889710727777, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6497133511279798, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.5895315664361112, + "decision_coverage": 0.6666666666666666, + "early_accept": 20, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.5895315664361112, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5546374843344664, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8888888888888888, + "balanced_accuracy": 0.75, + "brier": 0.0769104484710601, + "confusion": { + "false_negative": 0, + "false_positive": 4, + "true_negative": 4, + "true_positive": 28 + }, + "log_loss": 0.23230947656220863 + }, + "policies": [ + { + "abstain_continue_full": 6, + "correctly_saved_h20_hours": 0.6709655458816667, + "decision_coverage": 0.8333333333333334, + "early_accept": 26, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.8578028168458919, + "remaining_h20_hours": 0.0014445317586111114 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0014445317586111114, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6724100776402777, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 11, + "correctly_saved_h20_hours": 0.626660078601389, + "decision_coverage": 0.6944444444444444, + "early_accept": 22, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.626660078601389, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.5895683782116529, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 16, + "correctly_saved_h20_hours": 0.5494646261305556, + "decision_coverage": 0.5555555555555556, + "early_accept": 17, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.5494646261305556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5169420864266074, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 2, + "instrumentation_only_correct": 2, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 1 + } + }, + "headline_regularization": 1.0, + "label_sanity": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 36, + "negative": 8, + "positive": 28, + "primary_adjudicated_disagreements": 3 + }, + "regularization_sensitivity": { + "0.1": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.13793103448275856 + ], + "point": 0.05555555555555558 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.08924754562701061, + 0.007148232838160279 + ], + "point": -0.03187976039254431 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9444444444444444, + "balanced_accuracy": 0.875, + "brier": 0.03162164266499348, + "confusion": { + "false_negative": 0, + "false_positive": 2, + "true_negative": 6, + "true_positive": 28 + }, + "log_loss": 0.08770869035960555 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7164359347680556, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.7164359347680556, + "threshold": 0.8, + "valid_cost_reduction_fraction": 0.6740304458143536, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7164359347680556, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.7164359347680556, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6740304458143536, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.6910240435036111, + "decision_coverage": 0.8888888888888888, + "early_accept": 26, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6910240435036111, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6501226718366222, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8888888888888888, + "balanced_accuracy": 0.7946428571428572, + "brier": 0.06350140305753779, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 5, + "true_positive": 27 + }, + "log_loss": 0.2433253313244278 + }, + "policies": [ + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.689579511745, + "decision_coverage": 0.8888888888888888, + "early_accept": 27, + "early_reject": 5, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9922655130114992, + "remaining_h20_hours": 0.0014445317586111114 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0014445317586111114, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6910240435036111, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.6783313931494445, + "decision_coverage": 0.8611111111111112, + "early_accept": 27, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9922655130114992, + "remaining_h20_hours": 0.0014445317586111114 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0014445317586111114, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6797759249080555, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.6449625425236111, + "decision_coverage": 0.7777777777777778, + "early_accept": 24, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9922655130114992, + "remaining_h20_hours": 0.0014445317586111114 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0014445317586111114, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6464070742822221, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 32, + "both_wrong": 2, + "instrumentation_only_correct": 2, + "mcnemar_exact_two_sided_p": 0.5, + "outcome_only_correct": 0 + } + }, + "1.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.06896551724137923, + 0.11428571428571421 + ], + "point": 0.02777777777777779 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.07972316141025643, + -0.004101962239149991 + ], + "point": -0.03744665409603411 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9166666666666666, + "balanced_accuracy": 0.8571428571428572, + "brier": 0.03946379437502599, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.12484297369457635 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.7164359347680556, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.7164359347680556, + "threshold": 0.8, + "valid_cost_reduction_fraction": 0.6740304458143536, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 6, + "correctly_saved_h20_hours": 0.6905889710727777, + "decision_coverage": 0.8333333333333334, + "early_accept": 25, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6905889710727777, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.6497133511279798, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.5895315664361112, + "decision_coverage": 0.6666666666666666, + "early_accept": 20, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.5895315664361112, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5546374843344664, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8888888888888888, + "balanced_accuracy": 0.75, + "brier": 0.0769104484710601, + "confusion": { + "false_negative": 0, + "false_positive": 4, + "true_negative": 4, + "true_positive": 28 + }, + "log_loss": 0.23230947656220863 + }, + "policies": [ + { + "abstain_continue_full": 6, + "correctly_saved_h20_hours": 0.6709655458816667, + "decision_coverage": 0.8333333333333334, + "early_accept": 26, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.8578028168458919, + "remaining_h20_hours": 0.0014445317586111114 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0014445317586111114, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6724100776402777, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 11, + "correctly_saved_h20_hours": 0.626660078601389, + "decision_coverage": 0.6944444444444444, + "early_accept": 22, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.626660078601389, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.5895683782116529, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 16, + "correctly_saved_h20_hours": 0.5494646261305556, + "decision_coverage": 0.5555555555555556, + "early_accept": 17, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.5494646261305556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5169420864266074, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 2, + "instrumentation_only_correct": 2, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 1 + } + }, + "10.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.17142857142857137 + ], + "point": 0.08333333333333326 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.051136314558473485, + -0.009942733239401492 + ], + "point": -0.029809575955846893 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8888888888888888, + "balanced_accuracy": 0.8392857142857143, + "brier": 0.07023194254174596, + "confusion": { + "false_negative": 2, + "false_positive": 2, + "true_negative": 6, + "true_positive": 26 + }, + "log_loss": 0.23287532021698182 + }, + "policies": [ + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.6924315406702778, + "decision_coverage": 0.8611111111111112, + "early_accept": 27, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.801686701254931, + "remaining_h20_hours": 0.004004611363888889 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.004004611363888889, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6964361520341666, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 14, + "correctly_saved_h20_hours": 0.5771830138255556, + "decision_coverage": 0.6111111111111112, + "early_accept": 19, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.5771830138255556, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.5430198364509199, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 26, + "correctly_saved_h20_hours": 0.30550869967055555, + "decision_coverage": 0.2777777777777778, + "early_accept": 8, + "early_reject": 2, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.30550869967055555, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.2874257907034979, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8055555555555556, + "balanced_accuracy": 0.6517857142857143, + "brier": 0.10004151849759285, + "confusion": { + "false_negative": 2, + "false_positive": 5, + "true_negative": 3, + "true_positive": 26 + }, + "log_loss": 0.32055333394171265 + }, + "policies": [ + { + "abstain_continue_full": 7, + "correctly_saved_h20_hours": 0.6877041925958333, + "decision_coverage": 0.8055555555555556, + "early_accept": 27, + "early_reject": 2, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8140722809073944, + "remaining_h20_hours": 0.004562354750555555 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.004562354750555555, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.6922665473463888, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 24, + "correctly_saved_h20_hours": 0.4570601465275, + "decision_coverage": 0.3333333333333333, + "early_accept": 11, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.4570601465275, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.43000698230978923, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 33, + "correctly_saved_h20_hours": 0.15267571839666666, + "decision_coverage": 0.08333333333333333, + "early_accept": 3, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.7379133138080555, + "saved_h20_hours_if_decisions_used": 0.15267571839666666, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.14363891806913368, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 29, + "both_wrong": 4, + "instrumentation_only_correct": 3, + "mcnemar_exact_two_sided_p": 0.25, + "outcome_only_correct": 0 + } + } + }, + "remaining_h20_hours": { + "distinct_n": 36, + "max": 0.05164427065888889, + "min": 0.0014445317586111114, + "n": 36 + } + }, + "20": { + "cells": 12, + "completion_time_sources": { + "reconstructed_from_latency": 36 + }, + "examples": 36, + "headline": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.12195121951219512 + ], + "point": 0.05555555555555547 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.09816256823569453, + -0.009602741441109449 + ], + "point": -0.04807041919163683 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9166666666666666, + "balanced_accuracy": 0.8571428571428572, + "brier": 0.05902138516195827, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.24132948833387438 + }, + "policies": [ + { + "abstain_continue_full": 2, + "correctly_saved_h20_hours": 0.6191580696761111, + "decision_coverage": 0.9444444444444444, + "early_accept": 28, + "early_reject": 6, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8399807697230681, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.009139434342777777, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6282975040188888, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.6101058238716667, + "decision_coverage": 0.8611111111111112, + "early_accept": 27, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6174606812416666, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.5390392875763889, + "decision_coverage": 0.75, + "early_accept": 23, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5463941449463888, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8611111111111112, + "balanced_accuracy": 0.7321428571428572, + "brier": 0.1070918043535951, + "confusion": { + "false_negative": 1, + "false_positive": 4, + "true_negative": 4, + "true_positive": 27 + }, + "log_loss": 0.437208492871466 + }, + "policies": [ + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.5864084509805555, + "decision_coverage": 0.8888888888888888, + "early_accept": 28, + "early_reject": 4, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9419008835359353, + "remaining_h20_hours": 5.564286972222258e-05 + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.8529942983175343, + "remaining_h20_hours": 0.0012268335861111111 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8268198678404454, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.01042191079861111, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5968303617791667, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.5682670563286111, + "decision_coverage": 0.7777777777777778, + "early_accept": 25, + "early_reject": 3, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9419008835359353, + "remaining_h20_hours": 5.564286972222258e-05 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.007410500239722223, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5756775565683333, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 17, + "correctly_saved_h20_hours": 0.40106617994777777, + "decision_coverage": 0.5277777777777778, + "early_accept": 16, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.4084210373177778, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 3, + "instrumentation_only_correct": 2, + "mcnemar_exact_two_sided_p": 0.5, + "outcome_only_correct": 0 + } + }, + "headline_regularization": 1.0, + "label_sanity": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 36, + "negative": 8, + "positive": 28, + "primary_adjudicated_disagreements": 3 + }, + "regularization_sensitivity": { + "0.1": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.1842105263157895 + ], + "point": 0.08333333333333326 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.1289106028344674, + -0.00039980805027801426 + ], + "point": -0.0571263639718715 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9444444444444444, + "balanced_accuracy": 0.9196428571428572, + "brier": 0.05738846072877349, + "confusion": { + "false_negative": 1, + "false_positive": 1, + "true_negative": 7, + "true_positive": 27 + }, + "log_loss": 0.3180853188579326 + }, + "policies": [ + { + "abstain_continue_full": 2, + "correctly_saved_h20_hours": 0.6191580696761111, + "decision_coverage": 0.9444444444444444, + "early_accept": 28, + "early_reject": 6, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8655614945720641, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.00028842593944516384, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.009139434342777777, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6282975040188888, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.6191580696761111, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.00028842593944516384, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6265129270461112, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.6074562703719444, + "decision_coverage": 0.8888888888888888, + "early_accept": 26, + "early_reject": 6, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.00028842593944516384, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6148111277419444, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8611111111111112, + "balanced_accuracy": 0.7767857142857143, + "brier": 0.11451482470064499, + "confusion": { + "false_negative": 2, + "false_positive": 3, + "true_negative": 5, + "true_positive": 26 + }, + "log_loss": 0.7747412375287079 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.5848221568852777, + "decision_coverage": 0.9166666666666666, + "early_accept": 27, + "early_reject": 6, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9999405808371276, + "remaining_h20_hours": 5.564286972222258e-05 + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.9739951015698483, + "remaining_h20_hours": 0.0012268335861111111 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 1.6376668165938014e-05, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.008637333825833335, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5934594907111111, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.5791295938452777, + "decision_coverage": 0.8888888888888888, + "early_accept": 27, + "early_reject": 5, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9999405808371276, + "remaining_h20_hours": 5.564286972222258e-05 + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.9739951015698483, + "remaining_h20_hours": 0.0012268335861111111 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 1.6376668165938014e-05, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.008637333825833335, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5877669276711112, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.5676638636697222, + "decision_coverage": 0.8611111111111112, + "early_accept": 26, + "early_reject": 5, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9999405808371276, + "remaining_h20_hours": 5.564286972222258e-05 + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.9739951015698483, + "remaining_h20_hours": 0.0012268335861111111 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 1.6376668165938014e-05, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.008637333825833335, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5763011974955555, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 2, + "instrumentation_only_correct": 3, + "mcnemar_exact_two_sided_p": 0.25, + "outcome_only_correct": 0 + } + }, + "1.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.12195121951219512 + ], + "point": 0.05555555555555547 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.09816256823569453, + -0.009602741441109449 + ], + "point": -0.04807041919163683 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9166666666666666, + "balanced_accuracy": 0.8571428571428572, + "brier": 0.05902138516195827, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.24132948833387438 + }, + "policies": [ + { + "abstain_continue_full": 2, + "correctly_saved_h20_hours": 0.6191580696761111, + "decision_coverage": 0.9444444444444444, + "early_accept": 28, + "early_reject": 6, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8399807697230681, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.009139434342777777, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6282975040188888, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.6101058238716667, + "decision_coverage": 0.8611111111111112, + "early_accept": 27, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6174606812416666, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.5390392875763889, + "decision_coverage": 0.75, + "early_accept": 23, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.008919034978436878, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5463941449463888, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8611111111111112, + "balanced_accuracy": 0.7321428571428572, + "brier": 0.1070918043535951, + "confusion": { + "false_negative": 1, + "false_positive": 4, + "true_negative": 4, + "true_positive": 27 + }, + "log_loss": 0.437208492871466 + }, + "policies": [ + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.5864084509805555, + "decision_coverage": 0.8888888888888888, + "early_accept": 28, + "early_reject": 4, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9419008835359353, + "remaining_h20_hours": 5.564286972222258e-05 + }, + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.8529942983175343, + "remaining_h20_hours": 0.0012268335861111111 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8268198678404454, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.01042191079861111, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5968303617791667, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.5682670563286111, + "decision_coverage": 0.7777777777777778, + "early_accept": 25, + "early_reject": 3, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.5, + "cell": "tp1_mns64", + "label_feasible": false, + "probability_feasible": 0.9419008835359353, + "remaining_h20_hours": 5.564286972222258e-05 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.007410500239722223, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5756775565683333, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 17, + "correctly_saved_h20_hours": 0.40106617994777777, + "decision_coverage": 0.5277777777777778, + "early_accept": 16, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0017338236710752416, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.4084210373177778, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 31, + "both_wrong": 3, + "instrumentation_only_correct": 2, + "mcnemar_exact_two_sided_p": 0.5, + "outcome_only_correct": 0 + } + }, + "10.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.17142857142857137 + ], + "point": 0.08333333333333326 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.047457441513371, + -0.01422543161645023 + ], + "point": -0.029913894255621076 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.9166666666666666, + "balanced_accuracy": 0.8571428571428572, + "brier": 0.07346598213380073, + "confusion": { + "false_negative": 1, + "false_positive": 2, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.24820863982463115 + }, + "policies": [ + { + "abstain_continue_full": 5, + "correctly_saved_h20_hours": 0.6026586119891666, + "decision_coverage": 0.8611111111111112, + "early_accept": 27, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8171600256181173, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.08736122237187642, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.009139434342777777, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6117980463319445, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.5909011698152777, + "decision_coverage": 0.7777777777777778, + "early_accept": 25, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.08736122237187642, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5982560271852778, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 27, + "correctly_saved_h20_hours": 0.27948134858055557, + "decision_coverage": 0.25, + "early_accept": 9, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.27948134858055557, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.26293898566315754, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.7142857142857143, + "brier": 0.10337987638942181, + "confusion": { + "false_negative": 2, + "false_positive": 4, + "true_negative": 4, + "true_positive": 26 + }, + "log_loss": 0.34599307331049317 + }, + "policies": [ + { + "abstain_continue_full": 7, + "correctly_saved_h20_hours": 0.6007090416925, + "decision_coverage": 0.8055555555555556, + "early_accept": 27, + "early_reject": 2, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8140697427804628, + "remaining_h20_hours": 0.0017845769727777771 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0444153615651932, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.009139434342777777, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.6098484760352777, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.5199227922302778, + "decision_coverage": 0.6666666666666666, + "early_accept": 23, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0444153615651932, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.5272776496002778, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 33, + "correctly_saved_h20_hours": 0.08992033662666665, + "decision_coverage": 0.08333333333333333, + "early_accept": 2, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.033717411016, + "cell": "tp4_mns32", + "label_feasible": true, + "probability_feasible": 0.0444153615651932, + "remaining_h20_hours": 0.0073548573700000005 + } + ], + "full_trial_h20_hours": 1.0629133138080555, + "invalidly_saved_h20_hours": 0.0073548573700000005, + "remaining_h20_hours_at_cutoff": 0.6295799804747222, + "saved_h20_hours_if_decisions_used": 0.09727519399666666, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 3, + "instrumentation_only_correct": 3, + "mcnemar_exact_two_sided_p": 0.25, + "outcome_only_correct": 0 + } + } + }, + "remaining_h20_hours": { + "distinct_n": 36, + "max": 0.04608871510333333, + "min": 5.564286972222258e-05, + "n": 36 + } + }, + "5": { + "cells": 12, + "completion_time_sources": { + "none_completed": 2, + "reconstructed_from_latency": 35 + }, + "examples": 37, + "headline": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.02777777777777779, + 0.19999999999999996 + ], + "point": 0.10810810810810811 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.08739451923629489, + 0.010803161364314168 + ], + "point": -0.03964626050589401 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8154761904761905, + "brier": 0.09006542265758688, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.2730810945072346 + }, + "policies": [ + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.7828422774630556, + "decision_coverage": 0.7567567567567568, + "early_accept": 23, + "early_reject": 5, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.918171067470718, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7929601877691667, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.7136471349655555, + "decision_coverage": 0.6756756756756757, + "early_accept": 21, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.918171067470718, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7237650452716666, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 15, + "correctly_saved_h20_hours": 0.6518691238655556, + "decision_coverage": 0.5945945945945946, + "early_accept": 18, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.6518691238655556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6109595741422738, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.7837837837837838, + "balanced_accuracy": 0.7063492063492063, + "brier": 0.1297116831634809, + "confusion": { + "false_negative": 4, + "false_positive": 4, + "true_negative": 5, + "true_positive": 24 + }, + "log_loss": 0.38484417997476916 + }, + "policies": [ + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.745153891865, + "decision_coverage": 0.7567567567567568, + "early_accept": 24, + "early_reject": 4, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.8051358474368825, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9221804334720108, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.8933807379877649, + "remaining_h20_hours": 0.022359229706666667 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.042037306932222226, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7871911987972221, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 14, + "correctly_saved_h20_hours": 0.6362291990002777, + "decision_coverage": 0.6216216216216216, + "early_accept": 19, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9221804334720108, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.6463471093063888, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 21, + "correctly_saved_h20_hours": 0.3878475556194444, + "decision_coverage": 0.43243243243243246, + "early_accept": 13, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.3878475556194444, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.3635072881013599, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 29, + "both_wrong": 4, + "instrumentation_only_correct": 4, + "mcnemar_exact_two_sided_p": 0.125, + "outcome_only_correct": 0 + } + }, + "headline_regularization": 1.0, + "label_sanity": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 37, + "negative": 9, + "positive": 28, + "primary_adjudicated_disagreements": 3 + }, + "regularization_sensitivity": { + "0.1": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.06980897009966774, + 0.21875 + ], + "point": 0.08108108108108103 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.1649967051920624, + 0.06447487686406284 + ], + "point": -0.05737762784013967 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7797619047619048, + "brier": 0.13521691422103324, + "confusion": { + "false_negative": 3, + "false_positive": 3, + "true_negative": 6, + "true_positive": 25 + }, + "log_loss": 0.4511157566254769 + }, + "policies": [ + { + "abstain_continue_full": 3, + "correctly_saved_h20_hours": 0.8122652196319445, + "decision_coverage": 0.918918918918919, + "early_accept": 26, + "early_reject": 8, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.9405657729239719, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783420846880035, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns16", + "label_feasible": false, + "probability_feasible": 0.9635702290378427, + "remaining_h20_hours": 0.023106262014444445 + } + ], + "false_reject": 2, + "false_reject_examples": [ + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.07387488532278061, + "remaining_h20_hours": 0.015631309493055554 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns16", + "label_feasible": true, + "probability_feasible": 0.17417966056624304, + "remaining_h20_hours": 0.03127135866722222 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.08968700740027778, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.9019522270322223, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 6, + "correctly_saved_h20_hours": 0.765429306841111, + "decision_coverage": 0.8378378378378378, + "early_accept": 24, + "early_reject": 7, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.9405657729239719, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783420846880035, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns16", + "label_feasible": false, + "probability_feasible": 0.9635702290378427, + "remaining_h20_hours": 0.023106262014444445 + } + ], + "false_reject": 1, + "false_reject_examples": [ + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.07387488532278061, + "remaining_h20_hours": 0.015631309493055554 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.05841564873305555, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.8238449555741666, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.7469523484622221, + "decision_coverage": 0.7567567567567568, + "early_accept": 22, + "early_reject": 6, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783420846880035, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns16", + "label_feasible": false, + "probability_feasible": 0.9635702290378427, + "remaining_h20_hours": 0.023106262014444445 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.03322417232055556, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7801765207827777, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.7567567567567568, + "balanced_accuracy": 0.7261904761904762, + "brier": 0.1925945420611729, + "confusion": { + "false_negative": 6, + "false_positive": 3, + "true_negative": 6, + "true_positive": 22 + }, + "log_loss": 0.7361068786964186 + }, + "policies": [ + { + "abstain_continue_full": 4, + "correctly_saved_h20_hours": 0.7745768340338888, + "decision_coverage": 0.8918918918918919, + "early_accept": 25, + "early_reject": 8, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.977871407650825, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783309849310482, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.9900156648758626, + "remaining_h20_hours": 0.022359229706666667 + } + ], + "false_reject": 3, + "false_reject_examples": [ + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.012882187111638552, + "remaining_h20_hours": 0.015631309493055554 + }, + { + "anchor": 0.2265625, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.016042537020255383, + "remaining_h20_hours": 0.01563306773861111 + }, + { + "anchor": 0.23046875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.15321211351284938, + "remaining_h20_hours": 0.015631133334166665 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.08893281749805555, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.8635096515319445, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 8, + "correctly_saved_h20_hours": 0.6784465414575, + "decision_coverage": 0.7837837837837838, + "early_accept": 22, + "early_reject": 7, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.977871407650825, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783309849310482, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.9900156648758626, + "remaining_h20_hours": 0.022359229706666667 + } + ], + "false_reject": 2, + "false_reject_examples": [ + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.012882187111638552, + "remaining_h20_hours": 0.015631309493055554 + }, + { + "anchor": 0.2265625, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.016042537020255383, + "remaining_h20_hours": 0.01563306773861111 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0733016841638889, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7517482256213891, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.6471751827902779, + "decision_coverage": 0.7567567567567568, + "early_accept": 21, + "early_reject": 7, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.977871407650825, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9783309849310482, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.9900156648758626, + "remaining_h20_hours": 0.022359229706666667 + } + ], + "false_reject": 2, + "false_reject_examples": [ + { + "anchor": 0.21875, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.012882187111638552, + "remaining_h20_hours": 0.015631309493055554 + }, + { + "anchor": 0.2265625, + "cell": "tp1_mns8", + "label_feasible": true, + "probability_feasible": 0.016042537020255383, + "remaining_h20_hours": 0.01563306773861111 + } + ], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0733016841638889, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7204768669541667, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + ] + }, + "paired_correctness": { + "both_correct": 26, + "both_wrong": 4, + "instrumentation_only_correct": 5, + "mcnemar_exact_two_sided_p": 0.453125, + "outcome_only_correct": 2 + } + }, + "1.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.02777777777777779, + 0.19999999999999996 + ], + "point": 0.10810810810810811 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.08739451923629489, + 0.010803161364314168 + ], + "point": -0.03964626050589401 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8154761904761905, + "brier": 0.09006542265758688, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.2730810945072346 + }, + "policies": [ + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.7828422774630556, + "decision_coverage": 0.7567567567567568, + "early_accept": 23, + "early_reject": 5, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.918171067470718, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7929601877691667, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.7136471349655555, + "decision_coverage": 0.6756756756756757, + "early_accept": 21, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.918171067470718, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7237650452716666, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 15, + "correctly_saved_h20_hours": 0.6518691238655556, + "decision_coverage": 0.5945945945945946, + "early_accept": 18, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.6518691238655556, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6109595741422738, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.7837837837837838, + "balanced_accuracy": 0.7063492063492063, + "brier": 0.1297116831634809, + "confusion": { + "false_negative": 4, + "false_positive": 4, + "true_negative": 5, + "true_positive": 24 + }, + "log_loss": 0.38484417997476916 + }, + "policies": [ + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.745153891865, + "decision_coverage": 0.7567567567567568, + "early_accept": 24, + "early_reject": 4, + "false_accept": 3, + "false_accept_examples": [ + { + "anchor": 0.75, + "cell": "tp2_mns64", + "label_feasible": false, + "probability_feasible": 0.8051358474368825, + "remaining_h20_hours": 0.009560166919444444 + }, + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9221804334720108, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.034252608017, + "cell": "tp4_mns8", + "label_feasible": false, + "probability_feasible": 0.8933807379877649, + "remaining_h20_hours": 0.022359229706666667 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.042037306932222226, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7871911987972221, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 14, + "correctly_saved_h20_hours": 0.6362291990002777, + "decision_coverage": 0.6216216216216216, + "early_accept": 19, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9221804334720108, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.6463471093063888, + "threshold": 0.9, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 21, + "correctly_saved_h20_hours": 0.3878475556194444, + "decision_coverage": 0.43243243243243246, + "early_accept": 13, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.3878475556194444, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.3635072881013599, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 29, + "both_wrong": 4, + "instrumentation_only_correct": 4, + "mcnemar_exact_two_sided_p": 0.125, + "outcome_only_correct": 0 + } + }, + "10.0": { + "bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.08698849104859337, + 0.07499999999999996 + ], + "point": 0.0 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.0419671863122155, + 0.010410939632453053 + ], + "point": -0.015389088872407489 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "instrumentation_aware": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7420634920634921, + "brier": 0.09695183324390907, + "confusion": { + "false_negative": 2, + "false_positive": 4, + "true_negative": 5, + "true_positive": 26 + }, + "log_loss": 0.31480154295615903 + }, + "policies": [ + { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.7917544064236111, + "decision_coverage": 0.7567567567567568, + "early_accept": 24, + "early_reject": 4, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8905819069991598, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.8018723167297224, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 21, + "correctly_saved_h20_hours": 0.5056768371675, + "decision_coverage": 0.43243243243243246, + "early_accept": 13, + "early_reject": 3, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.5056768371675, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.4739419214357306, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 30, + "correctly_saved_h20_hours": 0.19128570447222223, + "decision_coverage": 0.1891891891891892, + "early_accept": 6, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.19128570447222223, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.17928112908743485, + "valid_zero_error_policy": true + } + ] + }, + "outcome_only": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7043650793650793, + "brier": 0.11234092211631656, + "confusion": { + "false_negative": 1, + "false_positive": 5, + "true_negative": 4, + "true_positive": 27 + }, + "log_loss": 0.36773394130409404 + }, + "policies": [ + { + "abstain_continue_full": 15, + "correctly_saved_h20_hours": 0.6981008277897223, + "decision_coverage": 0.5945945945945946, + "early_accept": 21, + "early_reject": 1, + "false_accept": 1, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.8520031544496152, + "remaining_h20_hours": 0.010117910306111111 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.010117910306111111, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7082187380958334, + "threshold": 0.8, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + }, + { + "abstain_continue_full": 31, + "correctly_saved_h20_hours": 0.23290921672555553, + "decision_coverage": 0.16216216216216217, + "early_accept": 6, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.23290921672555553, + "threshold": 0.9, + "valid_cost_reduction_fraction": 0.21829246186816503, + "valid_zero_error_policy": true + }, + { + "abstain_continue_full": 34, + "correctly_saved_h20_hours": 0.13888648644666668, + "decision_coverage": 0.08108108108108109, + "early_accept": 3, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.13888648644666668, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.13017034479312573, + "valid_zero_error_policy": true + } + ] + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 5, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "outcome_only_correct": 1 + } + } + }, + "remaining_h20_hours": { + "distinct_n": 37, + "max": 0.06275538177, + "min": 0.0026573007705555556, + "n": 37 + } + } + }, + "cutoffs_s": [ + 5.0, + 10.0, + 15.0, + 20.0 + ], + "decision": { + "contribution_established": false, + "reason": "This dataset contains one workload and reconstructed rather than exact request completion times. Three TP4 primary trials also disagree with their 2-of-3 labels. It can reject a missing-signal premise but cannot establish generalization or a paper-facing cost reduction." + }, + "estimand": "2-of-3 adjudicated anchor feasibility from the first primary trial's identical short real prefix", + "headline_incremental_deltas": { + "10": { + "accuracy": -0.027027027027027084, + "brier": -0.0026830647286030934 + }, + "15": { + "accuracy": 0.02777777777777779, + "brier": -0.03744665409603411 + }, + "20": { + "accuracy": 0.05555555555555547, + "brier": -0.04807041919163683 + }, + "5": { + "accuracy": 0.10810810810810811, + "brier": -0.03964626050589401 + } + }, + "instrumentation_features": [ + "model_steps_per_second", + "waiting_mean", + "waiting_max", + "waiting_nonzero_share", + "running_mean", + "running_max", + "decode_batch_mean", + "decode_batch_max", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "preemptions" + ], + "model": "same L2 logistic model and folds; instrumentation model appends Layer-1 features", + "outcome_features": [ + "log_offered_rate_per_gpu", + "log2_tp", + "log2_max_num_seqs", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit" + ], + "provenance": { + "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", + "phase6_metrics_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", + "raw_root": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/solo-authoritative/cells" + }, + "sanity": { + "cutoff_count": 4, + "invariants": { + "cutoffs_positive": true, + "full_trial_label_not_used_as_feature": true, + "paired_same_model_family": true, + "probabilities_checked_in_unit_interval": true, + "records_strictly_prefix_sliced": true + }, + "red_flags": [] + }, + "schema": "fidelity-prefix-v1", + "scope": "retrospective single-workload prefix diagnostic; model selection, threshold choice, and contribution claims require held-out prospective tasks", + "split": "leave-one-configuration-cell-out", + "status": "PASS" +} diff --git a/runs/fidelity-headroom/prepare_pilot.py b/runs/fidelity-headroom/prepare_pilot.py new file mode 100644 index 0000000..168beb6 --- /dev/null +++ b/runs/fidelity-headroom/prepare_pilot.py @@ -0,0 +1,355 @@ +#!/usr/bin/env python3 +"""Materialize session-disjoint pilot repeats and freeze attainable anchors. + +The private outputs retain prompt text and stay on the experiment host. The +public manifest contains only aggregate counts, hashes, paths, and parameters. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import sys +from pathlib import Path +from typing import Any + + +AITUNER_ROOT = Path(os.environ.get("AITUNER_ROOT", Path(__file__).resolve().parents[2])) +sys.path.insert(0, str(AITUNER_ROOT / "src")) + +from aituner.spec import load_study_spec # noqa: E402 +from aituner.trace import load_trace_requests, select_requests_for_threshold # noqa: E402 + + +ROLES = ("burnin", "low1", "high1", "low2", "high2", "low3", "high3") +CELLS = { + "tp1_mns8": {"tp": 1, "mns": 8, "frontier_req_s_gpu": 2.3833333333333333}, + "tp1_mns64": {"tp": 1, "mns": 64, "frontier_req_s_gpu": 2.3833333333333333}, + "tp2_mns8": {"tp": 2, "mns": 8, "frontier_req_s_gpu": 2.2416666666666667}, + "tp2_mns64": {"tp": 2, "mns": 64, "frontier_req_s_gpu": 2.3}, + "tp4_mns16": {"tp": 4, "mns": 16, "frontier_req_s_gpu": 2.5}, + "tp4_mns64": {"tp": 4, "mns": 64, "frontier_req_s_gpu": 2.5}, +} +TARGET_MULTIPLIERS = {"low": 0.85, "high": 1.25} + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + tmp = path.with_suffix(path.suffix + ".tmp") + tmp.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8") + os.replace(tmp, path) + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def order_hash(values: list[str]) -> str: + return hashlib.sha256("\n".join(values).encode()).hexdigest() + + +def resolve_source_trace(windows_path: Path, window_id: str) -> tuple[dict[str, Any], Path]: + payload = json.loads(windows_path.read_text(encoding="utf-8")) + for window in payload["windows"]: + if window["window_id"] != window_id: + continue + trace = Path(window["trace_file"]) + if not trace.is_absolute(): + trace = (windows_path.parent / trace).resolve() + return window, trace + raise ValueError(f"window not found: {window_id}") + + +def materialize_bands( + source_trace: Path, + source_window: dict[str, Any], + private_root: Path, +) -> tuple[Path, dict[str, Any]]: + traces_root = private_root / "traces" + traces_root.mkdir(parents=True, exist_ok=True) + temporary = {role: traces_root / f".{role}.jsonl.tmp" for role in ROLES} + final = {role: traces_root / f"{role}.jsonl" for role in ROLES} + handles = {role: temporary[role].open("w", encoding="utf-8") for role in ROLES} + stats = { + role: { + "rows": 0, + "sum_input_tokens": 0, + "min_timestamp": None, + "max_timestamp": None, + } + for role in ROLES + } + try: + with source_trace.open(encoding="utf-8") as source: + for line_number, line in enumerate(source): + row = json.loads(line) + value = float(row["sampling_u"]) + if not 0.0 <= value <= 1.0: + raise ValueError(f"sampling_u outside [0,1] at line {line_number}") + band = min(len(ROLES) - 1, int(value * len(ROLES))) + role = ROLES[band] + remapped = value * len(ROLES) - band + row["sampling_u"] = min(remapped, math.nextafter(1.0, 0.0)) + row["fidelity_pilot_band"] = role + handles[role].write(json.dumps(row, ensure_ascii=False) + "\n") + timestamp = float(row["timestamp"]) + item = stats[role] + item["rows"] += 1 + item["sum_input_tokens"] += int(row.get("input_length") or 0) + item["min_timestamp"] = ( + timestamp if item["min_timestamp"] is None + else min(float(item["min_timestamp"]), timestamp) + ) + item["max_timestamp"] = ( + timestamp if item["max_timestamp"] is None + else max(float(item["max_timestamp"]), timestamp) + ) + finally: + for handle in handles.values(): + handle.close() + for role in ROLES: + os.replace(temporary[role], final[role]) + stats[role]["sha256"] = sha256_file(final[role]) + stats[role]["bytes"] = final[role].stat().st_size + + windows = [] + for role in ROLES: + window = dict(source_window) + window["window_id"] = f"fidelity_pilot_{role}" + window["trace_file"] = f"traces/{role}.jsonl" + window["num_requests"] = stats[role]["rows"] + window["sum_input_length"] = stats[role]["sum_input_tokens"] + window["sampling_strategy"] = "session_uniform_seven_disjoint_bands_remapped" + window["fidelity_pilot_role"] = role + windows.append(window) + private_windows = private_root / "windows.json" + atomic_json( + private_windows, + { + "schema": "fidelity-pilot-private-windows-v1", + "roles": list(ROLES), + "windows": windows, + }, + ) + return private_windows, stats + + +def write_studies( + *, + base_primary: Path, + base_tp4: Path, + private_windows: Path, + private_root: Path, +) -> dict[str, dict[str, Path]]: + bases = { + "primary": json.loads(base_primary.read_text(encoding="utf-8")), + "tp4": json.loads(base_tp4.read_text(encoding="utf-8")), + } + result: dict[str, dict[str, Path]] = {} + for role in ROLES: + result[role] = {} + for tier, base in bases.items(): + payload = json.loads(json.dumps(base)) + payload["study_id"] = f"fidelity-prefix-pilot-{role}-{tier}" + payload["hardware"]["host_candidates"] = ["dash0"] + payload["engine"]["engine_version"] = "0.24.1.dev3+opprof" + payload["trace"]["windows_path"] = str(private_windows) + payload["trace"]["window_id"] = f"fidelity_pilot_{role}" + path = private_root / "studies" / f"{role}-{tier}.json" + atomic_json(path, payload) + result[role][tier] = path + return result + + +def attainable_anchor(requests: list[Any], target_count: int) -> tuple[float, list[Any]]: + ordered = sorted(float(request.sampling_u) for request in requests) + if not ordered: + raise ValueError("no requests after study filtering") + candidate_indices = sorted({ + max(0, min(len(ordered) - 1, target_count - 1)), + max(0, min(len(ordered) - 1, target_count)), + }) + candidates = [] + for index in candidate_indices: + anchor = ordered[index] + selected = select_requests_for_threshold(requests, threshold=anchor) + candidates.append((abs(len(selected) - target_count), len(selected), anchor, selected)) + _error, _count, anchor, selected = min(candidates, key=lambda item: (item[0], item[1])) + return anchor, selected + + +def selected_record(selected: list[Any], *, tp: int, duration_s: float) -> dict[str, Any]: + return { + "anchor": max(float(request.sampling_u) for request in selected), + "selected_count": len(selected), + "offered_req_s": len(selected) / duration_s, + "offered_req_s_per_gpu": len(selected) / duration_s / tp, + "request_id_order_sha256": order_hash([request.row_id for request in selected]), + "arrival_order_sha256": order_hash([f"{request.arrival_s:.12f}" for request in selected]), + "input_length_order_sha256": order_hash( + [str(request.prompt_tokens_hint) for request in selected] + ), + } + + +def build_manifest( + *, + studies: dict[str, dict[str, Path]], + private_windows: Path, + band_stats: dict[str, Any], + source_trace: Path, + source_windows: Path, + source_window_id: str, +) -> dict[str, Any]: + loaded = {} + durations = {} + for role, tiers in studies.items(): + loaded[role] = {} + for tier, path in tiers.items(): + study = load_study_spec(path) + window, requests = load_trace_requests(study, study_spec_path=path) + loaded[role][tier] = requests + durations[role] = float(window.window_end - window.window_start) + + cells = {} + all_hashes = [] + for cell, config in CELLS.items(): + tp = int(config["tp"]) + tier = "tp4" if tp == 4 else "primary" + targets = {} + for level, multiplier in TARGET_MULTIPLIERS.items(): + target_rate = float(config["frontier_req_s_gpu"]) * multiplier + target_count = round(target_rate * durations["low1"] * tp) + roles = [ + role + for role in ROLES + if role.startswith(level) or (level == "low" and role == "burnin") + ] + selections = {} + for role in roles: + anchor, selected = attainable_anchor(loaded[role][tier], target_count) + record = selected_record(selected, tp=tp, duration_s=durations[role]) + record["anchor"] = anchor + record["study"] = str(studies[role][tier]) + selections[role] = record + all_hashes.append(record["request_id_order_sha256"]) + targets[level] = { + "multiplier": multiplier, + "target_req_s_per_gpu": target_rate, + "target_count": target_count, + "selections": selections, + } + cells[cell] = {**config, "targets": targets} + + red_flags = [] + for cell, config in cells.items(): + for level, target in config["targets"].items(): + if not target["selections"]: + red_flags.append(f"missing_{cell}_{level}") + for selection in target["selections"].values(): + if selection["selected_count"] <= 0: + red_flags.append(f"empty_{cell}_{level}") + per_cell_distinct = {} + for cell, config in cells.items(): + hashes = [ + selection["request_id_order_sha256"] + for target in config["targets"].values() + for selection in target["selections"].values() + ] + per_cell_distinct[cell] = len(hashes) == len(set(hashes)) + if not per_cell_distinct[cell]: + red_flags.append(f"session_bands_overlap_{cell}") + return { + "schema": "fidelity-prefix-pilot-manifest-v1", + "status": "PASS" if not red_flags else "STOP", + "source": { + "windows": str(source_windows), + "window_id": source_window_id, + "trace": str(source_trace), + "trace_sha256": sha256_file(source_trace), + }, + "private": { + "windows": str(private_windows), + "windows_sha256": sha256_file(private_windows), + "band_stats": band_stats, + "studies": { + role: {tier: str(path) for tier, path in tiers.items()} + for role, tiers in studies.items() + }, + }, + "roles": list(ROLES), + "cells": cells, + "execution": { + "cutoff_s": 5.0, + "replicates_per_level": 3, + "label": "2-of-3 session-disjoint repetitions", + "even_cell_order": ["low1", "high1", "high2", "low2", "low3", "high3"], + "odd_cell_order": ["high1", "low1", "low2", "high2", "high3", "low3"], + "hard_cap_h20_hours": 3.5, + }, + "sanity": { + "red_flags": red_flags, + "n_cells": len(cells), + "n_roles": len(ROLES), + "selected_sets": len(all_hashes), + "distinct_selected_sets": len(set(all_hashes)), + "per_cell_selected_sets_distinct": per_cell_distinct, + "invariants": { + "cells_6": len(cells) == 6, + "roles_7": len(ROLES) == 7, + "band_rows_nonzero": all(stats["rows"] > 0 for stats in band_stats.values()), + "session_bands_disjoint_per_cell": all(per_cell_distinct.values()), + }, + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--source-windows", type=Path, required=True) + parser.add_argument("--source-window-id", default="chat_w20260312_1000") + parser.add_argument("--base-primary-study", type=Path, required=True) + parser.add_argument("--base-tp4-study", type=Path, required=True) + parser.add_argument("--private-root", type=Path, required=True) + parser.add_argument("--public-manifest", type=Path, required=True) + args = parser.parse_args() + + source_window, source_trace = resolve_source_trace( + args.source_windows, args.source_window_id + ) + private_windows, band_stats = materialize_bands( + source_trace, source_window, args.private_root + ) + studies = write_studies( + base_primary=args.base_primary_study, + base_tp4=args.base_tp4_study, + private_windows=private_windows, + private_root=args.private_root, + ) + manifest = build_manifest( + studies=studies, + private_windows=private_windows, + band_stats=band_stats, + source_trace=source_trace, + source_windows=args.source_windows, + source_window_id=args.source_window_id, + ) + atomic_json(args.public_manifest, manifest) + print(json.dumps({ + "status": manifest["status"], + "manifest": str(args.public_manifest), + "sanity": manifest["sanity"], + }, sort_keys=True)) + if manifest["status"] != "PASS": + raise RuntimeError(f"pilot preflight failed: {manifest['sanity']['red_flags']}") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/prepare_pilot_simulator.py b/runs/fidelity-headroom/prepare_pilot_simulator.py new file mode 100644 index 0000000..fc84bde --- /dev/null +++ b/runs/fidelity-headroom/prepare_pilot_simulator.py @@ -0,0 +1,334 @@ +#!/usr/bin/env python3 +"""Prepare exact Frontier fixtures for the P1 primary low/high probes. + +Prompt-bearing band traces remain under ``--private-root``. The emitted +fixtures and public manifest contain token IDs, block IDs, hashes, and +aggregate metadata, but no prompt text. +""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +import subprocess +import sys +from pathlib import Path +from typing import Any + +from transformers import AutoTokenizer + + +HERE = Path(__file__).resolve().parent +AITUNER_ROOT = HERE.parents[1] +sys.path.insert(0, str(HERE)) + +import prepare_pilot as pilot # noqa: E402 + + +PRIMARY_ROLES = ("low1", "high1") + + +def load_module(path: Path): + module_root = str(path.parent.resolve()) + if module_root not in sys.path: + sys.path.insert(0, module_root) + spec = importlib.util.spec_from_file_location("simfid_s2rb_prepare", path) + if spec is None or spec.loader is None: + raise ImportError(path) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def order_hash(values: list[str]) -> str: + return hashlib.sha256("\n".join(values).encode()).hexdigest() + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def git_capture(root: Path, *arguments: str) -> str: + return subprocess.run( + ["git", "-C", str(root), *arguments], + check=True, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + ).stdout + + +def raw_rows(path: Path) -> dict[int, dict[str, Any]]: + result = {} + with path.open(encoding="utf-8") as source: + for index, line in enumerate(source): + if line.strip(): + result[index] = json.loads(line) + return result + + +def selected_hashes( + selected: list[Any], rows: dict[int, dict[str, Any]] +) -> dict[str, str]: + identifiers = [] + arrivals = [] + lengths = [] + for item in selected: + row = rows[item.row_index] + identifiers.append(str(row.get("request_id") or row.get("id") or item.row_index)) + arrivals.append(f"{float(item.timestamp) * 0.1:.12f}") + lengths.append(str(int(item.input_length))) + return { + "request_id_order_sha256": order_hash(identifiers), + "arrival_order_sha256": order_hash(arrivals), + "input_length_order_sha256": order_hash(lengths), + } + + +def kv_blocks(raw_root: Path, cell: str) -> int: + stream = next((raw_root / cell / "opprof").glob("*.jsonl")) + with stream.open(encoding="utf-8") as source: + for line in source: + record = json.loads(line) + if "step_index" in record: + return int(record["kv"]["total_blocks"]) + raise ValueError(f"no Layer-1 record for {cell}") + + +def source_window(windows_path: Path, window_id: str) -> tuple[dict[str, Any], Path]: + return pilot.resolve_source_trace(windows_path, window_id) + + +def prepare(args: argparse.Namespace) -> dict[str, Any]: + simulator = load_module(args.replayserve_root / "tools/simfid_s2rb_prepare.py") + manifest = json.loads(args.pilot_manifest.read_text(encoding="utf-8")) + window, trace = source_window(args.source_windows, args.source_window_id) + if args.band_root is not None: + role_paths = { + role: (args.band_root / f"{role}.jsonl").resolve() + for role in PRIMARY_ROLES + } + band_stats = { + role: manifest["private"]["band_stats"][role] + for role in PRIMARY_ROLES + } + for role, path in role_paths.items(): + if sha256_file(path) != band_stats[role]["sha256"]: + raise ValueError(f"pre-materialized band hash mismatch: {role}") + private_windows = None + else: + private_windows, all_band_stats = pilot.materialize_bands( + trace, window, args.private_root + ) + private_payload = json.loads(private_windows.read_text(encoding="utf-8")) + role_paths = { + item["fidelity_pilot_role"]: ( + private_windows.parent / item["trace_file"] + ).resolve() + for item in private_payload["windows"] + } + band_stats = { + role: all_band_stats[role] + for role in PRIMARY_ROLES + } + + tokenizer = AutoTokenizer.from_pretrained( + args.tokenizer, local_files_only=True, use_fast=True + ) + fixture_root = args.output / "fixtures" + config_root = args.output / "configs" + fixture_root.mkdir(parents=True, exist_ok=True) + config_root.mkdir(parents=True, exist_ok=True) + entries = [] + red_flags = [] + for role in PRIMARY_ROLES: + trace_path = role_paths[role] + retained, trace_stats = simulator.scan_trace(trace_path) + rows = raw_rows(trace_path) + primary_pool = [retained[(index * len(retained)) // 512] for index in range(512)] + selections: dict[str, list[Any]] = {} + selected_union: set[int] = set() + for cell, cell_manifest in sorted(manifest["cells"].items()): + level = "low" if role.startswith("low") else "high" + expected = cell_manifest["targets"][level]["selections"][role] + pool = retained if int(cell_manifest["tp"]) == 4 else primary_pool + selected = [item for item in pool if item.sampling_u <= float(expected["anchor"])] + selections[cell] = selected + selected_union.update(item.row_index for item in selected) + hashes = selected_hashes(selected, rows) + if len(selected) != int(expected["selected_count"]): + red_flags.append(f"selection_count_{cell}_{role}") + for key, value in hashes.items(): + if value != expected[key]: + red_flags.append(f"selection_hash_{cell}_{role}_{key}") + + token_gates, selected_records, block_stats = simulator.tokenize_and_hash( + trace=trace_path, + tokenizer=tokenizer, + retained=retained, + selected_union=selected_union, + ) + if any(gate["status"] != "pass" for gate in token_gates.values()): + red_flags.append(f"token_gate_{role}") + for cell, selected in selections.items(): + cell_manifest = manifest["cells"][cell] + level = "low" if role.startswith("low") else "high" + expected = cell_manifest["targets"][level]["selections"][role] + fixture_id = f"fidelity_p1_{cell}_{role}" + cell_record = { + "cell_id": cell, + "tensor_parallel_size": int(cell_manifest["tp"]), + "max_num_seqs": int(cell_manifest["mns"]), + "store_role": "companion" if int(cell_manifest["tp"]) == 4 else "primary", + "kv_capacity": { + "block_size_tokens": 16, + "num_blocks": kv_blocks(args.phase6_raw_root, cell), + }, + } + probe = { + "probe_index": 0 if role == "low1" else 1, + "sampling_u": float(expected["anchor"]), + } + fixture = simulator.create_fixture( + fixture_root=fixture_root, + fixture_id=fixture_id, + cell=cell_record, + probe=probe, + row_indexes=[item.row_index for item in selected], + meta_by_index={item.row_index: item for item in retained}, + selected_records=selected_records, + ) + config_path = config_root / f"{fixture_id}.json" + config = simulator.build_config( + path=config_path, + cell=cell_record, + mode="frozen-calibrated", + fixture_ids=[fixture_id], + frontier_root=args.frontier_root, + cache_dir=args.cache_dir, + ) + entries.append( + { + "cell": cell, + "role": role, + "level": level, + "anchor": expected["anchor"], + "selected_count": len(selected), + "fixture_id": fixture_id, + "fixture_manifest": str( + (fixture_root / fixture_id / "fixture_manifest.json").resolve() + ), + "frontier_csv": fixture["frontier_csv"]["path"], + "sidecar": fixture["sidecar_jsonl"]["path"], + "config": str(config_path.resolve()), + "calibration_scale": config["calibration"]["a_tp"], + } + ) + if block_stats["selected_union_records"] != len(selected_union): + red_flags.append(f"selected_union_{role}") + if trace_stats["retained_inclusive_0_8192"] < 512: + red_flags.append(f"retained_too_small_{role}") + + selected_counts = [int(entry["selected_count"]) for entry in entries] + calibration = [float(entry["calibration_scale"]) for entry in entries] + result = { + "schema": "fidelity-p1-frontier-prepared-v1", + "status": "PASS" if not red_flags else "STOP", + "source": { + "pilot_manifest": str(args.pilot_manifest.resolve()), + "source_windows": str(args.source_windows.resolve()), + "source_window_id": args.source_window_id, + "source_trace": str(trace.resolve()), + "private_windows": ( + str(private_windows.resolve()) if private_windows is not None else None + ), + "pre_materialized_band_root": ( + str(args.band_root.resolve()) if args.band_root is not None else None + ), + "band_stats": band_stats, + }, + "simulator": { + "replayserve_root": str(args.replayserve_root.resolve()), + "frontier_root": str(args.frontier_root.resolve()), + "tokenizer": str(args.tokenizer.resolve()), + "mode": "frozen-calibrated", + }, + "generator": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": git_capture(AITUNER_ROOT, "rev-parse", "HEAD").strip(), + "aituner_git_status_short": git_capture(AITUNER_ROOT, "status", "--short"), + }, + "entries": entries, + "sanity": { + "red_flags": red_flags, + "n": len(entries), + "selected_count": { + "n": len(selected_counts), + "min": min(selected_counts), + "max": max(selected_counts), + "distinct_n": len(set(selected_counts)), + }, + "calibration_scale": { + "n": len(calibration), + "min": min(calibration), + "max": max(calibration), + "distinct_n": len(set(calibration)), + }, + "invariants": { + "entries_12": len(entries) == 12, + "roles_2": {entry["role"] for entry in entries} == set(PRIMARY_ROLES), + "cells_6": len({entry["cell"] for entry in entries}) == 6, + "selected_nonnegative": all(value > 0 for value in selected_counts), + "per_config_not_identical": len(set(selected_counts)) > 1, + }, + }, + } + args.public_manifest.parent.mkdir(parents=True, exist_ok=True) + args.public_manifest.write_text(json.dumps(result, indent=2, sort_keys=True) + "\n") + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--pilot-manifest", type=Path, required=True) + result.add_argument("--source-windows", type=Path, required=True) + result.add_argument("--source-window-id", required=True) + result.add_argument("--private-root", type=Path, required=True) + result.add_argument("--band-root", type=Path) + result.add_argument("--output", type=Path, required=True) + result.add_argument("--public-manifest", type=Path, required=True) + result.add_argument("--phase6-raw-root", type=Path, required=True) + result.add_argument("--replayserve-root", type=Path, required=True) + result.add_argument("--frontier-root", type=Path, required=True) + result.add_argument("--cache-dir", type=Path, required=True) + result.add_argument("--tokenizer", type=Path, required=True) + return result + + +def main() -> None: + result = prepare(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "entries": len(result["entries"]), + "red_flags": result["sanity"]["red_flags"], + }, + sort_keys=True, + ) + ) + if result["status"] != "PASS": + raise RuntimeError(result["sanity"]["red_flags"]) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/run_pilot_simulator.py b/runs/fidelity-headroom/run_pilot_simulator.py new file mode 100644 index 0000000..d1833ec --- /dev/null +++ b/runs/fidelity-headroom/run_pilot_simulator.py @@ -0,0 +1,303 @@ +#!/usr/bin/env python3 +"""Run and score a frozen Frontier probe manifest, CPU only.""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import os +import subprocess +import sys +import time +from pathlib import Path +from typing import Any + + +def load_module(name: str, path: Path): + module_root = str(path.parent.resolve()) + if module_root not in sys.path: + sys.path.insert(0, module_root) + spec = importlib.util.spec_from_file_location(name, path) + if spec is None or spec.loader is None: + raise ImportError(path) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def git_capture(root: Path, *arguments: str) -> str: + return subprocess.run( + ["git", "-C", str(root), *arguments], + check=True, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + ).stdout + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + args.output = args.output.resolve() + prepared = json.loads(args.prepared_manifest.read_text(encoding="utf-8")) + if prepared["status"] != "PASS": + raise RuntimeError("prepared simulator manifest did not pass") + expected_runs = int(prepared.get("expected_runs", len(prepared["entries"]))) + if expected_runs != len(prepared["entries"]): + raise RuntimeError( + f"prepared manifest expected {expected_runs} runs but contains " + f"{len(prepared['entries'])} entries" + ) + driver = load_module( + "simfid_execution_driver", + args.replayserve_root + / "runs/simfid_s2rb/results/execution_driver.py", + ) + head = git_capture(args.frontier_root, "rev-parse", "HEAD").strip() + status_short = git_capture(args.frontier_root, "status", "--short") + aituner_root = Path(__file__).resolve().parents[2] + aituner_head = git_capture(aituner_root, "rev-parse", "HEAD").strip() + aituner_status_short = git_capture(aituner_root, "status", "--short") + results = [] + failures = [] + gpu_visibility_disabled = True + for sequence, entry in enumerate(prepared["entries"]): + run_root = args.output / f"{sequence:03d}_{entry['fixture_id']}" + scorer_path = run_root / "scorer_output.json" + if scorer_path.is_file() and args.resume: + scorer = json.loads(scorer_path.read_text(encoding="utf-8")) + results.append({**entry, "sequence": sequence, "scorer": scorer, "resumed": True}) + continue + run_root.mkdir(parents=True, exist_ok=True) + config_path = Path(entry["config"]) + config = json.loads(config_path.read_text(encoding="utf-8")) + fixture_manifest_path = Path(entry["fixture_manifest"]) + fixture = json.loads(fixture_manifest_path.read_text(encoding="utf-8")) + trace_path = Path(entry["frontier_csv"]) + sidecar_path = Path(entry["sidecar"]) + metrics_root = run_root / "frontier_metrics" + run_id = f"fidelity_p1_frontier_{sequence:02d}_{entry['cell']}_{entry['role']}" + knobs = config["frontier"]["knobs"] + command = driver.build_command( + trace_path=trace_path, + metrics_root=metrics_root, + run_id=run_id, + knobs=knobs, + ) + driver.audit_command(command, knobs) + row = { + "hook_path": config["calibration"]["hook_path"], + "applied_a_tp": config["calibration"]["a_tp"], + "sidecar_path": str(sidecar_path), + "request_count": int(fixture["request_count"]), + "tensor_parallel_size": int(fixture["tensor_parallel_size"]), + } + environment = driver.environment_for(row) + gpu_visibility_disabled = gpu_visibility_disabled and ( + environment.get("CUDA_VISIBLE_DEVICES") == "" + and environment.get("NVIDIA_VISIBLE_DEVICES") == "void" + ) + run_manifest = { + "schema": "fidelity-p1-frontier-run-v1", + "sequence": sequence, + "cell": entry["cell"], + "role": entry["role"], + "anchor": entry["anchor"], + "request_count": entry["selected_count"], + "frontier": { + "root": str(args.frontier_root.resolve()), + "git_head": head, + "git_status_short": status_short, + }, + "runner": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": aituner_head, + "aituner_git_status_short": aituner_status_short, + }, + "inputs": { + "config": str(config_path), + "config_sha256": sha256_file(config_path), + "fixture_manifest": str(fixture_manifest_path), + "fixture_manifest_sha256": sha256_file(fixture_manifest_path), + "frontier_csv": str(trace_path), + "frontier_csv_sha256": sha256_file(trace_path), + "sidecar": str(sidecar_path), + "sidecar_sha256": sha256_file(sidecar_path), + }, + "environment": { + key: environment[key] + for key in ( + "PYTHONPATH", + "FRONTIER_EXECUTION_TIME_SCALE", + "CUDA_VISIBLE_DEVICES", + "NVIDIA_VISIBLE_DEVICES", + "FRONTIER_LOG_LEVEL", + ) + }, + "command": command, + "contains_prompt_text": False, + } + atomic_json(run_root / "run_manifest.json", run_manifest) + start = time.time() + with (run_root / "stdout.log").open("w", encoding="utf-8") as stdout, ( + run_root / "stderr.log" + ).open("w", encoding="utf-8") as stderr: + try: + process = subprocess.run( + command, + cwd=args.frontier_root, + env=environment, + stdout=stdout, + stderr=stderr, + timeout=args.timeout_s, + ) + return_code = int(process.returncode) + except subprocess.TimeoutExpired: + return_code = 124 + runtime = time.time() - start + if return_code != 0: + failure = { + "sequence": sequence, + "cell": entry["cell"], + "role": entry["role"], + "return_code": return_code, + "runtime_s": runtime, + } + failures.append(failure) + atomic_json(run_root / "failure.json", failure) + break + system_path, request_path = driver.find_metrics(run_root) + scorer = driver.score_trial(row, system_path, request_path) + scorer["runtime_s"] = runtime + atomic_json(scorer_path, scorer) + results.append({**entry, "sequence": sequence, "scorer": scorer, "resumed": False}) + print( + json.dumps( + { + "sequence": sequence, + "cell": entry["cell"], + "role": entry["role"], + "runtime_s": runtime, + "sim_pass_rate": scorer["slo"]["pass_rate"], + "sim_feasible": scorer["slo"]["feasible"], + }, + sort_keys=True, + ), + flush=True, + ) + + pass_rates = [float(item["scorer"]["slo"]["pass_rate"]) for item in results] + throughputs = [ + float(item["scorer"]["throughput_requests_per_second_per_gpu"]) + for item in results + ] + runtimes = [float(item["scorer"]["runtime_s"]) for item in results] + red_flags = [] + if failures: + red_flags.append("frontier_run_failure") + if len(results) != expected_runs: + red_flags.append("runs_not_expected") + if any(not 0.0 <= value <= 1.0 for value in pass_rates): + red_flags.append("pass_rate_out_of_range") + if any(value <= 0 for value in throughputs): + red_flags.append("nonpositive_throughput") + result = { + "schema": "fidelity-p1-frontier-result-v1", + "status": "PASS" if not red_flags else "STOP", + "prepared_manifest": str(args.prepared_manifest.resolve()), + "prepared_manifest_sha256": sha256_file(args.prepared_manifest), + "frontier": { + "root": str(args.frontier_root.resolve()), + "git_head": head, + "git_status_short": status_short, + }, + "runner": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": aituner_head, + "aituner_git_status_short": aituner_status_short, + }, + "results": results, + "failures": failures, + "sanity": { + "red_flags": red_flags, + "n": len(results), + "pass_rate": { + "n": len(pass_rates), + "min": min(pass_rates) if pass_rates else None, + "max": max(pass_rates) if pass_rates else None, + "distinct_n": len(set(pass_rates)), + }, + "throughput_per_gpu": { + "n": len(throughputs), + "min": min(throughputs) if throughputs else None, + "max": max(throughputs) if throughputs else None, + "distinct_n": len(set(throughputs)), + }, + "runtime_s": { + "n": len(runtimes), + "min": min(runtimes) if runtimes else None, + "max": max(runtimes) if runtimes else None, + "distinct_n": len(set(runtimes)), + }, + "invariants": { + "runs_expected": len(results) == expected_runs, + "expected_runs": expected_runs, + "zero_failures": not failures, + "ratios_bounded": all(0.0 <= value <= 1.0 for value in pass_rates), + "nonnegative_metrics": all(value > 0 for value in throughputs), + "per_config_not_identical": len(set(pass_rates)) > 1, + "gpu_visibility_disabled": gpu_visibility_disabled, + }, + }, + } + atomic_json(args.output / "metrics.json", result) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--prepared-manifest", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + result.add_argument("--replayserve-root", type=Path, required=True) + result.add_argument("--frontier-root", type=Path, required=True) + result.add_argument("--timeout-s", type=float, default=900.0) + result.add_argument("--resume", action="store_true") + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "runs": len(result["results"]), + "red_flags": result["sanity"]["red_flags"], + }, + sort_keys=True, + ) + ) + if result["status"] != "PASS": + raise RuntimeError(result["sanity"]["red_flags"]) + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/strong-baseline-metrics.json b/runs/fidelity-headroom/strong-baseline-metrics.json new file mode 100644 index 0000000..bf681ad --- /dev/null +++ b/runs/fidelity-headroom/strong-baseline-metrics.json @@ -0,0 +1,477 @@ +{ + "comparison": "same 5-second prefix, folds, logistic family, regularization, and frozen Frontier outputs; the only nested difference is real Layer-1 engine state", + "decision": { + "contribution_established": false, + "prospective_requirement": "repeat sim+outcome versus sim+outcome+instrumentation on complete held-out tasks" + }, + "features": { + "instrumentation_only": [ + "model_steps_per_second", + "waiting_mean", + "waiting_max", + "waiting_nonzero_share", + "running_mean", + "running_max", + "decode_batch_mean", + "decode_batch_max", + "decode_batch_cv", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "padding_fraction", + "prefill_token_fraction", + "preemptions" + ], + "shared_outcome": [ + "log_offered_rate_per_gpu", + "log2_tp", + "log2_max_num_seqs", + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit" + ], + "shared_simulator": [ + "log_sim_completed_throughput_per_gpu", + "sim_slo_pass_rate", + "sim_slo_feasible" + ] + }, + "headline": { + "group_bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.18181818181818188 + ], + "point": 0.08108108108108103 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.04292727744470806, + 0.019924730979981074 + ], + "point": -0.010145365131402809 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 4, + "instrumentation_only_correct": 3, + "mcnemar_exact_two_sided_p": 0.25, + "sim_outcome_only_correct": 0 + }, + "sim_plus_outcome": { + "classification": { + "accuracy": 0.8108108108108109, + "balanced_accuracy": 0.7242063492063493, + "brier": 0.1058226346682949, + "confusion": { + "false_negative": 3, + "false_positive": 4, + "true_negative": 5, + "true_positive": 25 + }, + "log_loss": 0.3011048455679668 + }, + "policy_0p95": { + "abstain_continue_full": 17, + "correctly_saved_h20_hours": 0.5429431818208333, + "decision_coverage": 0.5405405405405406, + "early_accept": 16, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.5429431818208333, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5088695307144538, + "valid_zero_error_policy": true + } + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8154761904761905, + "brier": 0.0956772695368921, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.288823031828762 + }, + "policy_0p95": { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.7360063646722222, + "decision_coverage": 0.6756756756756757, + "early_accept": 20, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7360063646722222, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6898165884274738, + "valid_zero_error_policy": true + } + } + }, + "headline_regularization": 1.0, + "provenance": { + "frozen_simulator_manifest_scorer_set_sha256": "833842d96ecaa0b059ef99852621752f7989e63d100118b6025425fb119b7a55", + "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", + "phase6_metrics_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", + "phase6_raw_root": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/solo-authoritative/cells", + "simulator_metrics": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json", + "simulator_metrics_sha256": "55edb37d5692e979ab6f6dc6c65913a9db0aa0a836c350e4c05d9c38eee78206", + "simulator_raw_root": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/raw" + }, + "regularization_sensitivity": { + "0.1": { + "group_bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.17500000000000004, + 0.0 + ], + "point": -0.08108108108108103 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.026383192545085435, + 0.0607951286646285 + ], + "point": 0.019228316404518567 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 4, + "instrumentation_only_correct": 0, + "mcnemar_exact_two_sided_p": 0.25, + "sim_outcome_only_correct": 3 + }, + "sim_plus_outcome": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8154761904761905, + "brier": 0.10990776306815446, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.328357763455984 + }, + "policy_0p95": { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.7402314096841667, + "decision_coverage": 0.6756756756756757, + "early_accept": 20, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7402314096841667, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6937764809990414, + "valid_zero_error_policy": true + } + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": { + "accuracy": 0.8108108108108109, + "balanced_accuracy": 0.7619047619047619, + "brier": 0.12913607947267303, + "confusion": { + "false_negative": 4, + "false_positive": 3, + "true_negative": 6, + "true_positive": 24 + }, + "log_loss": 0.4373556318820343 + }, + "policy_0p95": { + "abstain_continue_full": 9, + "correctly_saved_h20_hours": 0.7469523484622221, + "decision_coverage": 0.7567567567567568, + "early_accept": 22, + "early_reject": 6, + "false_accept": 2, + "false_accept_examples": [ + { + "anchor": 0.49609375, + "cell": "tp2_mns8", + "label_feasible": false, + "probability_feasible": 0.9869795738005246, + "remaining_h20_hours": 0.010117910306111111 + }, + { + "anchor": 0.033717411016, + "cell": "tp4_mns16", + "label_feasible": false, + "probability_feasible": 0.9855364057197005, + "remaining_h20_hours": 0.023106262014444445 + } + ], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.03322417232055556, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7801765207827777, + "threshold": 0.95, + "valid_cost_reduction_fraction": null, + "valid_zero_error_policy": false + } + } + }, + "1.0": { + "group_bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + 0.0, + 0.18181818181818188 + ], + "point": 0.08108108108108103 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.04292727744470806, + 0.019924730979981074 + ], + "point": -0.010145365131402809 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 4, + "instrumentation_only_correct": 3, + "mcnemar_exact_two_sided_p": 0.25, + "sim_outcome_only_correct": 0 + }, + "sim_plus_outcome": { + "classification": { + "accuracy": 0.8108108108108109, + "balanced_accuracy": 0.7242063492063493, + "brier": 0.1058226346682949, + "confusion": { + "false_negative": 3, + "false_positive": 4, + "true_negative": 5, + "true_positive": 25 + }, + "log_loss": 0.3011048455679668 + }, + "policy_0p95": { + "abstain_continue_full": 17, + "correctly_saved_h20_hours": 0.5429431818208333, + "decision_coverage": 0.5405405405405406, + "early_accept": 16, + "early_reject": 4, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.5429431818208333, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.5088695307144538, + "valid_zero_error_policy": true + } + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": { + "accuracy": 0.8918918918918919, + "balanced_accuracy": 0.8154761904761905, + "brier": 0.0956772695368921, + "confusion": { + "false_negative": 1, + "false_positive": 3, + "true_negative": 6, + "true_positive": 27 + }, + "log_loss": 0.288823031828762 + }, + "policy_0p95": { + "abstain_continue_full": 12, + "correctly_saved_h20_hours": 0.7360063646722222, + "decision_coverage": 0.6756756756756757, + "early_accept": 20, + "early_reject": 5, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.7360063646722222, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.6898165884274738, + "valid_zero_error_policy": true + } + } + }, + "10.0": { + "group_bootstrap": { + "accuracy_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.13333333333333341, + 0.05555555555555558 + ], + "point": -0.027027027027027084 + }, + "brier_delta_instrumentation_minus_outcome": { + "ci95": [ + -0.03091105649870874, + 0.01684192005239855 + ], + "point": -0.007318433328714388 + }, + "replicates": 10000, + "seed": 20260714, + "semantics": "group bootstrap over cells; diagnostic confidence interval" + }, + "paired_correctness": { + "both_correct": 30, + "both_wrong": 4, + "instrumentation_only_correct": 1, + "mcnemar_exact_two_sided_p": 1.0, + "sim_outcome_only_correct": 2 + }, + "sim_plus_outcome": { + "classification": { + "accuracy": 0.8648648648648649, + "balanced_accuracy": 0.7222222222222222, + "brier": 0.10613344425735322, + "confusion": { + "false_negative": 0, + "false_positive": 5, + "true_negative": 4, + "true_positive": 28 + }, + "log_loss": 0.3404203142465075 + }, + "policy_0p95": { + "abstain_continue_full": 32, + "correctly_saved_h20_hours": 0.21727432337249997, + "decision_coverage": 0.13513513513513514, + "early_accept": 5, + "early_reject": 0, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.21727432337249997, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.20363877229302757, + "valid_zero_error_policy": true + } + }, + "sim_plus_outcome_plus_instrumentation": { + "classification": { + "accuracy": 0.8378378378378378, + "balanced_accuracy": 0.7420634920634921, + "brier": 0.09881501092863883, + "confusion": { + "false_negative": 2, + "false_positive": 4, + "true_negative": 5, + "true_positive": 26 + }, + "log_loss": 0.312914193285738 + }, + "policy_0p95": { + "abstain_continue_full": 30, + "correctly_saved_h20_hours": 0.2384080185036111, + "decision_coverage": 0.1891891891891892, + "early_accept": 6, + "early_reject": 1, + "false_accept": 0, + "false_accept_examples": [], + "false_reject": 0, + "false_reject_examples": [], + "full_trial_h20_hours": 1.0669595034675, + "invalidly_saved_h20_hours": 0.0, + "remaining_h20_hours_at_cutoff": 0.957237281245278, + "saved_h20_hours_if_decisions_used": 0.2384080185036111, + "threshold": 0.95, + "valid_cost_reduction_fraction": 0.22344617366339725, + "valid_zero_error_policy": true + } + } + } + }, + "sanity": { + "examples": { + "distinct_n": 1, + "max": 1.0, + "min": 1.0, + "n": 37 + }, + "frozen_simulator_runs": 92, + "invariants": { + "all_examples_matched_once": true, + "labels_not_identical": true, + "per_config_results_not_all_identical": true, + "same_nested_folds": true, + "simulator_ratios_bounded": true + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 37, + "negative": 9, + "positive": 28 + }, + "matched_simulator_pass_rate": { + "distinct_n": 12, + "max": 1.0, + "min": 0.06884057971014493, + "n": 37 + }, + "red_flags": [] + }, + "schema": "fidelity-strong-baseline-v1", + "scope": "retrospective one-task headroom audit; not contribution evidence", + "status": "PASS" +} diff --git a/runs/fidelity-headroom/test_analysis.py b/runs/fidelity-headroom/test_analysis.py new file mode 100644 index 0000000..434cd37 --- /dev/null +++ b/runs/fidelity-headroom/test_analysis.py @@ -0,0 +1,40 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import math +import sys +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load_analysis(): + spec = importlib.util.spec_from_file_location("fidelity_headroom", HERE / "analyze_existing.py") + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def main() -> None: + analysis = load_analysis() + curve = analysis.topk_curve( + {"a": 3.0, "b": 2.0, "c": 1.0}, + {"a": 1.0, "b": 2.0, "c": 2.0}, + 2e-6, + ) + assert curve["points"][0]["expanded_k"] == 2 + assert curve["points"][0]["candidates"] == ["b", "c"] + assert math.isclose(curve["points"][0]["real_regret"], 1.0 / 3.0) + assert curve["points"][2]["real_regret"] == 0.0 + assert curve["minimum_k"]["five_percent"] == {"nominal_k": 3, "expanded_k": 3} + assert analysis._mcnemar_exact_p(0, 1) == 1.0 + assert analysis._mcnemar_exact_p(0, 5) == 0.0625 + print("fidelity headroom analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/test_pilot_e2e.py b/runs/fidelity-headroom/test_pilot_e2e.py new file mode 100644 index 0000000..e837b42 --- /dev/null +++ b/runs/fidelity-headroom/test_pilot_e2e.py @@ -0,0 +1,50 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +from analyze_pilot_e2e import expanded_top_k, replay + + +def candidate( + cell: str, + sim_score: float, + real_feasible: bool, + probability: float, +) -> dict[str, object]: + return { + "cell": cell, + "level": "high", + "sim_throughput_req_s_per_gpu": sim_score, + "real_goodput_req_s_per_gpu": sim_score, + "real_feasible": real_feasible, + "setup_h20_hours": 0.1, + "full_trial_h20_hours": 0.05, + "prefix_h20_hours": 0.01, + "instrument_probability": probability, + } + + +def main() -> None: + candidates = [ + candidate("a", 3.0, True, 0.5), + candidate("b", 2.0, False, 0.01), + candidate("c", 2.0, True, 0.99), + ] + shortlist = expanded_top_k(candidates, 2) + assert [item["cell"] for item in shortlist] == ["a", "b", "c"] + result = replay( + shortlist, + probability_key="instrument_probability", + oracle_goodput=3.0, + common_failure_h20_hours=0.02, + ) + assert result["selected_cell"] == "a" + assert result["false_accept"] == 0 + assert result["false_reject"] == 0 + assert result["early_accept"] == 1 + assert result["early_reject"] == 1 + assert result["online_h20_hours"] > 0 + print("fidelity pilot e2e: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/test_pilot_tools.py b/runs/fidelity-headroom/test_pilot_tools.py new file mode 100644 index 0000000..46d9cba --- /dev/null +++ b/runs/fidelity-headroom/test_pilot_tools.py @@ -0,0 +1,163 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import math +import sys +import tempfile +from dataclasses import dataclass +from pathlib import Path + + +HERE = Path(__file__).resolve().parent +sys.path.insert(0, str(HERE)) + +import pilot_controller as controller # noqa: E402 +import prepare_pilot as prepare # noqa: E402 +from analyze_pilot import campaign_gpu_accounting # noqa: E402 + + +@dataclass +class Request: + row_id: str + sampling_u: float + arrival_s: float = 0.0 + prompt_tokens_hint: int = 1 + + +def main() -> None: + requests = [ + Request("a", 0.1), + Request("b", 0.2), + Request("c", 0.2), + Request("d", 0.9), + ] + anchor, selected = prepare.attainable_anchor(requests, target_count=2) + assert anchor == 0.2 + assert [request.row_id for request in selected] == ["a", "b", "c"] + + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + source = root / "source.jsonl" + rows = [] + for index, role in enumerate(prepare.ROLES): + rows.append( + { + "request_id": role, + "timestamp": float(index), + "sampling_u": (index + 0.5) / len(prepare.ROLES), + "input_length": 16 + index, + "messages": [{"role": "user", "content": role}], + } + ) + source.write_text( + "".join(json.dumps(row) + "\n" for row in rows), encoding="utf-8" + ) + windows, stats = prepare.materialize_bands( + source, + { + "window_id": "source", + "trace_type": "chat", + "window_start": 0.0, + "window_end": 600.0, + }, + root / "private", + ) + assert windows.is_file() + assert all(stats[role]["rows"] == 1 for role in prepare.ROLES) + for role in prepare.ROLES: + row = json.loads((root / "private" / "traces" / f"{role}.jsonl").read_text()) + assert row["fidelity_pilot_band"] == role + assert abs(float(row["sampling_u"]) - 0.5) < 1e-12 + + prior = root / "prior-state.json" + primary = root / "primary-state.json" + prior.write_text( + json.dumps( + { + "status": "failed", + "gpu_hours_total": 0.02, + "hard_cap_h20_hours": 3.5, + } + ), + encoding="utf-8", + ) + primary.write_text( + json.dumps( + { + "status": "complete", + "gpu_hours_total": 1.5, + "hard_cap_h20_hours": 3.5, + } + ), + encoding="utf-8", + ) + accounting = campaign_gpu_accounting(primary, (prior,)) + assert math.isclose(accounting["aggregate_h20_hours"], 1.52) + assert all(accounting["invariants"].values()) + + assert len(controller.ORDER) == 6 + assert set(controller.ORDER) == set(prepare.CELLS) + assert math.isclose( + sum( + controller.CELL_ESTIMATE_H20_HOURS[int(config["tp"])] + for config in prepare.CELLS.values() + ) + controller.SAFETY_H20_HOURS, + 3.0, + ) + selection = { + "selected_count": 122, + "request_id_order_sha256": "request-hash", + "arrival_order_sha256": "arrival-hash", + "input_length_order_sha256": "length-hash", + } + warmup = { + "kind": "warmup", + "selection": {"count": 16}, + "invariants": { + "warmup_16": True, + "warmup_exact_16": True, + "warmup_long": True, + }, + } + controller.validate_result_selection( + result=warmup, + selection=selection, + cell="tp1_mns8", + role="burnin", + warmup=True, + ) + measured = { + "kind": "anchor", + "selection": { + "count": 122, + "request_id_order_sha256": "request-hash", + "arrival_order_sha256": "arrival-hash", + "raw_length_order_sha256": "length-hash", + }, + "invariants": {}, + } + controller.validate_result_selection( + result=measured, + selection=selection, + cell="tp1_mns8", + role="low1", + warmup=False, + ) + try: + controller.validate_result_selection( + result=warmup, + selection=selection, + cell="tp1_mns8", + role="low1", + warmup=False, + ) + except RuntimeError as error: + assert "selection count mismatch" in str(error) + else: + raise AssertionError("measured selection accepted a warmup subset") + print("fidelity pilot tools: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/test_prefix_analysis.py b/runs/fidelity-headroom/test_prefix_analysis.py new file mode 100644 index 0000000..bc812d6 --- /dev/null +++ b/runs/fidelity-headroom/test_prefix_analysis.py @@ -0,0 +1,72 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import math +import sys +from pathlib import Path + + +HERE = Path(__file__).resolve().parent +sys.path.insert(0, str(HERE)) + +import analyze_prefixes as analysis # noqa: E402 + + +def main() -> None: + exact, exact_source = analysis.completion_elapsed_s( + {"completed_elapsed_s": 7.25} + ) + assert exact == 7.25 and exact_source == "exact_monotonic" + + reconstructed, reconstructed_source = analysis.completion_elapsed_s( + { + "success": True, + "arrival_s": 2.0, + "ttft_ms": 100.0, + "tpot_ms": 10.0, + "completion_tokens": 11, + } + ) + assert math.isclose(reconstructed or 0.0, 2.2) + assert reconstructed_source == "reconstructed_from_latency" + missing, missing_source = analysis.completion_elapsed_s({"success": False}) + assert missing is None and missing_source == "unobserved_failure" + + examples = [ + analysis.PrefixExample( + cell=f"c{index}", + anchor=float(index), + cutoff_s=5.0, + tp=1, + full_elapsed_s=65.0, + feasible=label, + primary_feasible=label, + outcome=(float(index),), + instrumentation=(float(index % 2),), + completion_time_source="exact_monotonic", + ) + for index, label in enumerate((0, 1, 1)) + ] + labels = analysis.np.asarray([0, 1, 1]) + probabilities = analysis.np.asarray([0.01, 0.99, 0.60]) + policy = analysis.policy_metrics(examples, labels, probabilities, 0.95) + assert policy["early_accept"] == 1 + assert policy["early_reject"] == 1 + assert policy["abstain_continue_full"] == 1 + assert policy["false_accept"] == 0 and policy["false_reject"] == 0 + assert policy["valid_zero_error_policy"] + assert policy["valid_cost_reduction_fraction"] is not None + model = analysis.fit_frozen_model( + examples, + instrumentation_aware=True, + regularization=1.0, + ) + frozen_probability = analysis.predict_frozen_model(model, examples) + assert len(frozen_probability) == len(examples) + assert analysis.np.all(frozen_probability >= 0.0) + assert analysis.np.all(frozen_probability <= 1.0) + print("fidelity prefix analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/test_strong_baseline.py b/runs/fidelity-headroom/test_strong_baseline.py new file mode 100644 index 0000000..5c1e9f1 --- /dev/null +++ b/runs/fidelity-headroom/test_strong_baseline.py @@ -0,0 +1,37 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +from pathlib import Path + +from analyze_strong_baseline import analyze + + +ROOT = Path(__file__).resolve().parents[2] +REPLAYSERVE = ROOT.parent / "replayserve" + + +def main() -> None: + result = analyze( + ROOT / "runs/opprof-phase6/phase6/metrics.json", + ROOT / "runs/opprof-phase6/phase6/solo-authoritative/cells", + REPLAYSERVE / "runs/simfid_s2rb/results/raw", + REPLAYSERVE / "runs/simfid_s2rb/results/metrics.json", + ) + assert result["status"] == "PASS", json.dumps(result["sanity"], indent=2) + assert result["sanity"]["frozen_simulator_runs"] == 92 + assert result["sanity"]["labels"]["n"] == 37 + headline = result["headline"] + assert headline["sim_plus_outcome"]["policy_0p95"]["false_accept"] == 0 + assert headline["sim_plus_outcome"]["policy_0p95"]["false_reject"] == 0 + assert ( + headline["sim_plus_outcome_plus_instrumentation"]["policy_0p95"][ + "false_accept" + ] + == 0 + ) + print("fidelity strong baseline: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/fidelity-headroom/test_strong_pilot.py b/runs/fidelity-headroom/test_strong_pilot.py new file mode 100644 index 0000000..c888438 --- /dev/null +++ b/runs/fidelity-headroom/test_strong_pilot.py @@ -0,0 +1,100 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import json +import tempfile +from pathlib import Path + +import numpy as np + +from analyze_prefixes import PrefixExample +from prepare_pilot_simulator import load_module as load_prepare_module +from run_pilot_simulator import load_module as load_run_module +from analyze_strong_pilot import ( + covariate_shift, + fit_model, + load_pilot_simulator, + predict_model, +) + + +def example(index: int) -> PrefixExample: + label = int(index >= 4) + return PrefixExample( + cell=f"cell-{index // 2}", + anchor=float(index), + cutoff_s=5.0, + tp=1, + full_elapsed_s=10.0, + feasible=label, + primary_feasible=label, + outcome=tuple(float(index + offset) for offset in range(13)), + instrumentation=tuple(float(index * offset + 1) for offset in range(17)), + completion_time_source="exact_monotonic", + ) + + +def main() -> None: + examples = [example(index) for index in range(8)] + simulator = [(float(index), index / 10.0, float(index >= 4)) for index in range(8)] + for instrumentation_aware in (False, True): + model = fit_model( + examples, + simulator, + instrumentation_aware=instrumentation_aware, + regularization=1.0, + ) + probability = predict_model(model, examples, simulator) + assert probability.shape == (8,) + assert np.all((probability >= 0.0) & (probability <= 1.0)) + shift = covariate_shift( + examples, + simulator, + examples, + simulator, + instrumentation_aware=instrumentation_aware, + ) + assert shift["values"]["min"] >= 0.0 + assert shift["count_gt_3"] == 0 + + payload = { + "status": "PASS", + "results": [ + { + "cell": f"cell-{index // 2}", + "role": "low1" if index % 2 == 0 else "high1", + "scorer": { + "throughput_requests_per_second_per_gpu": 1.0 + index, + "slo": { + "pass_rate": index / 12.0, + "feasible": index % 2 == 0, + }, + }, + } + for index in range(12) + ], + } + with tempfile.TemporaryDirectory() as temporary: + path = Path(temporary) / "metrics.json" + path.write_text(json.dumps(payload), encoding="utf-8") + features, red_flags = load_pilot_simulator(path) + assert len(features) == 12 + assert red_flags == [] + + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + (root / "prepare_dependency.py").write_text("VALUE = 17\n", encoding="utf-8") + (root / "prepare_target.py").write_text( + "from prepare_dependency import VALUE\n", encoding="utf-8" + ) + assert load_prepare_module(root / "prepare_target.py").VALUE == 17 + (root / "run_dependency.py").write_text("VALUE = 23\n", encoding="utf-8") + (root / "run_target.py").write_text( + "from run_dependency import VALUE\n", encoding="utf-8" + ) + assert load_run_module("run_target", root / "run_target.py").VALUE == 23 + print("fidelity strong pilot: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/.gitignore b/runs/frontier-fidelity-envelope-v1/.gitignore new file mode 100644 index 0000000..4de000f --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/.gitignore @@ -0,0 +1,15 @@ +__pycache__/ +a1-native-smoke/ +a2-measured-fix-smoke/ +simulator-a1/ +simulator-a2/ +simulator-a3/ +simulator-f1/ +simulator-f1-steady/ +simulator-f2/ +simulator-t1/ +fleet-state/ +fleet-artifacts/ +profiles/profile-v3-batch/ +trace-exact-v1-public/ +workloads/f1-isl512-osl1-steady10s/*.csv diff --git a/runs/frontier-fidelity-envelope-v1/analyze_graph_piecewise_surface.py b/runs/frontier-fidelity-envelope-v1/analyze_graph_piecewise_surface.py new file mode 100644 index 0000000..968e8bb --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/analyze_graph_piecewise_surface.py @@ -0,0 +1,272 @@ +#!/usr/bin/env python3 +"""Compare a graph-aligned Frontier Qwen30 surface with audited real vLLM data. + +The input surface may have been produced by independent TP jobs. Therefore +this script deliberately reads the per-cell ``result.json`` files rather than +the runner's root manifest, which is only a convenience artifact and can be +overwritten by concurrent dispatch. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from itertools import combinations +from pathlib import Path +from typing import Any + + +CONFIGS = tuple(f"tp{tp}_mns{mns}" for tp in (1, 2, 4) for mns in (8, 16, 32, 64)) +METRICS = ("ttft_ms", "tpot_ms", "e2e_ms") +STATISTICS = ("mean", "p90") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--sim-root", type=Path, required=True) + parser.add_argument("--real-audit", type=Path, required=True) + parser.add_argument("--json-output", type=Path, required=True) + parser.add_argument("--markdown-output", type=Path, required=True) + return parser.parse_args() + + +def read_json(path: Path) -> dict[str, Any]: + with path.open(encoding="utf-8") as source: + payload = json.load(source) + if not isinstance(payload, dict): + raise ValueError(f"expected JSON object: {path}") + return payload + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def numeric(value: Any, label: str) -> float: + if not isinstance(value, (int, float)) or isinstance(value, bool): + raise ValueError(f"{label} is not numeric: {value!r}") + result = float(value) + if not math.isfinite(result) or result < 0: + raise ValueError(f"{label} is invalid: {value!r}") + return result + + +def parse_config(name: str) -> tuple[int, int]: + try: + tp_part, mns_part = name.split("_", 1) + return int(tp_part.removeprefix("tp")), int(mns_part.removeprefix("mns")) + except ValueError as error: + raise ValueError(f"invalid config name: {name}") from error + + +def load_sim_cell(root: Path, name: str) -> dict[str, Any]: + tp, mns = parse_config(name) + path = root / "runs" / name / f"tp{tp}" / "result.json" + result = read_json(path) + if result.get("status") != "completed": + raise ValueError(f"{path}: status is not completed") + config = result.get("config") + if config != {"tp": tp, "mns": mns, "name": name}: + raise ValueError(f"{path}: config drift: {config!r}") + if result.get("trace_label") != f"tp{tp}": + raise ValueError(f"{path}: trace label does not match TP") + if result.get("request_count") != 129: + raise ValueError(f"{path}: expected 129 requests") + score = result.get("score") + if not isinstance(score, dict): + raise ValueError(f"{path}: missing score") + metrics: dict[str, dict[str, float]] = {} + for metric in METRICS: + prefix = metric.removesuffix("_ms") + metrics[metric] = { + statistic: numeric(score.get(f"{prefix}_{statistic}_ms"), f"{path}:{metric}:{statistic}") + for statistic in STATISTICS + } + return { + "config": name, + "tp": tp, + "mns": mns, + "trace_label": result["trace_label"], + "trace_sha256": result.get("trace_sha256"), + "offered_request_rate": numeric(result.get("offered_request_rate"), f"{path}:rate"), + "offered_request_rate_per_gpu": numeric( + result.get("offered_request_rate_per_gpu"), f"{path}:rate_per_gpu" + ), + "request_count": result["request_count"], + "elapsed_seconds": numeric(result.get("elapsed_seconds"), f"{path}:elapsed"), + "result_path": str(path.resolve()), + "result_sha256": sha256_file(path), + "metrics": metrics, + } + + +def load_real_cell(audit: dict[str, Any], name: str) -> dict[str, dict[str, float]]: + configs = audit.get("configs") + if not isinstance(configs, dict) or name not in configs: + raise ValueError(f"real audit lacks {name}") + result = configs[name] + metrics = result.get("metrics") + if not isinstance(metrics, dict): + raise ValueError(f"real audit {name} lacks metrics") + parsed: dict[str, dict[str, float]] = {} + for metric in METRICS: + values = metrics.get(metric) + if not isinstance(values, dict): + raise ValueError(f"real audit {name} lacks {metric}") + parsed[metric] = { + "mean": numeric(values.get("pooled_mean_ms"), f"real:{name}:{metric}:mean"), + "p90": numeric(values.get("pooled_p90_ms"), f"real:{name}:{metric}:p90"), + } + return parsed + + +def ranking(values: dict[str, float]) -> list[str]: + return [name for name, _ in sorted(values.items(), key=lambda item: (item[1], item[0]))] + + +def pairwise_agreement(sim: dict[str, float], real: dict[str, float]) -> dict[str, int]: + concordant = discordant = ties = 0 + for left, right in combinations(sorted(sim), 2): + sim_delta = sim[left] - sim[right] + real_delta = real[left] - real[right] + if math.isclose(sim_delta, 0.0, abs_tol=1e-9) or math.isclose( + real_delta, 0.0, abs_tol=1e-9 + ): + ties += 1 + elif (sim_delta > 0) == (real_delta > 0): + concordant += 1 + else: + discordant += 1 + return { + "concordant_pairs": concordant, + "discordant_pairs": discordant, + "tied_pairs": ties, + "informative_pairs": concordant + discordant, + } + + +def comparison_summary(cells: dict[str, dict[str, Any]], real_audit: dict[str, Any]) -> dict[str, Any]: + summaries: dict[str, Any] = {} + for metric in METRICS: + for statistic in STATISTICS: + sim_values = { + name: cells[name]["metrics"][metric][f"sim_{statistic}_ms"] + for name in CONFIGS + } + real_values = { + name: load_real_cell(real_audit, name)[metric][statistic] for name in CONFIGS + } + sim_ranking = ranking(sim_values) + real_ranking = ranking(real_values) + pairwise = pairwise_agreement(sim_values, real_values) + summaries[f"{metric}:{statistic}"] = { + "sim_winner": sim_ranking[0], + "real_winner": real_ranking[0], + "winner_match": sim_ranking[0] == real_ranking[0], + "sim_ranking": sim_ranking, + "real_ranking": real_ranking, + **pairwise, + "pairwise_agreement_fraction": ( + pairwise["concordant_pairs"] / pairwise["informative_pairs"] + if pairwise["informative_pairs"] + else None + ), + } + return summaries + + +def render_markdown(payload: dict[str, Any]) -> str: + lines = [ + "# Frontier piecewise graph-profile vs. real vLLM", + "", + "Each cell uses its TP-normalized, 129-request trace. Frontier values are one deterministic simulation; " + "real values pool three fresh-server trials (387 requests/cell).", + "", + "| Config | TTFT sim / real mean (ms) | TPOT sim / real mean (ms) | E2E sim / real mean (ms) |", + "|---|---:|---:|---:|", + ] + for name in CONFIGS: + cell = payload["cells"][name] + entries = [] + for metric in METRICS: + values = cell["metrics"][metric] + entries.append(f"{values['sim_mean_ms']:.1f} / {values['real_mean_ms']:.1f}") + lines.append(f"| {name} | " + " | ".join(entries) + " |") + + lines.extend(["", "## Selection agreement", "", "| Target | Frontier winner | Real winner | Match | Pairwise agreement |", "|---|---|---|---:|---:|"]) + for target, summary in payload["selection"].items(): + agreement = summary["pairwise_agreement_fraction"] + agreement_text = "n/a" if agreement is None else f"{agreement:.1%}" + lines.append( + f"| {target} | {summary['sim_winner']} | {summary['real_winner']} | " + f"{'yes' if summary['winner_match'] else 'no'} | {agreement_text} " + f"({summary['concordant_pairs']}/{summary['informative_pairs']}) |" + ) + lines.append("") + return "\n".join(lines) + + +def main() -> None: + args = parse_args() + sim_root = args.sim_root.resolve() + real_path = args.real_audit.resolve() + real_audit = read_json(real_path) + if set(real_audit.get("configs", {})) != set(CONFIGS): + raise ValueError("real audit config set does not match the 12-cell surface") + + cells: dict[str, dict[str, Any]] = {} + for name in CONFIGS: + sim = load_sim_cell(sim_root, name) + real = load_real_cell(real_audit, name) + metrics = {} + for metric in METRICS: + metrics[metric] = { + f"sim_{statistic}_ms": sim["metrics"][metric][statistic] + for statistic in STATISTICS + } + metrics[metric].update( + { + f"real_{statistic}_ms": real[metric][statistic] + for statistic in STATISTICS + } + ) + metrics[metric].update( + { + f"{statistic}_ratio_sim_over_real": sim["metrics"][metric][statistic] + / real[metric][statistic] + for statistic in STATISTICS + } + ) + cells[name] = {key: value for key, value in sim.items() if key != "metrics"} + cells[name]["metrics"] = metrics + + payload = { + "schema": "frontier-qwen30-piecewise-graph-comparison-v1", + "contract": { + "configs": list(CONFIGS), + "request_count_per_cell": 129, + "real_trials_per_cell": 3, + "trace_policy": "TP-normalized arrivals with full 16-token prefix blocks only", + "graph_semantics": "Frontier piecewise; CUDA_EVENT for prefill/mixed and KERNEL_ONLY for captured decode", + }, + "sim_root": str(sim_root), + "real_audit": str(real_path), + "real_audit_sha256": sha256_file(real_path), + "cells": cells, + "selection": comparison_summary(cells, real_audit), + } + args.json_output.parent.mkdir(parents=True, exist_ok=True) + args.markdown_output.parent.mkdir(parents=True, exist_ok=True) + args.json_output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8") + args.markdown_output.write_text(render_markdown(payload), encoding="utf-8") + print(json.dumps(payload["selection"], indent=2, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/analyze_qwen235_v020_campaign.py b/runs/frontier-fidelity-envelope-v1/analyze_qwen235_v020_campaign.py new file mode 100644 index 0000000..25853e6 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/analyze_qwen235_v020_campaign.py @@ -0,0 +1,128 @@ +#!/usr/bin/env python3 +"""Compare pooled real and Frontier Qwen235 latency-selection surfaces.""" + +from __future__ import annotations + +import argparse +import itertools +import json +import math +from pathlib import Path + + +CONFIGS = ("tp4_ep1_mns64", "tp4_ep1_mns128", "tp8_ep8_mns64", "tp8_ep8_mns128") +CASES = ("fixed-pd", "fixed-po", "trace-pd", "trace-po") + + +def percentile(values: list[float], fraction: float): + if not values: + return None + return sorted(values)[math.ceil(len(values) * fraction) - 1] + + +def summarize(values: list[float]): + if not values: + return {"mean": None, "p90": None} + return {"mean": sum(values) / len(values), "p90": percentile(values, 0.9)} + + +def real_surface(root: Path, case: str): + surface = {} + for config in CONFIGS: + requests = [] + trials = [] + for trial in (1, 2, 3): + path = root / "real" / case / "real" / config / f"trial{trial}" / "results" / "result.json" + payload = json.loads(path.read_text()) + summary = payload["summary"] + if summary["failed"] or summary["completed"] != len(payload["requests"]): + raise ValueError(f"invalid real result: {path}") + if any(not row["success"] for row in payload["requests"]): + raise ValueError(f"failed request: {path}") + requests.extend(payload["requests"]) + trials.append(str(path.resolve())) + ttft = summarize([float(row["ttft_ms"]) for row in requests]) + tpot = summarize([float(row["tpot_ms"]) for row in requests if row["tpot_ms"] is not None]) + e2e = summarize([float(row["e2e_ms"]) for row in requests]) + surface[config] = { + "ttft_mean_ms": ttft["mean"], "ttft_p90_ms": ttft["p90"], + "tpot_mean_ms": tpot["mean"], "tpot_p90_ms": tpot["p90"], + "e2e_mean_ms": e2e["mean"], "e2e_p90_ms": e2e["p90"], + "request_samples": len(requests), "trials": trials, + } + return surface + + +def sim_surface(root: Path, case: str): + payload = json.loads((root / "sim" / case / "frontier_surface.json").read_text()) + surface = {} + for result in payload["results"]: + if result["status"] != "completed": + continue + config = result["config"]["name"] + if config in surface: + raise ValueError(f"duplicate simulator config: {config}") + surface[config] = {key: value for key, value in result["metrics"].items() if key.endswith("_ms")} + if set(surface) != set(CONFIGS): + raise ValueError(f"simulator coverage failure for {case}: {sorted(surface)}") + return surface + + +def compare_metric(real: dict, sim: dict, metric: str): + applicable = [config for config in CONFIGS if real[config][metric] is not None and sim[config][metric] is not None] + real_winner = min(applicable, key=lambda config: (real[config][metric], config)) + sim_winner = min(applicable, key=lambda config: (sim[config][metric], config)) + regret = real[sim_winner][metric] / real[real_winner][metric] - 1 + informative = agreement = 0 + reversals = [] + for left, right in itertools.combinations(applicable, 2): + real_direction = (real[left][metric] > real[right][metric]) - (real[left][metric] < real[right][metric]) + sim_direction = (sim[left][metric] > sim[right][metric]) - (sim[left][metric] < sim[right][metric]) + if real_direction and sim_direction: + informative += 1 + agreement += int(real_direction == sim_direction) + if real_direction != sim_direction: + reversals.append([left, right]) + return { + "real_winner": real_winner, + "sim_winner": sim_winner, + "winner_match": real_winner == sim_winner, + "selected_real_regret": regret, + "informative_pairs": informative, + "pair_direction_agreement": agreement / informative if informative else None, + "reversals": reversals, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--campaign-root", type=Path, required=True) + parser.add_argument("--json-output", type=Path, required=True) + parser.add_argument("--markdown-output", type=Path, required=True) + args = parser.parse_args() + cases = {} + lines = ["# Qwen235 vLLM 0.20 Frontier vs real", "", "| case | metric | Frontier winner | real winner | match | regret | pair agreement |", "|---|---|---|---|---:|---:|---:|"] + for case in CASES: + real = real_surface(args.campaign_root, case) + sim = sim_surface(args.campaign_root, case) + metrics = ["ttft_mean_ms", "ttft_p90_ms", "e2e_mean_ms", "e2e_p90_ms"] + if case.endswith("pd"): + metrics[2:2] = ["tpot_mean_ms", "tpot_p90_ms"] + comparisons = {} + for metric in metrics: + item = compare_metric(real, sim, metric) + comparisons[metric] = item + lines.append( + f"| {case} | {metric} | {item['sim_winner']} | {item['real_winner']} | " + f"{'yes' if item['winner_match'] else 'no'} | {item['selected_real_regret']:.1%} | " + f"{item['pair_direction_agreement']:.1%} |" + ) + cases[case] = {"real": real, "sim": sim, "comparison": comparisons} + payload = {"schema": "qwen235-v020-simulator-real-comparison-v1", "cases": cases} + args.json_output.parent.mkdir(parents=True, exist_ok=True) + args.json_output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + args.markdown_output.write_text("\n".join(lines) + "\n") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/analyze_qwen30_latency_case.py b/runs/frontier-fidelity-envelope-v1/analyze_qwen30_latency_case.py new file mode 100644 index 0000000..2b7e734 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/analyze_qwen30_latency_case.py @@ -0,0 +1,136 @@ +#!/usr/bin/env python3 +"""Compare one complete Qwen30 Frontier latency surface with real vLLM.""" + +from __future__ import annotations + +import argparse +import json +import math +from itertools import combinations +from pathlib import Path +from typing import Any + + +CONFIGS = tuple(f"tp{tp}_mns{mns}" for tp in (1, 2, 4) for mns in (8, 16, 32, 64)) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--sim-root", type=Path, required=True) + parser.add_argument("--real-audit", type=Path, required=True) + parser.add_argument("--json-output", type=Path, required=True) + parser.add_argument("--markdown-output", type=Path, required=True) + return parser.parse_args() + + +def number(value: Any, field: str) -> float: + if not isinstance(value, (int, float)) or isinstance(value, bool): + raise ValueError(f"{field} is not numeric: {value!r}") + result = float(value) + if not math.isfinite(result) or result < 0: + raise ValueError(f"{field} is invalid: {value!r}") + return result + + +def parse_config(name: str) -> tuple[int, int]: + tp, mns = name.split("_", 1) + return int(tp.removeprefix("tp")), int(mns.removeprefix("mns")) + + +def rank(values: dict[str, float]) -> list[str]: + return [name for name, _ in sorted(values.items(), key=lambda item: (item[1], item[0]))] + + +def pairwise(sim: dict[str, float], real: dict[str, float]) -> dict[str, int | float | None]: + concordant = discordant = ties = 0 + for left, right in combinations(CONFIGS, 2): + sim_delta = sim[left] - sim[right] + real_delta = real[left] - real[right] + if math.isclose(sim_delta, 0.0, abs_tol=1e-9) or math.isclose(real_delta, 0.0, abs_tol=1e-9): + ties += 1 + elif (sim_delta > 0) == (real_delta > 0): + concordant += 1 + else: + discordant += 1 + informative = concordant + discordant + return { + "concordant_pairs": concordant, + "discordant_pairs": discordant, + "tied_pairs": ties, + "informative_pairs": informative, + "agreement": concordant / informative if informative else None, + } + + +def main() -> None: + args = parse_args() + root = args.sim_root.resolve() + audit = json.loads(args.real_audit.read_text()) + metrics = tuple(audit.get("applicable_metrics") or []) + if not metrics or set(audit.get("configs") or {}) != set(CONFIGS): + raise ValueError("real audit is incomplete") + expected_requests = int(audit["trace_manifests"]["tp1"]["requests"]) + cells: dict[str, Any] = {} + for name in CONFIGS: + tp, mns = parse_config(name) + path = root / "runs" / name / f"tp{tp}" / "result.json" + result = json.loads(path.read_text()) + if result.get("status") != "completed" or result.get("request_count") != expected_requests: + raise ValueError(f"incomplete simulator result: {path}") + if result.get("config") != {"tp": tp, "mns": mns, "name": name}: + raise ValueError(f"simulator config drift: {path}") + score = result.get("score") + if not isinstance(score, dict): + raise ValueError(f"simulator score missing: {path}") + values: dict[str, Any] = {} + for metric in metrics: + prefix = metric.removesuffix("_ms") + values[metric] = { + statistic: number(score.get(f"{prefix}_{statistic}_ms"), f"{path}:{metric}:{statistic}") + for statistic in ("mean", "p90") + } + cells[name] = {"path": str(path), "metrics": values} + selection: dict[str, Any] = {} + for metric in metrics: + for statistic in ("mean", "p90"): + sim_values = {name: cells[name]["metrics"][metric][statistic] for name in CONFIGS} + real_values = { + name: number( + audit["configs"][name]["metrics"][metric][f"pooled_{statistic}_ms"], + f"real:{name}:{metric}:{statistic}", + ) + for name in CONFIGS + } + sim_ranking, real_ranking = rank(sim_values), rank(real_values) + sim_choice, real_best = sim_ranking[0], real_ranking[0] + selection[f"{metric}:{statistic}"] = { + "sim_winner": sim_choice, + "real_winner": real_best, + "winner_match": sim_choice == real_best, + "selected_config_real_regret": (real_values[sim_choice] - real_values[real_best]) / real_values[real_best], + "sim_ranking": sim_ranking, + "real_ranking": real_ranking, + **pairwise(sim_values, real_values), + } + payload = { + "schema": "qwen30-latency-case-frontier-real-comparison-v1", + "sim_root": str(root), + "real_audit": str(args.real_audit.resolve()), + "prefill_only": bool(audit["prefill_only"]), + "applicable_metrics": metrics, + "selection": selection, + "cells": cells, + } + lines = ["# Qwen30 Frontier vs real latency selection", "", "| Objective | Frontier | Real | Match | Regret | Pairwise |", "|---|---|---|---:|---:|---:|"] + for objective, row in selection.items(): + agreement = "N/A" if row["agreement"] is None else f"{row['agreement']:.1%}" + lines.append(f"| {objective} | {row['sim_winner']} | {row['real_winner']} | {'yes' if row['winner_match'] else 'no'} | {row['selected_config_real_regret']:.1%} | {agreement} |") + args.json_output.parent.mkdir(parents=True, exist_ok=True) + args.markdown_output.parent.mkdir(parents=True, exist_ok=True) + args.json_output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + args.markdown_output.write_text("\n".join(lines) + "\n") + print(json.dumps(selection, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/assemble_qwen235_v020_profiles.py b/runs/frontier-fidelity-envelope-v1/assemble_qwen235_v020_profiles.py new file mode 100644 index 0000000..e9fdb96 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/assemble_qwen235_v020_profiles.py @@ -0,0 +1,93 @@ +#!/usr/bin/env python3 +"""Assemble immutable flat profiles for the Qwen235 Frontier runner.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import shutil +from pathlib import Path + + +MODEL = "Qwen3-235B-A22B" + + +def parse_args(): + parser = argparse.ArgumentParser() + for name in ("cuda_common", "cuda_moe_tp4", "cuda_moe_ep8", "kernel_common", "kernel_moe_tp4", "kernel_moe_ep8"): + parser.add_argument(f"--{name.replace('_', '-')}", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + return parser.parse_args() + + +def digest(path: Path) -> str: + value = hashlib.sha256() + value.update(path.read_bytes()) + return value.hexdigest() + + +def model_file(root: Path, name: str) -> Path: + path = root / "compute" / "h20" / MODEL / name + if not path.is_file(): + raise FileNotFoundError(path) + return path + + +def merge_csv(inputs: list[Path], output: Path) -> None: + fields = None + rows = [] + for path in inputs: + with path.open(newline="") as source: + reader = csv.DictReader(source) + if fields is None: + fields = reader.fieldnames + elif reader.fieldnames != fields: + raise ValueError(f"CSV schema mismatch: {path}") + rows.extend(reader) + if not fields or not rows: + raise ValueError("cannot merge empty profile CSV") + with output.open("w", newline="") as target: + writer = csv.DictWriter(target, fieldnames=fields, lineterminator="\n") + writer.writeheader() + writer.writerows(rows) + + +def main() -> None: + args = parse_args() + output = args.output_root.resolve() + if output.exists(): + raise FileExistsError(output) + output.mkdir(parents=True) + sources = { + "linear_op.csv": model_file(args.cuda_common, "linear_op.csv"), + "attention.csv": model_file(args.cuda_common, "attention.csv"), + "linear_op_kernel_only.csv": model_file(args.kernel_common, "linear_op_kernel_only.csv"), + "attention_kernel_only.csv": model_file(args.kernel_common, "attention_kernel_only.csv"), + } + for name, source in sources.items(): + shutil.copy2(source, output / name) + merge_csv( + [model_file(args.cuda_moe_tp4, "moe.csv"), model_file(args.cuda_moe_ep8, "moe.csv")], + output / "moe.csv", + ) + merge_csv( + [model_file(args.kernel_moe_tp4, "moe_kernel_only.csv"), model_file(args.kernel_moe_ep8, "moe_kernel_only.csv")], + output / "moe_kernel_only.csv", + ) + outputs = {path.name: digest(path) for path in sorted(output.glob("*.csv"))} + manifest = { + "schema": "qwen235-v020-frontier-profile-v1", + "model": MODEL, + "measurement_families": ["CUDA_EVENT", "KERNEL_ONLY"], + "moe_runtime_paths": {"tp4_ep1": "TRITON", "tp8_ep8": "FLASHINFER_CUTLASS"}, + "inputs": {name: str(getattr(args, name).resolve()) for name in ("cuda_common", "cuda_moe_tp4", "cuda_moe_ep8", "kernel_common", "kernel_moe_tp4", "kernel_moe_ep8")}, + "outputs": outputs, + } + (output / "manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps(manifest, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/audit_qwen30_latency_case.py b/runs/frontier-fidelity-envelope-v1/audit_qwen30_latency_case.py new file mode 100644 index 0000000..617b443 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/audit_qwen30_latency_case.py @@ -0,0 +1,193 @@ +#!/usr/bin/env python3 +"""Audit one completed Qwen30 latency-selection real surface. + +This reader intentionally treats a prefill-only case as a four-objective +surface: TTFT/E2E mean and p90. TPOT is omitted rather than coerced to zero. +""" + +from __future__ import annotations + +import argparse +import json +import math +import statistics +from pathlib import Path +from typing import Any + + +CONFIGS = tuple(f"tp{tp}_mns{mns}" for tp in (1, 2, 4) for mns in (8, 16, 32, 64)) +TRIALS = (1, 2, 3) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--case-root", type=Path, required=True) + parser.add_argument( + "--traces-root", + type=Path, + help="Frozen TP-normalized traces; defaults to /traces.", + ) + parser.add_argument("--json-output", type=Path, required=True) + parser.add_argument("--markdown-output", type=Path, required=True) + return parser.parse_args() + + +def nearest_rank(values: list[float], fraction: float) -> float: + if not values: + raise ValueError("cannot calculate a percentile of an empty metric") + return sorted(values)[math.ceil(len(values) * fraction) - 1] + + +def finite(value: Any, field: str) -> float: + if not isinstance(value, (int, float)) or isinstance(value, bool): + raise ValueError(f"{field} is not numeric: {value!r}") + value = float(value) + if not math.isfinite(value) or value < 0: + raise ValueError(f"{field} is invalid: {value!r}") + return value + + +def read_manifest(traces_root: Path, tp: int) -> dict[str, Any]: + path = traces_root / f"tp{tp}" / "public" / "manifest.json" + manifest = json.loads(path.read_text()) + if manifest.get("schema") != "qwen30-latency-case-v1": + raise ValueError(f"unexpected trace schema in {path}") + if manifest.get("tensor_parallel_size") != tp or int(manifest.get("requests", 0)) <= 0: + raise ValueError(f"invalid trace contract in {path}") + outputs = manifest.get("output_tokens") + if not isinstance(outputs, list) or len(outputs) != 1 or int(outputs[0]) <= 0: + raise ValueError(f"non-uniform output contract in {path}") + return manifest + + +def metric_stats(values: list[float]) -> dict[str, float]: + return {"samples": len(values), "mean_ms": statistics.fmean(values), "p90_ms": nearest_rank(values, 0.90)} + + +def validate_trial(path: Path, manifest: dict[str, Any]) -> tuple[dict[str, list[float]], dict[str, Any]]: + payload = json.loads(path.read_text()) + if payload.get("schema") != "qwen30-exact-trace-anchor-v1": + raise ValueError(f"unexpected real result schema in {path}") + contract = payload.get("contract") + summary = payload.get("summary") + records = payload.get("requests") + if not isinstance(contract, dict) or not isinstance(summary, dict) or not isinstance(records, list): + raise ValueError(f"malformed result payload: {path}") + expected = int(manifest["requests"]) + expected_contract = { + "requests": expected, + "requests_file_sha256": manifest["private_jsonl_sha256"], + "row_vector_sha256": manifest["row_vector_sha256"], + "first_arrival_s": manifest["first_arrival_s"], + "last_arrival_s": manifest["last_arrival_s"], + } + for name, wanted in expected_contract.items(): + actual = contract.get(name) + if isinstance(wanted, float): + if not isinstance(actual, (int, float)) or not math.isclose(float(actual), wanted, abs_tol=1e-9): + raise ValueError(f"{path}: contract drift for {name}") + elif actual != wanted: + raise ValueError(f"{path}: contract drift for {name}") + if len(records) != expected or summary.get("completed") != expected or summary.get("failed") != 0: + raise ValueError(f"{path}: incomplete real replay") + + metrics: dict[str, list[float]] = {"ttft_ms": [], "tpot_ms": [], "e2e_ms": []} + indices: set[int] = set() + expected_output = int(manifest["output_tokens"][0]) + for record in records: + if record.get("success") is not True: + raise ValueError(f"{path}: failed request record") + index = record.get("source_index") + if not isinstance(index, int) or index in indices: + raise ValueError(f"{path}: duplicate/non-integer source index") + indices.add(index) + if record.get("actual_input_tokens") != record.get("input_tokens"): + raise ValueError(f"{path}: input usage drift") + if record.get("requested_output_tokens") != expected_output or record.get("actual_output_tokens") != expected_output: + raise ValueError(f"{path}: output usage drift") + for metric in ("ttft_ms", "e2e_ms"): + metrics[metric].append(finite(record.get(metric), metric)) + tpot = record.get("tpot_ms") + if expected_output == 1: + if tpot is not None: + raise ValueError(f"{path}: OSL=1 must report TPOT=null") + else: + metrics["tpot_ms"].append(finite(tpot, "tpot_ms")) + if len(indices) != expected: + raise ValueError(f"{path}: missing source rows") + return metrics, {"result_path": str(path), "requests": expected} + + +def main() -> None: + args = parse_args() + root = args.case_root.resolve() + traces_root = (args.traces_root or root / "traces").resolve() + manifests = {tp: read_manifest(traces_root, tp) for tp in (1, 2, 4)} + prefill_only = int(manifests[1]["output_tokens"][0]) == 1 + if any((int(manifest["output_tokens"][0]) == 1) != prefill_only for manifest in manifests.values()): + raise ValueError("TP-specific output contracts differ") + applicable = ("ttft_ms", "e2e_ms") if prefill_only else ("ttft_ms", "tpot_ms", "e2e_ms") + configs: dict[str, Any] = {} + for name in CONFIGS: + tp = int(name.split("_", 1)[0].removeprefix("tp")) + trial_rows = [] + pooled = {metric: [] for metric in applicable} + for trial in TRIALS: + path = root / "real" / name / f"trial{trial}" / "results" / "result.json" + values, provenance = validate_trial(path, manifests[tp]) + trial_rows.append({ + **provenance, + "trial": trial, + "metrics": {metric: metric_stats(values[metric]) for metric in applicable}, + }) + for metric in applicable: + pooled[metric].extend(values[metric]) + configs[name] = { + "trials": trial_rows, + "metrics": { + metric: { + "pooled_samples": len(pooled[metric]), + "pooled_mean_ms": statistics.fmean(pooled[metric]), + "pooled_p90_ms": nearest_rank(pooled[metric], 0.90), + "trial_mean_of_means_ms": statistics.fmean( + row["metrics"][metric]["mean_ms"] for row in trial_rows + ), + "trial_stddev_of_means_ms": statistics.stdev( + row["metrics"][metric]["mean_ms"] for row in trial_rows + ), + } + for metric in applicable + }, + } + winners: dict[str, Any] = {} + for metric in applicable: + for statistic in ("pooled_mean_ms", "pooled_p90_ms"): + ranked = sorted((row["metrics"][metric][statistic], name) for name, row in configs.items()) + winners[f"{metric}:{statistic}"] = { + "winner": ranked[0][1], + "winner_value_ms": ranked[0][0], + "ranking": [name for _, name in ranked], + } + payload = { + "schema": "qwen30-latency-case-real-audit-v1", + "case_root": str(root), + "traces_root": str(traces_root), + "prefill_only": prefill_only, + "applicable_metrics": list(applicable), + "trace_manifests": {f"tp{tp}": manifests[tp] for tp in manifests}, + "configs": configs, + "winners": winners, + } + lines = ["# Qwen30 real latency case audit", "", f"Prefill-only: `{prefill_only}`.", ""] + lines += ["| Objective | Real winner | Value (ms) |", "|---|---|---:|"] + for objective, winner in winners.items(): + lines.append(f"| {objective} | {winner['winner']} | {winner['winner_value_ms']:.2f} |") + args.json_output.parent.mkdir(parents=True, exist_ok=True) + args.markdown_output.parent.mkdir(parents=True, exist_ok=True) + args.json_output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + args.markdown_output.write_text("\n".join(lines) + "\n") + print(json.dumps(winners, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/experiment-card.md b/runs/frontier-fidelity-envelope-v1/experiment-card.md new file mode 100644 index 0000000..0ef2d9b --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/experiment-card.md @@ -0,0 +1,84 @@ +# EXP-SIMFID-ENVELOPE-V1：Frontier best-effort fidelity envelope + +> **状态：** 已批准，准备执行（2026-07-17）。用户要求先把 simulator 现有能力跑到最好，并同时覆盖 fixed input/output 与真实 trace replay。 + +## Claim 与可证伪假设 + +- **研究问题：** 在不使用被评测 config/workload 的 serving E2E calibration 时，Frontier 的 measured operator/collective profiles 与 scheduler state abstraction，是否足以找到真机上的低-regret config？ +- **H-CC：** Qwen30 prefill-only 的 TP 排序错误主要来自默认 analytical all-reduce；注入同机、同 TP 的 measured collective 后，已知 `2048/1` surface 的 regret 降至不超过 5%，Kendall tau-b 升至至少 0.8。 +- **H-BATCH：** 若 H-CC 不足，错误主要来自 pure-prefill attention 只有 batch=1 profile，而 Frontier 在多请求 batch 上使用没有 coverage 的 `attn_prefill_mixed` 外推；增加 MBT 可达的真实 batch composition 后可恢复排序。 +- **H-STATE：** 若 measured collective 与 batch-composition profile 都不能恢复排序，则缺失量位于 isolated operators 之外的 scheduler-state-conditioned step composition；继续增加静态 kernel rows不是有效修复。 +- **成功门槛：** worst selected-config regret `<=5%`、tie-aware Kendall tau-b `>=0.8`、真机 capacity bracket 不足以反转 top decision，并且没有使用同一 surface 的 E2E scalar calibration。 + +## Simulator ablation（先用已有 ground truth，零新增 GPU 成本） + +| variant | compute profile | collective | 目的 | +|---|---|---|---| +| A0 | vLLM 0.20 frozen profile-v2 | Frontier analytical | 已冻结 baseline | +| A1 | 同 A0 | Frontier 原生 Vidur + measured TP2/TP4 CSV | 检查原生 profile consumption；大 payload fallback 保留并计数 | +| A2 | 同 A0 | measured Vidur，cache miss 直接调用已训练 estimator | 最小 correctness fix；消除 `>100k elements` 静默 analytical fallback | +| A3 | 增加 pure-prefill batch-composition rows | 同 A2 | 检验 batch-composition coverage 是否是剩余误差来源 | + +A1/A2/A3 都重新运行完整 `TP∈{1,2,4} × MNS∈{8,16,32,64}` surface；TP1 无 all-reduce。所有 simulator variants 在查看新增真机 case 前冻结。A2 是单独标注、带单测的 compatibility patch，不与 Frontier upstream 原生能力混写。 + +## Workload matrix + +| ID | workload | arrival / prefix | phase role | +|---|---|---|---| +| F0 | fixed `ISL=2048, OSL=1` | uniform QPS；distinct prefixes | 已有 real ground truth，选择 A0--A3 | +| F1 | fixed `ISL=512, OSL=1` | uniform QPS；distinct prefixes | short-prefill、多请求 batch composition | +| F2 | fixed `ISL=2048, OSL=128` | uniform QPS；distinct prefixes | true prefill+decode mixed serving | +| T1 | `thinking_w20260327_1000` eligible trace | 原 timestamp/order；exact prompt/output/session/hash；prefix on | production joint distribution 与 cache/scheduler feedback | + +T1 只排除已经审计的 72 个超 40,960 context rows 和 6 个 zero-output rows，eligible universe 为 `15,401/15,479`。负载轴只使用 trace 已有、同 session 共享的 `sampling_u`；入选 request 的 arrival、input/output、prompt、hash 和相对次序不变。真实 runtime 设置 `min_tokens=max_tokens=output_length` 且 `ignore_eos=true`，逐请求核对 usage。 + +## 固定系统与 config surface + +| 项目 | 冻结设置 | +|---|---| +| machine | 仅 `dash0`，8×NVIDIA H20 | +| model/runtime | `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`；community vLLM 0.20.0+cu129；BF16 weight/activation/KV | +| simulator | Frontier `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + manifest 中列出的现有 compatibility patches；A2 patch 独立 hash | +| configs | `TP∈{1,2,4} × MNS∈{8,16,32,64}`；DP=PP=EP=1；MBT=8192；block=16 | +| runtime | chunked prefill on；fixed cases prefix off；T1 prefix on；fresh server per `(config, load, round)` | +| score | `capacity(c)=max tested offered req/s with joint SLO pass rate >=0.95`；primary `capacity/actual TP GPUs` | +| SLO | TTFT `<=1000ms + 1000×ISL/8000`；mixed case同时要求 TPOT `<=150ms`；另报告 50/100/180ms sensitivity，不用 sensitivity 改写 primary | + +F0 沿用已经冻结的 rate lattice与两个 fresh-server rounds。F1/F2 先由冻结 simulator 给出 boundary，再加入共同 per-GPU guard anchors，避免只测 simulator 预测附近而漏掉真实最优。每个 boundary anchor 两个 fresh-server rounds，二者都 pass 才算 feasible。 + +T1 保持原 600 秒 arrival window。先在 simulator 上冻结 `sampling_u` bracket；真机只运行 topological guard set `{TP1,TP2,TP4} × {MNS8,MNS32,MNS64}`，若 top set 或 bracket 仍可能被未测 MNS 反转再补 MNS16。每个入选 source-row vector在 real/sim 两侧必须有相同 digest；至少两个 session-hash folds，若本轮只完成一个 window则明确标为 single-window evidence。 + +> **Pre-bracket smoke amendment（2026-07-17，任何 T1 真机结果产生前）：** +> 允许预先运行一个固定的 `TP4/MNS32 × u=0.005` real harness smoke。该点已属于上述 +> guard set，选择不依赖 simulator 输出；在完整 simulator surface 冻结前，性能结果保持 +> sealed，只检查进程退出状态、artifact 完整性、request usage 与 row-vector digest。该 smoke +> 不用于选择 config、移动负载 bracket 或形成 fidelity verdict；full real anchors 仍必须等待 +> simulator bracket 冻结。目的仅是让 600 秒真机 harness 验证与 CPU simulator surface 并行， +> 缩短 critical path。 + +## 诊断与停止规则 + +1. A1 必须报告每次 collective prediction 的 measured-model hit 与 analytical fallback 次数；不能只看最终 rank。 +2. A2 对超过 100k elements 的 payload 必须由单测证明走 estimator；A2 若不改变任何 TP2/TP4 step，立即停止并检查 CLI/config 注入，不进入 GPU。 +3. A3 profile 只覆盖 MBT=8192 可达的 pure-prefill composition：F0 为 `1/2/4 × q2048`，F1 为 `1/2/4/8/16 × q512`，TP1/2/4 分别实测；不做无边界的 profile sweep。 +4. 若 A3 在 F0 仍不能达到 fidelity gate，先采集 `TP1@8, TP2@16, TP4@32` 的 per-step batch/queue/component residual；禁止用 per-TP E2E scale把答案拟合正确。 +5. 只有 best-effort simulator 在 F0 通过或形成可解释、可定位的失败后，才运行 F1/F2/T1 真机；任一 case 的结论不外推到其它 workload。 + +## 预期成本与产物 + +- simulator A1--A3：CPU only，约 1--3 小时总 CPU wall，0 GPU-hour。 +- attention composition profile：3 张 H20 并行，预计 5--10 分钟，`<0.5 H20-GPU-hour`。 +- F1/F2 real boundary：预计合计 12--24 H20-GPU-hours，smoke 后再锁定。 +- T1 real boundary：600 秒原 arrival window使单 anchor较贵；预计 30--60 H20-GPU-hours，必须在 simulator bracket 和一配置 smoke 后重新 echo 精确预算。 +- 产物：variant/profile manifests、full surfaces、anchor-level request metrics、rank/regret/confusion tables、profile-consumption counters，以及 fixed-vs-trace fidelity summary figure。 + +## Benchmark design audit + +| 风险 | 处理 | +|---|---| +| selective benchmarking | 预先冻结 F0/F1/F2/T1，不因结果删除失败 case | +| calibration=evaluation | 禁止使用同一 surface 的 serving E2E scalar；microprofile GPU 成本单独报告 | +| trace filtering | 只做 context/zero-output correctness exclusion和 session-coherent thinning，不按长度筛选 | +| simulator-guided real sampling | 使用共同 guard anchors；未闭合 bracket 不能宣布 top match | +| absolute-vs-rank metric | 同时报绝对 capacity/latency、rank、regret、tau-b、pair direction和 SLO confusion | +| hidden fallback | A1/A2 强制计数 measured-model hit/fallback，并写入 frozen manifest | diff --git a/runs/frontier-fidelity-envelope-v1/extract_qwen235_v020_runtime_contract.py b/runs/frontier-fidelity-envelope-v1/extract_qwen235_v020_runtime_contract.py new file mode 100644 index 0000000..1b7b4ee --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/extract_qwen235_v020_runtime_contract.py @@ -0,0 +1,44 @@ +#!/usr/bin/env python3 +"""Extract simulator-visible graph/KV metadata from Qwen235 server starts.""" + +from __future__ import annotations + +import argparse +import json +import re +from pathlib import Path + + +CONFIGS = ("tp4_ep1_mns64", "tp4_ep1_mns128", "tp8_ep8_mns64", "tp8_ep8_mns128") + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--case-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + configs = {} + for config in CONFIGS: + log = args.case_root / "real" / config / "trial1" / "logs" / "server.log" + text = log.read_text(errors="replace") + token_matches = re.findall(r"GPU KV cache size:\s*([0-9,]+) tokens", text) + capture_matches = re.findall(r"cudagraph_capture_sizes': \[([^]]+)\]", text) + if not token_matches or not capture_matches: + raise ValueError(f"missing runtime metadata in {log}") + tokens = int(token_matches[-1].replace(",", "")) + if tokens % 16: + raise ValueError(f"KV token count is not block aligned: {tokens}") + capture = [int(value.strip()) for value in capture_matches[-1].split(",")] + configs[config] = { + "num_gpu_blocks": tokens // 16, + "capture_sizes": capture, + "source_log": str(log.resolve()), + } + payload = {"schema": "qwen235-v020-runtime-contract-v1", "configs": configs} + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps(payload, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/fixed-pd-pressure-matching-card.md b/runs/frontier-fidelity-envelope-v1/fixed-pd-pressure-matching-card.md new file mode 100644 index 0000000..2efedad --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/fixed-pd-pressure-matching-card.md @@ -0,0 +1,89 @@ +# EXP-SIMFID-Q30-FIXED-PD-PRESSURE: calibrate a nontrivial fixed-shape workload + +> Status: complete (2026-07-19). This is a real-only workload calibration; its +> measurements are not part of a Frontier-versus-vLLM winner comparison. + +## Question + +The former Fixed-PD workload (`2048 -> 128`, 0.215 request/s/GPU) was nearly +single-request at TP4: its real mean E2E was 604 ms at a TP4 cluster rate of +0.86 request/s, or about 0.52 in-flight requests by Little's law. It therefore +does not test the decode batching regime of Trace-PD. + +Can a no-prefix, uniform `4096 -> 256` workload be assigned a fixed offered +rate that reaches the healthy Trace-PD operating point without using a +simulator result for calibration? + +## Controlled calibration + +The reference is the real Trace-PD `TP4/MNS64` cell, chosen before this probe +because it is the real TTFT/E2E winner on that surface: + +| Reference metric | Target | +|---|---:| +| mean TTFT | 245.95 ms | +| mean TPOT | 13.18 ms | +| mean E2E | 44.99 s | +| offered rate | 0.215 request/s/GPU; 0.86 request/s at TP4 | +| in-flight proxy | 38.69 requests | + +The probe fixes Qwen3-30B-A3B BF16, community vLLM 0.20.0, H20, TP4, MNS64, +MBT=8192, chunked prefill, and prefix caching off. An incomplete one-trial +range-finding pilot showed global 4 rps stable and global 8 rps already badly +overloaded; it is not used in the decision. The formal probe therefore compares +per-GPU rates `{1, 1.125, 1.25, 1.5}` (global TP4 rates `{4, 4.5, 5, 6}` +request/s). Every rate has 257 exact `4096 -> 256` requests in each of three +fresh-server trials; rate orders are rotated across trials. The launcher +requires a validated vLLM 0.20 FlashInfer kernel cache; this avoids including +one-off custom-kernel JIT in server startup and does not alter request latency +measurement. + +## Decision rule + +Pool the three trials for each rate. Among rates with every request completed, +choose the rate minimizing the Euclidean distance of the two *relative* errors +for mean TTFT and mean TPOT from the above targets. Report p90 TTFT/TPOT/E2E, +trial variation, and `global_rate * mean_E2E` as an in-flight proxy, but do not +turn any of them into an SLO. The selected rate becomes a frozen workload +contract. The subsequent 12-cell real/simulator surface must be fresh and is +not allowed to reuse this calibration data. + +## Interpretation boundary + +This selects a workload regime, not a simulator parameter and not a serving +capacity point. It may make Fixed-PD more comparable to Trace-PD in batching +pressure, but it deliberately continues to exclude trace-shaped arrivals and +prefix reuse; those are separately evaluated workload dimensions. + +## Result + +All 12 planned cells completed (`257` requests × `3` fresh-server trials per +rate, zero client failures). The raw artifact is +`dash0:/home/admin/cpfs/wjh/aituner/qwen30-fixed-pd-pressure-20260719-r5`; +it records the vLLM version, model/config checksums, H20 inventory, FlashInfer +workspace, and an artifact checksum manifest. Pooled results are: + +| Global / per-GPU rps | TTFT mean / p90 (ms) | TPOT mean / p90 (ms) | E2E mean / p90 (ms) | In-flight proxy | +|---|---:|---:|---:|---:| +| 4 / 1 | 121.53 / 125.12 | 10.75 / 11.83 | 2862.42 / 3138.92 | 11.45 | +| 4.5 / 1.125 | 122.76 / 126.29 | 11.76 / 12.55 | 3122.32 / 3322.26 | 14.05 | +| 5 / 1.25 | 125.09 / 129.66 | 15.16 / 16.58 | 3989.73 / 4354.22 | 19.95 | +| 6 / 1.5 | 129.27 / 134.69 | 21.33 / 23.03 | 5569.14 / 6005.12 | 33.41 | + +The predeclared two-metric rule picks `4.5` global rps (`1.125` rps/GPU), but +that is only the *least mismatched* tested point: TTFT is still 50.1% below +the Trace-PD target (122.76 vs 245.95 ms), while TPOT is 10.7% below it (11.76 +vs 13.18 ms). Raising load to 5 global rps moves TPOT past target (+15.0%) +while TTFT barely changes (125.09 ms); at 6 global rps TPOT is +61.9% while +TTFT remains 47.4% below target. Trial-level mean standard deviations are at +most 0.78 ms for TTFT and 0.34 ms for TPOT, so this is not trial-order noise. + +**Conclusion:** `4096 -> 256` with uniform no-prefix arrivals cannot be called +a pressure-matched Fixed-PD control simply by setting RPS. It has much shorter +request residence time (3.12 s versus 44.99 s) and lower concurrency (14.05 +versus 38.69 by the same Little-law proxy) at its closest point. For the +paper's fixed-versus-trace comparison, retain this case only as a deliberately +different fixed-shape workload; do not interpret any simulator gap as being +caused solely by trace fidelity. A future pressure-matched fixed control must +also increase output length (or otherwise preserve residence time), then repeat +this real-only calibration before reopening the Frontier-versus-vLLM surface. diff --git a/runs/frontier-fidelity-envelope-v1/fleet.toml b/runs/frontier-fidelity-envelope-v1/fleet.toml new file mode 100644 index 0000000..612e4f7 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/fleet.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-fidelity-envelope-v1/fleet-state" +artifacts_dir = "runs/frontier-fidelity-envelope-v1/fleet-artifacts" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 1024 +gpu_free_utilization_pct = 10 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-fidelity-envelope-v1/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/fidelity-envelope-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1" diff --git a/runs/frontier-fidelity-envelope-v1/graph-piecewise-experiment-card.md b/runs/frontier-fidelity-envelope-v1/graph-piecewise-experiment-card.md new file mode 100644 index 0000000..6a1e0e3 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/graph-piecewise-experiment-card.md @@ -0,0 +1,70 @@ +# EXP-SIMFID-Q30-GRAPH-PIECEWISE：graph-compatible kernel-only profile 是否修正 Frontier trace replay？ + +> **状态：** completed（2026-07-17）。本卡是已纠正 prefix-trace contract 后的最小判别实验；不复用此前 `decode_cuda_graph_mode=none` 的数值作 fidelity verdict。 + +## Purpose and hypotheses + +- **Parent claim：** Frontier 是否已经足以为 Qwen3-30B-A3B 的真实 trace serving surface 选择 config。 +- **Question：** 旧 Frontier replay 低估 decode service rate，是否主要是 simulator 使用 `none` 而真机使用 `FULL_AND_PIECEWISE`、并且没有向 Frontier 提供独立 `KERNEL_ONLY` profile family？ +- **G1 (graph-family omission)：** 用同一 vLLM 0.20/FA3/FlashInfer-CUTLASS stack 的 `RecordFunctionTracer` kernel-only measurements，加真实 capture buckets 和 Frontier `piecewise`，会显著缩小 TP2/MNS16 的 TPOT/service-rate gap，并至少改变一个 config 的 latency ranking。 +- **G2 (remaining composition error)：** 即使 graph family 对齐，TPOT、TTFT 或 E2E ranking 仍与真机不一致；则 graph omission 只是必要修正，不是 simulator 已解决 tuning 的证据。 + +## Controlled setup + +| Item | Frozen choice | +|---|---| +| model/runtime/hardware | Qwen3-30B-A3B BF16; community vLLM 0.20.0 (`88d34c…`); dash0 NVIDIA H20 | +| simulator | Frontier `deadc4a321f0baaa534c6ebd17f974123733cdc2`; no local source patch | +| workload | exact 129-request Trace-PD public projection; exact ISL/OSL/arrival order; TP-normalized arrival time and complete 16-token prefix blocks | +| surface | TP in {1,2,4}; MNS in {8,16,32,64}; MBT=8192; prefix/chunked prefill on | +| real graph contract | observed vLLM capture sizes: MNS8=[1,2,4,8,16], MNS16=[1,2,4,8,16,24,32], MNS32=[1,2,4,8,16,24,32,40,48,56,64], MNS64=[1,2,4,8,16,24,32,40,48,56,64,72,80,88,96,104,112,120,128] | +| profile intervention | CUDA-event profile stays frozen for prefill/mixed batches. New `KERNEL_ONLY` linear, FA3 decode + KV-update, MoE, and router rows use Frontier's actual `RecordFunctionTracer` semantics; no relabeling of CUDA-event numbers. | +| exact capacity | per-cell real observed KV block count and capture list; Frontier CPU-overhead model remains disabled on both old/new simulator runs because the intervention is GPU-kernel family only. | + +Frontier source inspection fixes the semantic boundary: `piecewise` emits `PIECEWISE` whenever a capture hits, but the MONOLITHIC predictor selects `KERNEL_ONLY` only when `num_prefill_tokens == 0`. Hence new profile coverage is pure decode only; captured mixed/prefill work continues to consume the existing CUDA-event family. + +## Measurement and decision rule + +- **Primary outputs：** per-config mean/p90 TTFT, TPOT, E2E; ranking for each metric; TP2/MNS16 per-request TPOT gap against the already frozen three-trial real audit. +- **Validity gates：** every kernel CSV hash matches its manifest; every row says `KERNEL_ONLY`; every TP/capture-bucket/KV-context required by the runner is present; command records `piecewise`, per-cell blocks and capture sizes; each simulator cell completes all 129 requests. +- **Decision:** G1 is supported only if the graph-aligned TP2/MNS16 TPOT median moves toward real **and** full-surface rank/error evidence improves. A single-cell timing improvement does not establish tuning sufficiency. If G2 holds, update the research claim to “Frontier has not solved tuning under trace-faithful MoE serving after graph-family alignment,” then profile stage/state composition rather than add arbitrary kernel rows. + +## Result (completed 2026-07-17) + +### Validity gates + +- The fresh profile family is exclusive `KERNEL_ONLY`: linear=57, attention=456, MoE=57 rows. Attention covers every TP in {1,2,4}, all MNS64 capture buckets, and the eight decode KV contexts; the final manifest records Frontier `deadc4a…`. +- All three TP traces contain the same 129 ISL/OSL requests and prompt semantic vector. Their offered rates are 0.215/0.430/0.860 req/s for TP=1/2/4, while the per-GPU rate is fixed. Prefix IDs use `floor(ISL/16)`: 36,321 legal complete blocks, not the invalid 36,443 `ceil` projection. +- All 12 Frontier cells completed 129/129 requests. No result has a Frontier error or scheduler stall. Commands record `piecewise`, the observed capture list, the real KV-block count, frozen CUDA-event prefill/mixed CSVs, and the new kernel-only CSVs. + +### Frontier versus real vLLM selection + +| Target | Frontier winner | Real winner | Pairwise order agreement | +|---|---|---|---:| +| TTFT mean / p90 | TP4/MNS64 / TP4/MNS64 | TP4/MNS64 / TP4/MNS64 | 56/66 (84.8%) / 56/66 (84.8%) | +| TPOT mean / p90 | TP4/MNS8 / TP4/MNS8 | TP4/MNS8 / TP4/MNS8 | 64/66 (97.0%) / 64/66 (97.0%) | +| E2E mean / p90 | TP4/MNS64 / TP4/MNS64 | TP4/MNS64 / TP4/MNS64 | 61/66 (92.4%) / 55/66 (83.3%) | + +Thus, for this one Qwen3-30B-A3B Trace-PD surface, graph-aligned Frontier selects the same argmin config for all six reported latency objectives. This is evidence for *selection fidelity on this surface*, not a general claim about dense models, fixed-shape traffic, prefill-only traffic, another serving engine, or SLO capacity. + +### Absolute latency is still unusable + +The same simulator is not calibrated for absolute latency. Across all 12 cells, sim/real ratios are: TTFT mean 8.13--510.88x (median 38.29x), TPOT mean 4.06--5.72x (median 4.86x), and E2E mean 4.66--13.99x (median 8.24x). At the TP2/MNS16 anchor, Frontier reports mean/p90 TTFT=606,885/1,194,098 ms, TPOT=64.42/66.54 ms, E2E=828,082/1,404,520 ms; real vLLM reports 34,443/77,922 ms, 14.10/16.15 ms, and 82,937/137,726 ms, respectively. + +The TPOT error is service-time error, not merely queueing: Frontier's `tpot_computation` equals its TPOT at the anchor. The large TTFT/E2E error then amplifies through the simulated queue. The profile rows themselves are measured kernel rows, so the next diagnosis must inspect Frontier's layer/operation composition and batching-state model rather than relabeling or adding arbitrary graph kernels. + +### Decision + +G1 is only partially supported: `piecewise` plus a true kernel-only decode family improves the prior `none` smoke's 96.37 ms TPOT median to 65.60 ms, but leaves a large absolute gap. G2 is rejected for the *argmin selection* question on this surface because all six winners match. The defensible claim is therefore: + +> Frontier has solved config selection for this evaluated MoE trace surface, despite failing absolute latency prediction; it has **not** yet been shown to solve the broader tuning problem. + +The next discriminating tests are the remaining planned dense/MoE × fixed-shape/trace-faithful × prefill+decode/prefill-only surfaces. In parallel, diagnose the decode service-time composition error; do not use Frontier's current absolute values for SLO, cost, or capacity claims. + +Artifacts: `/home/admin/cpfs/wjh/aituner/graph-piecewise-qwen30-20260717/full/frozen-kernel-only/manifest.json` and `/home/admin/cpfs/wjh/aituner/graph-piecewise-qwen30-20260717/simulator-piecewise-surface-v2/analysis/{comparison.json,comparison.md}`. + +## Cost and provenance + +- **GPU cost:** three 1-GPU FA3 decode profile shards, plus one 1-GPU linear shard and one 1-GPU MoE/router shard; expected 1.5--3.0 H20-GPU-hours, hard cap 4.0 GPU-hours. +- **CPU cost:** 12 exact-trace simulations, expected 20--45 CPU minutes; a one-cell TP2/MNS16 smoke precedes the full surface. +- **Calibration separation:** kernel microprofiles are independent measurements, never fitted to trace E2E latency. The frozen real trace audit is evaluation only. diff --git a/runs/frontier-fidelity-envelope-v1/graph-piecewise-profile-prototype.svg b/runs/frontier-fidelity-envelope-v1/graph-piecewise-profile-prototype.svg new file mode 100644 index 0000000..3713255 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/graph-piecewise-profile-prototype.svg @@ -0,0 +1,28 @@ + + + SCHEMATIC — no measured data + Does graph-compatible KERNEL_ONLY profiling make Frontier select the real trace-serving configuration? + + A. TP2/MNS16 TPOT prediction + + + latencymeasurement family + + + + nonepiecewisereal + old simG1: moves closer + + + B. Full 12-cell ranking agreement + + + rank errorsimulator variant + + + + nonepiecewisereal rank + G2: stays wrongG1: error falls + + Final figure reports mean/p90 TTFT, TPOT, E2E for the identical 129-request trace, not an SLO-derived proxy. + diff --git a/runs/frontier-fidelity-envelope-v1/jobs_attention_composition.toml b/runs/frontier-fidelity-envelope-v1/jobs_attention_composition.toml new file mode 100644 index 0000000..7e2aa3b --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/jobs_attention_composition.toml @@ -0,0 +1,43 @@ +version = 1 + +[[jobs]] +name = "qwen30-attention-composition-tp1-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_composition.sh" +artifacts = ["artifacts/attention-composition-tp1-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "1" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-composition-tp1-v1" + +[[jobs]] +name = "qwen30-attention-composition-tp2-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_composition.sh" +artifacts = ["artifacts/attention-composition-tp2-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "2" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-composition-tp2-v1" + +[[jobs]] +name = "qwen30-attention-composition-tp4-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_composition.sh" +artifacts = ["artifacts/attention-composition-tp4-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "4" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-composition-tp4-v1" diff --git a/runs/frontier-fidelity-envelope-v1/jobs_attention_long_context.toml b/runs/frontier-fidelity-envelope-v1/jobs_attention_long_context.toml new file mode 100644 index 0000000..ff94406 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/jobs_attention_long_context.toml @@ -0,0 +1,43 @@ +version = 1 + +[[jobs]] +name = "qwen30-attention-long-context-tp1-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_long_context.sh" +artifacts = ["artifacts/attention-long-context-tp1-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "1" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-long-context-tp1-v1" + +[[jobs]] +name = "qwen30-attention-long-context-tp2-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_long_context.sh" +artifacts = ["artifacts/attention-long-context-tp2-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "2" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-long-context-tp2-v1" + +[[jobs]] +name = "qwen30-attention-long-context-tp4-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_long_context.sh" +artifacts = ["artifacts/attention-long-context-tp4-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +TP = "4" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-fidelity-envelope-v1/artifacts/attention-long-context-tp4-v1" diff --git a/runs/frontier-fidelity-envelope-v1/jobs_t1_real_smoke.toml b/runs/frontier-fidelity-envelope-v1/jobs_t1_real_smoke.toml new file mode 100644 index 0000000..d366eb2 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/jobs_t1_real_smoke.toml @@ -0,0 +1,22 @@ +version = 1 + +[[jobs]] +name = "qwen30-t1-exact-real-smoke-tp4-mns32-u0p005-20260717-v2-cache-consistent" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-fidelity-envelope-v1 && timeout --signal=TERM --kill-after=30s 3600 bash run_qwen30_exact_trace_real_anchor.sh" + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +FLASHINFER_WORKSPACE_BASE = "/tmp/wjh" +TP = "4" +MNS = "32" +TRACE_LABEL = "u0p005" +SERVER_PORT = "8124" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/fidelity-envelope-private/trace-exact-v1/real-sealed-prebracket/tp4_mns32/u0p005/round1-retry2-cache-consistent" +REQUESTS_FILE = "/home/admin/cpfs/wjh/aituner/fidelity-envelope-private/trace-exact-v1/private/u0p005/real_requests.jsonl" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-fidelity-envelope-v1/materialize_frontier_allreduce.py b/runs/frontier-fidelity-envelope-v1/materialize_frontier_allreduce.py new file mode 100644 index 0000000..bb57bc0 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/materialize_frontier_allreduce.py @@ -0,0 +1,115 @@ +#!/usr/bin/env python3 +"""Convert frozen vLLM collective measurements to Frontier Vidur CC CSV.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +from pathlib import Path + + +FIELDS = ( + "time_stats.all_reduce.min", + "time_stats.all_reduce.max", + "time_stats.all_reduce.mean", + "time_stats.all_reduce.median", + "time_stats.all_reduce.std", + "rank", + "num_workers", + "size", + "collective", + "devices_per_node", + "max_devices_per_node", +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def convert(input_path: Path, output_path: Path) -> dict[str, object]: + payload = json.loads(input_path.read_text()) + if payload.get("schema_version") != "qwen30_vllm020_allreduce_frozen.v1": + raise ValueError(f"unexpected input schema: {payload.get('schema_version')!r}") + + rows = [] + seen = set() + for source in payload["rows"]: + tp = int(source["tensor_parallel_size"]) + tokens = int(source["num_tokens"]) + key = (tp, tokens) + if key in seen: + raise ValueError(f"duplicate collective row: {key}") + seen.add(key) + if tp not in (2, 4): + raise ValueError(f"unsupported TP: {tp}") + expected_bytes = tokens * int(source["hidden_dim"]) * 2 + if int(source["payload_bytes"]) != expected_bytes: + raise ValueError(f"payload mismatch for {key}") + + # Frontier Vidur consumes only the median target. The raw profiler kept + # per-rank distributions but not aligned per-repeat critical-path + # samples, so do not invent critical-path min/mean/max/std. Repeating + # the observed critical-path median in the unused fields keeps the CSV + # schema explicit without changing the trained target. + median = float(source["critical_path_median_ms"]) + rows.append( + { + "time_stats.all_reduce.min": median, + "time_stats.all_reduce.max": median, + "time_stats.all_reduce.mean": median, + "time_stats.all_reduce.median": median, + "time_stats.all_reduce.std": 0.0, + "rank": 0, + "num_workers": tp, + "size": expected_bytes, + "collective": "all_reduce", + "devices_per_node": tp, + "max_devices_per_node": 8, + } + ) + + expected = {(tp, tokens) for tp in (2, 4) for tokens in (1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192)} + if seen != expected: + raise ValueError(f"collective coverage mismatch: missing={expected - seen}, extra={seen - expected}") + + output_path.parent.mkdir(parents=True, exist_ok=True) + with output_path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=FIELDS, lineterminator="\n") + writer.writeheader() + writer.writerows(sorted(rows, key=lambda row: (row["num_workers"], row["size"]))) + + return { + "schema": "frontier-vidur-allreduce-materialization-v1", + "source": str(input_path.resolve()), + "source_sha256": sha256(input_path), + "output": str(output_path.resolve()), + "output_sha256": sha256(output_path), + "rows": len(rows), + "tp_coverage": [2, 4], + "target": "time_stats.all_reduce.median", + "unused_stat_policy": "repeat critical_path_median; std=0", + "payload_contract": "size=num_tokens*hidden_dim*2_bytes", + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--input", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + args = parser.parse_args() + manifest = convert(args.input, args.output) + args.manifest.parent.mkdir(parents=True, exist_ok=True) + args.manifest.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps(manifest, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/mock-fidelity-envelope.png b/runs/frontier-fidelity-envelope-v1/mock-fidelity-envelope.png new file mode 100644 index 0000000..0e63a4e Binary files /dev/null and b/runs/frontier-fidelity-envelope-v1/mock-fidelity-envelope.png differ diff --git a/runs/frontier-fidelity-envelope-v1/patches/0001-vidur-large-payload-model-prediction.patch b/runs/frontier-fidelity-envelope-v1/patches/0001-vidur-large-payload-model-prediction.patch new file mode 100644 index 0000000..56eac1f --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/patches/0001-vidur-large-payload-model-prediction.patch @@ -0,0 +1,84 @@ +diff --git a/frontier/cc_backend/backends/vidur_cc_backend.py b/frontier/cc_backend/backends/vidur_cc_backend.py +index ca1983a..0c57f05 100644 +--- a/frontier/cc_backend/backends/vidur_cc_backend.py ++++ b/frontier/cc_backend/backends/vidur_cc_backend.py +@@ -882,2 +882,21 @@ class VidurCCBackend(BaseCCBackend): +- # Fallback to analytical if not in cache +- logger.debug(f"num_tokens={num_tokens} not in cache, using analytical fallback") ++ # The precomputed lookup is capped at 100k elements, while realistic ++ # TP payloads are commonly much larger. A cache miss does not mean the ++ # measured-data model is unavailable: predict on demand and memoize the ++ # exact payload instead of silently switching model families. ++ with self._cache_lock: ++ model = self._models.get("all_reduce") ++ if model is not None: ++ features = pd.DataFrame({"num_tokens": [num_tokens]}) ++ result = float(model.predict(features)[0]) ++ with self._cache_lock: ++ self._predictions["all_reduce"][(num_tokens,)] = result ++ logger.debug( ++ f"predict_allreduce: data_size={data_size_bytes}, num_tokens={num_tokens}, " ++ f"result={result:.6f} ms (ML model, on-demand cache miss)" ++ ) ++ return max(0.0, result) ++ ++ logger.debug( ++ f"num_tokens={num_tokens} not in cache and model unavailable, " ++ "using analytical fallback" ++ ) +diff --git a/tests/unit/test_vidur_cc_large_payload.py b/tests/unit/test_vidur_cc_large_payload.py +new file mode 100644 +index 0000000..7e87aa7 +--- /dev/null ++++ b/tests/unit/test_vidur_cc_large_payload.py +@@ -0,0 +1,50 @@ ++from __future__ import annotations ++ ++import threading ++import unittest ++ ++import numpy as np ++ ++from frontier.cc_backend.backends.vidur_cc_backend import VidurCCBackend ++ ++ ++class RecordingModel: ++ def __init__(self, value: float) -> None: ++ self.value = value ++ self.features = [] ++ ++ def predict(self, features): ++ self.features.append(features.copy()) ++ return np.array([self.value]) ++ ++ ++class VidurCCLargePayloadTest(unittest.TestCase): ++ def test_cache_miss_uses_measured_model_and_memoizes(self) -> None: ++ backend = object.__new__(VidurCCBackend) ++ backend._cache_lock = threading.RLock() ++ backend._num_devices = 2 ++ backend._predictions = {"all_reduce": {(100000,): 0.1}} ++ model = RecordingModel(0.321) ++ backend._models = {"all_reduce": model} ++ backend._analytical_fallback_allreduce = lambda *_: self.fail( ++ "analytical fallback must not run when the measured model exists" ++ ) ++ ++ data_size_bytes = 2048 * 2048 * 2 ++ expected_elements = data_size_bytes // 2 ++ first = backend.predict_allreduce(data_size_bytes, num_devices=2) ++ second = backend.predict_allreduce(data_size_bytes, num_devices=2) ++ ++ self.assertEqual(first, 0.321) ++ self.assertEqual(second, 0.321) ++ self.assertEqual(len(model.features), 1) ++ self.assertEqual( ++ int(model.features[0].iloc[0]["num_tokens"]), expected_elements ++ ) ++ self.assertEqual( ++ backend._predictions["all_reduce"][(expected_elements,)], 0.321 ++ ) ++ ++ ++if __name__ == "__main__": ++ unittest.main() diff --git a/runs/frontier-fidelity-envelope-v1/plot_mock_envelope.py b/runs/frontier-fidelity-envelope-v1/plot_mock_envelope.py new file mode 100644 index 0000000..8d21765 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/plot_mock_envelope.py @@ -0,0 +1,75 @@ +#!/usr/bin/env python3 +"""Render the preregistered fidelity-envelope figure prototype.""" + +from pathlib import Path + +import matplotlib.pyplot as plt +import numpy as np + + +OUT = Path(__file__).with_name("mock-fidelity-envelope.png") + + +def main() -> None: + workloads = ["F0\n2048/1", "F1\n512/1", "F2\n2048/128", "T1\nexact trace"] + variants = ["A0 analytical", "A1 native measured", "A2 measured+fix", "A3 +batch profile"] + # Prototype values are deliberately marked as MOCK and encode possible, + # distinguishable outcomes only. They are never read by result analysis. + regret = np.array( + [ + [12.5, 14.0, 18.0, 25.0], + [11.0, 13.0, 16.0, 22.0], + [7.5, 10.0, 12.0, 18.0], + [2.0, 4.0, 7.0, 12.0], + ] + ) + + fig, (ax0, ax1) = plt.subplots( + 1, 2, figsize=(11.5, 4.5), gridspec_kw={"width_ratios": [1.35, 1.0]} + ) + x = np.arange(len(workloads)) + width = 0.19 + colors = ["#7f7f7f", "#4c78a8", "#f58518", "#54a24b"] + for index, (variant, color) in enumerate(zip(variants, colors)): + ax0.bar( + x + (index - 1.5) * width, + regret[index], + width, + label=variant, + color=color, + ) + ax0.axhline(5.0, color="#d62728", linestyle="--", linewidth=1.5, label="5% gate") + ax0.set_xticks(x, workloads) + ax0.set_ylabel("Worst selected-config regret (%)") + ax0.set_title("A. Rank fidelity across workload complexity") + ax0.legend(fontsize=8, ncol=2, frameon=False) + ax0.grid(axis="y", alpha=0.25) + + consumption = np.array( + [ + [0, 0, 0], + [35, 65, 0], + [100, 0, 0], + [100, 0, 100], + ] + ) + bottom = np.zeros(len(variants)) + labels = ["measured collective hit", "analytical fallback", "batch-profile coverage"] + stack_colors = ["#4c78a8", "#e45756", "#54a24b"] + for values, label, color in zip(consumption.T, labels, stack_colors): + ax1.barh(variants, values, left=bottom, label=label, color=color) + bottom += values + ax1.set_xlim(0, 200) + ax1.set_xlabel("Coverage counters (normalized; separate axes by mechanism)") + ax1.set_title("B. Profile consumption, not just final rank") + ax1.grid(axis="x", alpha=0.25) + ax1.legend(fontsize=8, frameon=False, loc="lower right") + + fig.suptitle("MOCK / preregistered layout — values are not experimental results", fontsize=12) + fig.tight_layout() + fig.savefig(OUT, dpi=180, bbox_inches="tight") + print(OUT) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/prepare_exact_trace.py b/runs/frontier-fidelity-envelope-v1/prepare_exact_trace.py new file mode 100644 index 0000000..405376c --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/prepare_exact_trace.py @@ -0,0 +1,302 @@ +#!/usr/bin/env python3 +"""Prepare exact Qwen trace replays with block-16 prefix identities. + +Prompt text is written only below ``private/``. Public manifests and Frontier +fixtures contain lengths, arrivals, session IDs, and deterministic block IDs. +""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +import platform +import socket +import time +from pathlib import Path +from typing import Any + + +CSV_FIELDS = ( + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def threshold_name(value: float) -> str: + return f"u{value:.12g}".replace(".", "p") + + +def token_payload(tokens: list[int]) -> bytes: + return len(tokens).to_bytes(2, "little") + b"".join( + int(token).to_bytes(4, "little", signed=False) for token in tokens + ) + + +def block_identity_records( + token_ids: list[int], block_size: int +) -> list[tuple[int, bytes]]: + """Return parent-sensitive identities and independent collision witnesses.""" + parent = b"FRONTIER_EXACT_TRACE_ROOT" + records = [] + for start in range(0, len(token_ids), block_size): + payload = token_payload(token_ids[start : start + block_size]) + identity_input = parent + b"\0" + payload + parent = hashlib.blake2b(identity_input, digest_size=16).digest() + records.append( + ( + int.from_bytes(parent, "big", signed=False), + hashlib.sha256(identity_input).digest(), + ) + ) + return records + + +def block_identities(token_ids: list[int], block_size: int) -> list[int]: + """Return parent-sensitive identities with the same prefix equivalence as vLLM.""" + return [identity for identity, _ in block_identity_records(token_ids, block_size)] + + +def root_sessions(rows: list[dict[str, Any]]) -> dict[int, int]: + roots: dict[int, int] = {} + for row in rows: + chat_id = int(row["chat_id"]) + parent = int(row["parent_chat_id"]) + roots[chat_id] = chat_id if parent == -1 else roots.get(parent, parent) + return roots + + +def update_digest(digest: Any, values: list[Any]) -> None: + digest.update(json.dumps(values, separators=(",", ":")).encode()) + digest.update(b"\n") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--trace", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument("--served-model-name", default="qwen3-30b-a3b") + parser.add_argument("--sampling-u-max", type=float, action="append", required=True) + parser.add_argument("--max-model-len", type=int, default=40960) + parser.add_argument("--source-block-size", type=int, default=64) + parser.add_argument("--runtime-block-size", type=int, default=16) + parser.add_argument("--batch-size", type=int, default=16) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + thresholds = sorted(set(args.sampling_u_max)) + if not thresholds or thresholds[0] < 0 or thresholds[-1] > 1: + raise ValueError("sampling thresholds must be in [0, 1]") + if args.source_block_size % args.runtime_block_size: + raise ValueError("source block size must be divisible by runtime block size") + if min(args.max_model_len, args.source_block_size, args.runtime_block_size, args.batch_size) <= 0: + raise ValueError("length and batch arguments must be positive") + + import transformers + from transformers import AutoTokenizer + + rows = [json.loads(line) for line in args.trace.open() if line.strip()] + roots = root_sessions(rows) + eligible = [ + (index, row) + for index, row in enumerate(rows) + if int(row["input_length"]) + int(row["output_length"]) <= args.max_model_len + and int(row["output_length"]) > 0 + ] + tokenizer = AutoTokenizer.from_pretrained(args.model, trust_remote_code=True) + prepared: list[dict[str, Any]] = [] + identity_to_digest: dict[int, bytes] = {} + source_to_runtime: dict[int, tuple[int, ...]] = {} + runtime_to_source: dict[tuple[int, ...], int] = {} + identity_collision_count = 0 + source_to_runtime_conflicts = 0 + runtime_to_source_conflicts = 0 + length_mismatches = 0 + source_hash_count_mismatches = 0 + started = time.time() + + for start in range(0, len(eligible), args.batch_size): + batch = eligible[start : start + args.batch_size] + encoded = tokenizer( + [row["prompt"] for _, row in batch], + add_special_tokens=False, + padding=False, + truncation=False, + )["input_ids"] + for (source_index, row), token_ids in zip(batch, encoded, strict=True): + token_ids = [int(token) for token in token_ids] + if len(token_ids) != int(row["input_length"]): + length_mismatches += 1 + source_hashes = [int(value) for value in row["hash_ids"]] + if len(source_hashes) != math.ceil(len(token_ids) / args.source_block_size): + source_hash_count_mismatches += 1 + identity_records = block_identity_records(token_ids, args.runtime_block_size) + runtime_ids = [identity for identity, _ in identity_records] + for runtime_id, witness in identity_records: + previous = identity_to_digest.setdefault(runtime_id, witness) + identity_collision_count += int(previous != witness) + + blocks_per_source = args.source_block_size // args.runtime_block_size + for block_index, source_id in enumerate(source_hashes): + begin = block_index * blocks_per_source + relation = tuple(runtime_ids[begin : begin + blocks_per_source]) + previous_relation = source_to_runtime.setdefault(source_id, relation) + source_to_runtime_conflicts += int(previous_relation != relation) + previous_source = runtime_to_source.setdefault(relation, source_id) + runtime_to_source_conflicts += int(previous_source != source_id) + + prepared.append( + { + "source_index": source_index, + "chat_id": int(row["chat_id"]), + "session_id": roots[int(row["chat_id"])], + "arrival": float(row["timestamp"]), + "input_length": int(row["input_length"]), + "output_length": int(row["output_length"]), + "sampling_u": float(row["sampling_u"]), + "prompt": row["prompt"], + "runtime_block_ids": runtime_ids, + } + ) + + args.output_root.mkdir(parents=True, exist_ok=True) + threshold_manifests = [] + for threshold in thresholds: + name = threshold_name(threshold) + public = args.output_root / "public" / name + private = args.output_root / "private" / name + public.mkdir(parents=True, exist_ok=True) + private.mkdir(parents=True, exist_ok=True) + selected = [row for row in prepared if row["sampling_u"] <= threshold] + frontier_path = public / "frontier.csv" + real_path = private / "real_requests.jsonl" + vector_digest = hashlib.sha256() + with frontier_path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=CSV_FIELDS, lineterminator="\n") + writer.writeheader() + for row in selected: + writer.writerow( + { + "arrived_at": f"{row['arrival']:.12f}", + "num_prefill_tokens": row["input_length"], + "num_decode_tokens": row["output_length"], + "session_id": row["session_id"], + "block_hash_ids": "|".join( + str(value) for value in row["runtime_block_ids"] + ), + } + ) + update_digest( + vector_digest, + [ + row["source_index"], + row["arrival"], + row["input_length"], + row["output_length"], + row["session_id"], + row["runtime_block_ids"], + ], + ) + with real_path.open("w") as output: + for row in selected: + output.write( + json.dumps( + { + "source_index": row["source_index"], + "arrived_at": row["arrival"], + "input_length": row["input_length"], + "output_length": row["output_length"], + "session_id": row["session_id"], + "runtime_block_ids": row["runtime_block_ids"], + "body": { + "model": args.served_model_name, + "prompt": row["prompt"], + "min_tokens": row["output_length"], + "max_tokens": row["output_length"], + "ignore_eos": True, + "stream": True, + }, + }, + separators=(",", ":"), + ) + + "\n" + ) + threshold_manifests.append( + { + "sampling_u_max": threshold, + "selected_requests": len(selected), + "request_rate": len(selected) / 600.0, + "frontier_csv": str(frontier_path.resolve()), + "frontier_csv_sha256": sha256(frontier_path), + "real_requests_private": str(real_path.resolve()), + "real_requests_private_sha256": sha256(real_path), + "row_vector_sha256": vector_digest.hexdigest(), + } + ) + + failures = { + "input_length_mismatches": length_mismatches, + "source_hash_count_mismatches": source_hash_count_mismatches, + "runtime_identity_collisions": identity_collision_count, + "source_to_runtime_relation_conflicts": source_to_runtime_conflicts, + "runtime_to_source_relation_conflicts": runtime_to_source_conflicts, + } + manifest = { + "schema": "qwen30-exact-trace-block16-v1", + "status": "pass" if not any(failures.values()) else "fail", + "execution": { + "host": socket.gethostname(), + "python": platform.python_version(), + "transformers": transformers.__version__, + "tokenizer": type(tokenizer).__name__, + "elapsed_seconds": round(time.time() - started, 3), + }, + "source": { + "trace": str(args.trace.resolve()), + "trace_sha256": sha256(args.trace), + "requests": len(rows), + "eligible_requests": len(eligible), + "model": str(args.model.resolve()), + "max_model_len": args.max_model_len, + }, + "block_contract": { + "source_block_size": args.source_block_size, + "runtime_block_size": args.runtime_block_size, + "identity": "BLAKE2b-128(parent runtime identity, exact token-id block)", + "unique_runtime_identities": len(identity_to_digest), + "unique_source_relations": len(source_to_runtime), + "failures": failures, + }, + "selection_contract": ( + "eligible universe followed by source sampling_u threshold; no length " + "selection or output override; original arrivals/order preserved" + ), + "thresholds": threshold_manifests, + "privacy": "prompt text exists only under private/ and must not be harvested", + } + manifest_path = args.output_root / "public" / "manifest.json" + manifest_path.parent.mkdir(parents=True, exist_ok=True) + manifest_path.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"status": manifest["status"], "failures": failures, "thresholds": threshold_manifests}, sort_keys=True)) + if manifest["status"] != "pass": + raise SystemExit(1) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/prepare_fixed_frontier_traces.py b/runs/frontier-fidelity-envelope-v1/prepare_fixed_frontier_traces.py new file mode 100644 index 0000000..60d62cc --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/prepare_fixed_frontier_traces.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Materialize deterministic, prefix-disjoint fixed-shape Frontier traces.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +from pathlib import Path + + +FIELDS = ( + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", +) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument("--input-tokens", type=int, required=True) + parser.add_argument("--output-tokens", type=int, required=True) + parser.add_argument("--requests", type=int, default=64) + parser.add_argument("--duration-seconds", type=float) + parser.add_argument("--rate", type=float, action="append", required=True) + parser.add_argument("--block-size", type=int, default=16) + return parser.parse_args() + + +def sha256(path: Path) -> str: + return hashlib.sha256(path.read_bytes()).hexdigest() + + +def rate_key(rate: float) -> str: + return f"r{rate:g}".replace(".", "p") + + +def materialize( + root: Path, + *, + input_tokens: int, + output_tokens: int, + requests: int, + rates: list[float], + block_size: int, + duration_seconds: float | None = None, +) -> dict[str, object]: + values = [input_tokens, output_tokens, requests, block_size] + if any(value <= 0 for value in values): + raise ValueError("token counts, requests, and block size must be positive") + if input_tokens + output_tokens > 40960: + raise ValueError("shape exceeds the frozen max model length") + if not rates or any(rate <= 0 or not math.isfinite(rate) for rate in rates): + raise ValueError("rates must be positive and finite") + if len(set(rates)) != len(rates): + raise ValueError("rates must be unique") + if duration_seconds is not None and duration_seconds <= 0: + raise ValueError("duration must be positive") + root.mkdir(parents=True, exist_ok=True) + anchors = [] + blocks = math.ceil(input_tokens / block_size) + for rate in rates: + request_count = max( + requests, + math.ceil(rate * duration_seconds) if duration_seconds is not None else 0, + ) + path = root / f"{rate_key(rate)}.csv" + with path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=FIELDS, lineterminator="\n") + writer.writeheader() + for request_id in range(request_count): + first_block = request_id * blocks + 1 + writer.writerow( + { + "arrived_at": f"{request_id / rate:.12f}", + "num_prefill_tokens": input_tokens, + "num_decode_tokens": output_tokens, + "session_id": request_id, + "block_hash_ids": "|".join( + str(first_block + offset) for offset in range(blocks) + ), + } + ) + anchors.append( + { + "label": rate_key(rate), + "rate": rate, + "requests": request_count, + "path": path.name, + "sha256": sha256(path), + } + ) + manifest = { + "schema": "qwen30-fixed-frontier-traces-v1", + "contract": { + "input_tokens": input_tokens, + "output_tokens": output_tokens, + "minimum_requests_per_anchor": requests, + "minimum_steady_duration_seconds": duration_seconds, + "block_size": block_size, + "arrival": "open_loop_uniform", + "prefix_caching": False, + "prefix_relation": "all request block identities are disjoint", + }, + "anchors": anchors, + } + (root / "manifest.json").write_text( + json.dumps(manifest, indent=2, sort_keys=True) + "\n" + ) + return manifest + + +def main() -> None: + args = parse_args() + manifest = materialize( + args.output_root.resolve(), + input_tokens=args.input_tokens, + output_tokens=args.output_tokens, + requests=args.requests, + rates=args.rate, + block_size=args.block_size, + duration_seconds=args.duration_seconds, + ) + print(json.dumps(manifest, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/prepare_qwen30_latency_case.py b/runs/frontier-fidelity-envelope-v1/prepare_qwen30_latency_case.py new file mode 100644 index 0000000..e4bfa6a --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/prepare_qwen30_latency_case.py @@ -0,0 +1,245 @@ +#!/usr/bin/env python3 +"""Materialize Qwen30 Fixed/Trace latency cases without changing their contract. + +The private request file deliberately contains either an exact source prompt +(trace cases) or deterministic token IDs (fixed cases). The public Frontier +fixture contains only arrivals, shapes, sessions, and legal complete prefix +block identities. +""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +from pathlib import Path +from typing import Any + + +FIELDS = ( + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def row_digest(rows: list[dict[str, Any]]) -> str: + digest = hashlib.sha256() + for row in rows: + digest.update( + json.dumps( + [ + row["source_index"], + row["arrived_at"], + row["input_length"], + row["output_length"], + row["session_id"], + row["runtime_block_ids"], + ], + separators=(",", ":"), + ).encode() + ) + digest.update(b"\n") + return digest.hexdigest() + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + sub = parser.add_subparsers(dest="kind", required=True) + + trace = sub.add_parser("trace") + trace.add_argument("--base-public", type=Path, required=True) + trace.add_argument("--base-private", type=Path, required=True) + trace.add_argument("--output-tokens", type=int, required=True) + trace.add_argument("--tp", type=int, required=True) + trace.add_argument("--output-root", type=Path, required=True) + + fixed = sub.add_parser("fixed") + fixed.add_argument("--model", type=Path, required=True) + fixed.add_argument("--input-tokens", type=int, required=True) + fixed.add_argument("--output-tokens", type=int, required=True) + fixed.add_argument("--requests", type=int, required=True) + fixed.add_argument("--per-gpu-rate", type=float, required=True) + fixed.add_argument("--tp", type=int, required=True) + fixed.add_argument("--output-root", type=Path, required=True) + return parser.parse_args() + + +def materialize_trace(args: argparse.Namespace) -> tuple[list[dict[str, Any]], bool, str, float]: + with args.base_public.open(newline="") as source: + public = list(csv.DictReader(source)) + private = [json.loads(line) for line in args.base_private.open() if line.strip()] + if not public or len(public) != len(private): + raise ValueError("trace public/private request count mismatch") + rows: list[dict[str, Any]] = [] + for public_row, private_row in zip(public, private, strict=True): + if int(public_row["num_prefill_tokens"]) != int(private_row["input_length"]): + raise ValueError("trace input-length drift") + if float(public_row["arrived_at"]) != float(private_row["arrived_at"]): + raise ValueError("trace arrival drift") + # The real vLLM request artifact retains its final partial prompt block + # for provenance, while the Frontier CSV deliberately projects only + # legal complete cache blocks. Use the latter as the shared + # simulator-facing identity vector; vLLM itself derives cache hashes + # from the exact private prompt text. + runtime_ids = [int(value) for value in public_row["block_hash_ids"].split("|") if value] + if len(runtime_ids) != int(private_row["input_length"]) // 16: + raise ValueError("public trace exposes an incomplete runtime prefix block") + if int(public_row["session_id"]) != int(private_row["session_id"]): + raise ValueError("trace session-id drift") + body = dict(private_row["body"]) + body.update( + { + "min_tokens": args.output_tokens, + "max_tokens": args.output_tokens, + "ignore_eos": True, + } + ) + rows.append( + { + "source_index": int(private_row["source_index"]), + "arrived_at": float(private_row["arrived_at"]), + "input_length": int(private_row["input_length"]), + "output_length": args.output_tokens, + "session_id": int(private_row["session_id"]), + "runtime_block_ids": runtime_ids, + "body": body, + } + ) + return ( + rows, + True, + "trace-derived: exact input/arrival/session/prefix; output override only", + len(rows) / 600.0 * args.tp, + ) + + +def materialize_fixed(args: argparse.Namespace) -> tuple[list[dict[str, Any]], bool, str, float]: + if min(args.input_tokens, args.output_tokens, args.requests, args.tp) <= 0: + raise ValueError("fixed dimensions must be positive") + if args.input_tokens + args.output_tokens > 40960: + raise ValueError("fixed shape exceeds max model length") + if not math.isfinite(args.per_gpu_rate) or args.per_gpu_rate <= 0: + raise ValueError("per-GPU rate must be positive") + from transformers import AutoTokenizer + + tokenizer = AutoTokenizer.from_pretrained(args.model, trust_remote_code=True) + candidates = [ + token for token in range(tokenizer.vocab_size) if token not in set(tokenizer.all_special_ids) + ] + if len(candidates) < args.requests + 1: + raise ValueError("tokenizer does not provide enough non-special tokens") + rate = args.per_gpu_rate * args.tp + base = candidates[0] + rows = [] + for index in range(args.requests): + rows.append( + { + "source_index": index, + "arrived_at": index / rate, + "input_length": args.input_tokens, + "output_length": args.output_tokens, + "session_id": index, + "runtime_block_ids": [], + "body": { + "prompt": [candidates[index + 1], *([base] * (args.input_tokens - 1))], + "min_tokens": args.output_tokens, + "max_tokens": args.output_tokens, + "ignore_eos": True, + }, + } + ) + return ( + rows, + False, + "fixed-shape: deterministic token IDs, uniform TP-normalized QPS, no prefix reuse", + rate, + ) + + +def write_case( + rows: list[dict[str, Any]], + *, + prefix_caching: bool, + description: str, + global_rate: float, + tp: int, + root: Path, +) -> None: + if not rows: + raise ValueError("empty case") + root = root.resolve() + public_root = root / "public" + private_root = root / "private" + public_root.mkdir(parents=True, exist_ok=True) + private_root.mkdir(parents=True, exist_ok=True) + public_path = public_root / "frontier.csv" + private_path = private_root / "real_requests.jsonl" + with public_path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=FIELDS, lineterminator="\n") + writer.writeheader() + for row in rows: + writer.writerow( + { + "arrived_at": f"{row['arrived_at']:.12f}", + "num_prefill_tokens": row["input_length"], + "num_decode_tokens": row["output_length"], + "session_id": row["session_id"], + "block_hash_ids": "|".join(str(value) for value in row["runtime_block_ids"]), + } + ) + with private_path.open("w") as output: + for row in rows: + output.write(json.dumps(row, separators=(",", ":")) + "\n") + arrivals = [float(row["arrived_at"]) for row in rows] + manifest = { + "schema": "qwen30-latency-case-v1", + "description": description, + "tensor_parallel_size": tp, + "requests": len(rows), + "output_tokens": sorted({int(row["output_length"]) for row in rows}), + "prefix_caching": prefix_caching, + "public_csv": str(public_path), + "public_csv_sha256": sha256(public_path), + "private_jsonl": str(private_path), + "private_jsonl_sha256": sha256(private_path), + "row_vector_sha256": row_digest(rows), + "first_arrival_s": arrivals[0], + "last_arrival_s": arrivals[-1], + "global_offered_request_rate": global_rate, + "per_gpu_offered_request_rate": global_rate / tp, + } + (public_root / "manifest.json").write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps(manifest, sort_keys=True)) + + +def main() -> None: + args = parse_args() + if args.kind == "trace": + rows, prefix, description, global_rate = materialize_trace(args) + else: + rows, prefix, description, global_rate = materialize_fixed(args) + write_case( + rows, + prefix_caching=prefix, + description=description, + global_rate=global_rate, + tp=args.tp, + root=args.output_root, + ) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv b/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv new file mode 100644 index 0000000..85fedab --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv @@ -0,0 +1,25 @@ +time_stats.all_reduce.min,time_stats.all_reduce.max,time_stats.all_reduce.mean,time_stats.all_reduce.median,time_stats.all_reduce.std,rank,num_workers,size,collective,devices_per_node,max_devices_per_node +0.08288000151515007,0.08288000151515007,0.08288000151515007,0.08288000151515007,0.0,0,2,4096,all_reduce,2,8 +0.0793600007891655,0.0793600007891655,0.0793600007891655,0.0793600007891655,0.0,0,2,32768,all_reduce,2,8 +0.0713919997215271,0.0713919997215271,0.0713919997215271,0.0713919997215271,0.0,0,2,65536,all_reduce,2,8 +0.08056000247597694,0.08056000247597694,0.08056000247597694,0.08056000247597694,0.0,0,2,131072,all_reduce,2,8 +0.0865279994904995,0.0865279994904995,0.0865279994904995,0.0865279994904995,0.0,0,2,262144,all_reduce,2,8 +0.07135999947786331,0.07135999947786331,0.07135999947786331,0.07135999947786331,0.0,0,2,524288,all_reduce,2,8 +0.07321599870920181,0.07321599870920181,0.07321599870920181,0.07321599870920181,0.0,0,2,1048576,all_reduce,2,8 +0.09025600180029869,0.09025600180029869,0.09025600180029869,0.09025600180029869,0.0,0,2,2097152,all_reduce,2,8 +0.08083200082182884,0.08083200082182884,0.08083200082182884,0.08083200082182884,0.0,0,2,4194304,all_reduce,2,8 +0.10891199856996536,0.10891199856996536,0.10891199856996536,0.10891199856996536,0.0,0,2,8388608,all_reduce,2,8 +0.1703840047121048,0.1703840047121048,0.1703840047121048,0.1703840047121048,0.0,0,2,16777216,all_reduce,2,8 +0.25539200007915497,0.25539200007915497,0.25539200007915497,0.25539200007915497,0.0,0,2,33554432,all_reduce,2,8 +0.1021759994328022,0.1021759994328022,0.1021759994328022,0.1021759994328022,0.0,0,4,4096,all_reduce,4,8 +0.12694399803876877,0.12694399803876877,0.12694399803876877,0.12694399803876877,0.0,0,4,32768,all_reduce,4,8 +0.09161599725484848,0.09161599725484848,0.09161599725484848,0.09161599725484848,0.0,0,4,65536,all_reduce,4,8 +0.08580800145864487,0.08580800145864487,0.08580800145864487,0.08580800145864487,0.0,0,4,131072,all_reduce,4,8 +0.09867199882864952,0.09867199882864952,0.09867199882864952,0.09867199882864952,0.0,0,4,262144,all_reduce,4,8 +0.09646400064229965,0.09646400064229965,0.09646400064229965,0.09646400064229965,0.0,0,4,524288,all_reduce,4,8 +0.08377600088715553,0.08377600088715553,0.08377600088715553,0.08377600088715553,0.0,0,4,1048576,all_reduce,4,8 +0.1128000020980835,0.1128000020980835,0.1128000020980835,0.1128000020980835,0.0,0,4,2097152,all_reduce,4,8 +0.08755199983716011,0.08755199983716011,0.08755199983716011,0.08755199983716011,0.0,0,4,4194304,all_reduce,4,8 +0.12361599877476692,0.12361599877476692,0.12361599877476692,0.12361599877476692,0.0,0,4,8388608,all_reduce,4,8 +0.20030399411916733,0.20030399411916733,0.20030399411916733,0.20030399411916733,0.0,0,4,16777216,all_reduce,4,8 +0.2924960106611252,0.2924960106611252,0.2924960106611252,0.2924960106611252,0.0,0,4,33554432,all_reduce,4,8 diff --git a/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.manifest.json b/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.manifest.json new file mode 100644 index 0000000..6ea2791 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.manifest.json @@ -0,0 +1,15 @@ +{ + "output": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv", + "output_sha256": "9d693fd406616b599e57bcde66c980c7fc2831b3acf37d3eb633cec80ea0070d", + "payload_contract": "size=num_tokens*hidden_dim*2_bytes", + "rows": 24, + "schema": "frontier-vidur-allreduce-materialization-v1", + "source": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/allreduce.json", + "source_sha256": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "target": "time_stats.all_reduce.median", + "tp_coverage": [ + 2, + 4 + ], + "unused_stat_policy": "repeat critical_path_median; std=0" +} diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/allreduce.json b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/allreduce.json new file mode 100644 index 0000000..aae1f0a --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/allreduce.json @@ -0,0 +1,824 @@ +{ + "environment": [ + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + } + ], + "frontier_consumption": "diagnostic_only_in_base_profile_only_run; measured lookup requires a separate CC-backend injection ablation", + "rows": [ + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08288000151515007, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.4872959852218628, + "mean": 0.1310015980154276, + "median": 0.07679999992251396, + "min": 0.06217600032687187, + "std": 0.12790721677293843 + }, + { + "max": 0.4402880072593689, + "mean": 0.12842560112476348, + "median": 0.08288000151515007, + "min": 0.0655680000782013, + "std": 0.11220176524616535 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0793600007891655, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.12716799974441528, + "mean": 0.07871360033750534, + "median": 0.0759200006723404, + "min": 0.06032000109553337, + "std": 0.019331314939874535 + }, + { + "max": 0.12380799651145935, + "mean": 0.08059840016067028, + "median": 0.0793600007891655, + "min": 0.06217600032687187, + "std": 0.01687088356254092 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0713919997215271, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12697599828243256, + "mean": 0.0767391998320818, + "median": 0.07078400254249573, + "min": 0.05910399928689003, + "std": 0.018775178979463278 + }, + { + "max": 0.11430399864912033, + "mean": 0.07594559974968433, + "median": 0.0713919997215271, + "min": 0.06124800071120262, + "std": 0.015719922150631036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08056000247597694, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.1037760004401207, + "mean": 0.07954559996724128, + "median": 0.08056000247597694, + "min": 0.05955199897289276, + "std": 0.0135697420393132 + }, + { + "max": 0.10608000308275223, + "mean": 0.07971520014107228, + "median": 0.07593599706888199, + "min": 0.06028800085186958, + "std": 0.015319849772775456 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0865279994904995, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.14470399916172028, + "mean": 0.09121599942445754, + "median": 0.0865279994904995, + "min": 0.06441599875688553, + "std": 0.024893837894277456 + }, + { + "max": 0.12438400089740753, + "mean": 0.08531199917197227, + "median": 0.08031999692320824, + "min": 0.06364800035953522, + "std": 0.01878029830059533 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07135999947786331, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.11753600090742111, + "mean": 0.07606079950928687, + "median": 0.07135999947786331, + "min": 0.05843200162053108, + "std": 0.01755519771639284 + }, + { + "max": 0.1103999987244606, + "mean": 0.07607359997928143, + "median": 0.07073600217700005, + "min": 0.05721599981188774, + "std": 0.016904445220949783 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07321599870920181, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.11740799993276596, + "mean": 0.07749119997024537, + "median": 0.07203200086951256, + "min": 0.05862399935722351, + "std": 0.017701381594822835 + }, + { + "max": 0.11382400244474411, + "mean": 0.07733759954571724, + "median": 0.07321599870920181, + "min": 0.059039998799562454, + "std": 0.017188890882557036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09025600180029869, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.13980799913406372, + "mean": 0.0950367994606495, + "median": 0.09025600180029869, + "min": 0.06815999746322632, + "std": 0.022829835127539378 + }, + { + "max": 0.14764800667762756, + "mean": 0.09710080176591873, + "median": 0.08720000088214874, + "min": 0.07152000069618225, + "std": 0.02500574954063789 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08083200082182884, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.15574400126934052, + "mean": 0.08760640025138855, + "median": 0.07846399769186974, + "min": 0.07097599655389786, + "std": 0.024487311423551025 + }, + { + "max": 0.16284799575805664, + "mean": 0.08963519930839539, + "median": 0.08083200082182884, + "min": 0.07100799679756165, + "std": 0.026047320562445356 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.10891199856996536, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.1582079976797104, + "mean": 0.11541439890861512, + "median": 0.10836799815297127, + "min": 0.09062399715185165, + "std": 0.018285136316576037 + }, + { + "max": 0.1578879952430725, + "mean": 0.11537599861621857, + "median": 0.10891199856996536, + "min": 0.0960640013217926, + "std": 0.018246538626977286 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1703840047121048, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.19327999651432037, + "mean": 0.1707327976822853, + "median": 0.1703840047121048, + "min": 0.14815999567508698, + "std": 0.014211056022719618 + }, + { + "max": 0.19276799261569977, + "mean": 0.1658592015504837, + "median": 0.16379200667142868, + "min": 0.14560000598430634, + "std": 0.013840249648693638 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.25539200007915497, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.2807359993457794, + "mean": 0.25750079900026324, + "median": 0.25539200007915497, + "min": 0.24624000489711761, + "std": 0.008925204570802302 + }, + { + "max": 0.2863999903202057, + "mean": 0.2585055992007256, + "median": 0.255280002951622, + "min": 0.24371199309825897, + "std": 0.012059738582656496 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1021759994328022, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.9770879745483398, + "mean": 0.1974783968180418, + "median": 0.10099200159311295, + "min": 0.05913599953055382, + "std": 0.2660581738745569 + }, + { + "max": 0.892799973487854, + "mean": 0.18164799660444259, + "median": 0.1021759994328022, + "min": 0.06435199826955795, + "std": 0.23947520188197013 + }, + { + "max": 0.6467199921607971, + "mean": 0.15839359983801843, + "median": 0.10100800171494484, + "min": 0.06800000369548798, + "std": 0.16617013141866102 + }, + { + "max": 0.6725760102272034, + "mean": 0.15686400160193442, + "median": 0.10044800117611885, + "min": 0.06063999980688095, + "std": 0.17523222161300497 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12694399803876877, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.6659520268440247, + "mean": 0.1927648030221462, + "median": 0.12694399803876877, + "min": 0.07609599828720093, + "std": 0.16697784531907736 + }, + { + "max": 0.695360004901886, + "mean": 0.19356480240821838, + "median": 0.11726400256156921, + "min": 0.0796160027384758, + "std": 0.17588096678867862 + }, + { + "max": 0.5939840078353882, + "mean": 0.1868800014257431, + "median": 0.12379200011491776, + "min": 0.07427199929952621, + "std": 0.1463231714943902 + }, + { + "max": 0.6635839939117432, + "mean": 0.1874335989356041, + "median": 0.12014400213956833, + "min": 0.07526399940252304, + "std": 0.16680959677760304 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09161599725484848, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12665599584579468, + "mean": 0.09443839862942696, + "median": 0.09124799817800522, + "min": 0.06431999802589417, + "std": 0.0205690155775906 + }, + { + "max": 0.1303039938211441, + "mean": 0.09712959825992584, + "median": 0.09161599725484848, + "min": 0.07648000121116638, + "std": 0.019072048129173236 + }, + { + "max": 0.13836799561977386, + "mean": 0.09821119979023933, + "median": 0.09148800000548363, + "min": 0.0727040022611618, + "std": 0.021314066545189116 + }, + { + "max": 0.12992000579833984, + "mean": 0.09272959977388381, + "median": 0.08931199833750725, + "min": 0.06406400352716446, + "std": 0.021360803830354765 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08580800145864487, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.13449600338935852, + "mean": 0.08960640132427215, + "median": 0.08299200236797333, + "min": 0.06796800345182419, + "std": 0.020633597898445998 + }, + { + "max": 0.14735999703407288, + "mean": 0.09248319901525974, + "median": 0.08580800145864487, + "min": 0.05913599953055382, + "std": 0.025960234928829564 + }, + { + "max": 0.13705599308013916, + "mean": 0.08947199806571007, + "median": 0.08460799977183342, + "min": 0.0634239986538887, + "std": 0.02122838946992339 + }, + { + "max": 0.13846400380134583, + "mean": 0.0859104000031948, + "median": 0.08308799937367439, + "min": 0.05974400043487549, + "std": 0.022373631424433445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09867199882864952, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.1279039978981018, + "mean": 0.09748799875378608, + "median": 0.09678399935364723, + "min": 0.0655359998345375, + "std": 0.02315719144614622 + }, + { + "max": 0.1356479972600937, + "mean": 0.10018239840865135, + "median": 0.09532799944281578, + "min": 0.06185600161552429, + "std": 0.02330700253650042 + }, + { + "max": 0.13142399489879608, + "mean": 0.09778879955410957, + "median": 0.09492799639701843, + "min": 0.06560000032186508, + "std": 0.02307579212430045 + }, + { + "max": 0.1276479959487915, + "mean": 0.09611519873142242, + "median": 0.09867199882864952, + "min": 0.0642239972949028, + "std": 0.02273612181973431 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09646400064229965, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.14060799777507782, + "mean": 0.09770880043506622, + "median": 0.09115200117230415, + "min": 0.06950400024652481, + "std": 0.024752645089849798 + }, + { + "max": 0.14377599954605103, + "mean": 0.09824960008263588, + "median": 0.08999999985098839, + "min": 0.07103999704122543, + "std": 0.025403407389046027 + }, + { + "max": 0.13680000603199005, + "mean": 0.09993600100278854, + "median": 0.09646400064229965, + "min": 0.06790400296449661, + "std": 0.022297424273985882 + }, + { + "max": 0.1391039937734604, + "mean": 0.09769919961690902, + "median": 0.09601600095629692, + "min": 0.06835199892520905, + "std": 0.023899922100804445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08377600088715553, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.1464959979057312, + "mean": 0.09276160076260567, + "median": 0.08377600088715553, + "min": 0.06777600198984146, + "std": 0.024751086465295658 + }, + { + "max": 0.14319999516010284, + "mean": 0.09080640003085136, + "median": 0.080400001257658, + "min": 0.06796800345182419, + "std": 0.023365718881708488 + }, + { + "max": 0.1382399946451187, + "mean": 0.09063360020518303, + "median": 0.08193599805235863, + "min": 0.06627199798822403, + "std": 0.02372618650854302 + }, + { + "max": 0.14313599467277527, + "mean": 0.09044799953699112, + "median": 0.08128000050783157, + "min": 0.06652799993753433, + "std": 0.02462486862033686 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1128000020980835, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.1600320041179657, + "mean": 0.11426239982247352, + "median": 0.1128000020980835, + "min": 0.07657600194215775, + "std": 0.030240087702350687 + }, + { + "max": 0.15881599485874176, + "mean": 0.11206399947404862, + "median": 0.10628800094127655, + "min": 0.0772159993648529, + "std": 0.029933135345483627 + }, + { + "max": 0.15612800419330597, + "mean": 0.10761600062251091, + "median": 0.09860799834132195, + "min": 0.07689599692821503, + "std": 0.02634237020678647 + }, + { + "max": 0.15865600109100342, + "mean": 0.11094079986214637, + "median": 0.10979199782013893, + "min": 0.07583999633789062, + "std": 0.028619217028542445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08755199983716011, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.11999999731779099, + "mean": 0.0843871995806694, + "median": 0.08755199983716011, + "min": 0.06332799792289734, + "std": 0.016638543890716024 + }, + { + "max": 0.1218239963054657, + "mean": 0.08518079966306687, + "median": 0.08032000064849854, + "min": 0.06393600255250931, + "std": 0.019900899429956945 + }, + { + "max": 0.11849600076675415, + "mean": 0.0843968003988266, + "median": 0.08702399954199791, + "min": 0.06297600269317627, + "std": 0.017031908773433545 + }, + { + "max": 0.12300799787044525, + "mean": 0.0846304003149271, + "median": 0.08139199763536453, + "min": 0.06195199862122536, + "std": 0.020106523698622265 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12361599877476692, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.17155200242996216, + "mean": 0.12168959975242614, + "median": 0.12055999785661697, + "min": 0.09609600156545639, + "std": 0.022565485532483928 + }, + { + "max": 0.9246399998664856, + "mean": 0.19978560134768486, + "median": 0.12361599877476692, + "min": 0.09715200215578079, + "std": 0.24230694305662265 + }, + { + "max": 0.9317439794540405, + "mean": 0.20037759989500045, + "median": 0.12327999994158745, + "min": 0.09603200107812881, + "std": 0.24450903278000383 + }, + { + "max": 0.9321280121803284, + "mean": 0.19875840097665787, + "median": 0.12230399996042252, + "min": 0.0950080007314682, + "std": 0.24519252220785093 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.20030399411916733, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.3261440098285675, + "mean": 0.20826880186796187, + "median": 0.19974400103092194, + "min": 0.1409280002117157, + "std": 0.051397264855718945 + }, + { + "max": 0.3248000144958496, + "mean": 0.20548800230026246, + "median": 0.1979840025305748, + "min": 0.141184002161026, + "std": 0.04980002399656717 + }, + { + "max": 0.32547199726104736, + "mean": 0.21280319690704347, + "median": 0.20030399411916733, + "min": 0.14127999544143677, + "std": 0.05304243085685509 + }, + { + "max": 0.26047998666763306, + "mean": 0.1969312012195587, + "median": 0.18079999834299088, + "min": 0.14057600498199463, + "std": 0.042174123638424224 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.2924960106611252, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.45052799582481384, + "mean": 0.3013375997543335, + "median": 0.28273600339889526, + "min": 0.21334399282932281, + "std": 0.07214223764527236 + }, + { + "max": 0.4466240108013153, + "mean": 0.2968191936612129, + "median": 0.27796798944473267, + "min": 0.21241599321365356, + "std": 0.07400678240849741 + }, + { + "max": 0.3830080032348633, + "mean": 0.2946112036705017, + "median": 0.2924960106611252, + "min": 0.21084800362586975, + "std": 0.05323627615746332 + }, + { + "max": 0.4609600007534027, + "mean": 0.3054272010922432, + "median": 0.289792001247406, + "min": 0.21062399446964264, + "std": 0.07608482904865672 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + } + ], + "schema_version": "qwen30_vllm020_allreduce_frozen.v1" +} diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention.csv new file mode 100644 index 0000000..cf785bf --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention.csv @@ -0,0 +1,151 @@ +time_stats.attn_input_reshape.min,time_stats.attn_input_reshape.max,time_stats.attn_input_reshape.mean,time_stats.attn_input_reshape.median,time_stats.attn_input_reshape.std,time_stats.attn_kv_cache_save.min,time_stats.attn_kv_cache_save.max,time_stats.attn_kv_cache_save.mean,time_stats.attn_kv_cache_save.median,time_stats.attn_kv_cache_save.std,time_stats.attn_prefill.min,time_stats.attn_prefill.max,time_stats.attn_prefill.mean,time_stats.attn_prefill.median,time_stats.attn_prefill.std,time_stats.attn_decode.min,time_stats.attn_decode.max,time_stats.attn_decode.mean,time_stats.attn_decode.median,time_stats.attn_decode.std,time_stats.attn_output_reshape.min,time_stats.attn_output_reshape.max,time_stats.attn_output_reshape.mean,time_stats.attn_output_reshape.median,time_stats.attn_output_reshape.std,n_embd,n_q_head,n_kv_head,block_size,num_tensor_parallel_workers,max_model_len,batch_size,prefill_chunk_size,kv_cache_size,is_prefill,attention_backend,is_mixed_batch,mode,seq_lens,total_tokens,max_seq_len,min_seq_len,avg_seq_len,equal_seq_len,seq_len_variance,seq_len_std,seq_len_cv,is_chunked_prefill_sample,chunk_start_token,chunk_end_token,total_prefill_tokens,profiling_precision,model_arch,quant_signature,measurement_type,is_true_mixed_batch,prefill_seq_lens,prefill_kv_cache_sizes,decode_kv_cache_sizes,num_prefill_seqs,num_decode_seqs,decode_batch_size,total_batch_size,total_decode_tokens,decode_avg_kv_cache_size,batch_composition_ratio,batch_spec,projection_policy +0.0,0.0,0.0,0.0,0.0,0.01414399966597557,0.028863999992609024,0.019705599918961526,0.01771199982613325,0.005157200849836681,0.047968000173568726,0.07046400010585785,0.05810240097343922,0.05810240097343922,0.007477463486041561,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04947200044989586,0.020412799902260303,0.01635199971497059,0.010107497379722417,0.046560000628232956,0.08323200047016144,0.05587520003318787,0.05587520003318787,0.011126758739503428,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.022207999601960182,0.017033600155264138,0.015312000177800655,0.002819235991970241,0.05104000121355057,0.07692799717187881,0.056396800279617305,0.056396800279617305,0.007481982178637539,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.022272000089287758,0.01706880023702979,0.01616000011563301,0.002460889579319197,0.06931199878454208,0.0838719978928566,0.07432000041007995,0.07432000041007995,0.004777766433175866,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.018592000007629395,0.028543999418616295,0.02095999978482723,0.019183999858796597,0.003198175496053494,0.12179200351238251,0.15408000349998474,0.1307712011039257,0.1307712011039257,0.00858807797538298,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027775999158620834,0.03385600075125694,0.030131200328469276,0.029680000618100166,0.0021152558103575215,0.32678401470184326,0.3450239896774292,0.33396480381488797,0.33396480381488797,0.0045872424917606375,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04438399896025658,0.05084799975156784,0.046540799736976626,0.04531199857592583,0.002277905811237223,1.0959680080413818,1.1151360273361206,1.0999775886535645,1.0999775886535645,0.005694403246120485,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.078015998005867,0.08691199868917465,0.08114239946007729,0.08019199967384338,0.00292795706334475,4.070400238037109,4.113152027130127,4.087088012695312,4.087088012695312,0.013660567012509554,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.05167999863624573,0.022115200012922286,0.017583999782800674,0.010340094822340818,0.05196800082921982,0.09011200070381165,0.06328320093452933,0.06328320093452933,0.012557341255467452,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01583999954164028,0.026623999699950218,0.018927999772131443,0.017376000061631203,0.003514650316260619,0.06681600213050842,0.07993599772453308,0.0725280001759529,0.0725280001759529,0.004343502558613716,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01648000068962574,0.030880000442266464,0.01945280022919178,0.017967999912798405,0.004096211183007485,0.1311360001564026,0.1546880006790161,0.13908160030841826,0.13908160030841826,0.007511906874366178,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.03558399900794029,0.020851199887692927,0.018559999763965607,0.005235911594130716,0.32950401306152344,0.350271999835968,0.33912960588932034,0.33912960588932034,0.006027400986663648,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.019328000023961067,0.040608000010252,0.022790400311350822,0.020704000256955624,0.006113051965778337,1.1415679454803467,1.1518720388412476,1.144483208656311,1.144483208656311,0.0032332311374389127,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015807999297976494,0.030688000842928886,0.019596799835562707,0.01774400006979704,0.004343384771033462,0.0,0.0,0.0,0.0,0.0,0.049056001007556915,0.07580800354480743,0.05948160067200661,0.05948160067200661,0.009031541471446955,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.02486399933695793,0.01923839971423149,0.018079999834299088,0.0032282528537266424,0.0,0.0,0.0,0.0,0.0,0.05142400041222572,0.07353600114583969,0.059328000620007516,0.059328000620007516,0.0073307807735143084,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016543999314308167,0.03977600112557411,0.021379199624061585,0.018511999398469925,0.006593576176246171,0.0,0.0,0.0,0.0,0.0,0.0488319993019104,0.06435199826955795,0.05479039996862411,0.05479039996862411,0.005672522998491864,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01635199971497059,0.02844800055027008,0.019267200119793416,0.017952000722289085,0.0035068687666949577,0.0,0.0,0.0,0.0,0.0,0.049855999648571014,0.07798399776220322,0.05986879989504815,0.05986879989504815,0.01043914754878828,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016383999958634377,0.026079999282956123,0.01923519968986511,0.017791999503970146,0.0032161974331284568,0.0,0.0,0.0,0.0,0.0,0.058111999183893204,0.1045759990811348,0.06708480007946492,0.06708480007946492,0.013479022462646494,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.04057599976658821,0.019100800156593323,0.015455999877303839,0.007512281243011577,0.0,0.0,0.0,0.0,0.0,0.05363199859857559,0.07782399654388428,0.06090559959411622,0.06090559959411622,0.007544620176348091,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.02191999927163124,0.016684799920767546,0.01563199982047081,0.0024293118621811216,0.0,0.0,0.0,0.0,0.0,0.0629120022058487,0.07891199737787247,0.06891520097851753,0.06891520097851753,0.005472695665695425,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.038943998515605927,0.018313600029796363,0.01561600062996149,0.007127270260115769,0.0,0.0,0.0,0.0,0.0,0.08675199747085571,0.10662399977445602,0.09391999915242194,0.09391999915242194,0.006988099589086635,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.054687999188899994,0.021439999900758268,0.01539199985563755,0.012052764849597775,0.0,0.0,0.0,0.0,0.0,0.13488000631332397,0.1528639942407608,0.1431359991431236,0.1431359991431236,0.005436271464033599,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.041919998824596405,0.01899839974939823,0.015343999955803156,0.007989843526623287,0.0,0.0,0.0,0.0,0.0,0.06176000088453293,0.08374399691820145,0.06747519969940186,0.06747519969940186,0.0066067747128778,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016256000846624374,0.11353600025177002,0.042761600017547606,0.028960000723600388,0.029104301538020762,0.0,0.0,0.0,0.0,0.0,0.09734400361776352,0.14422400295734406,0.11392960175871848,0.11392960175871848,0.013198594600417867,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.034143999218940735,0.018918400071561335,0.01643200032413006,0.005500943993080684,0.0,0.0,0.0,0.0,0.0,0.12918399274349213,0.15087999403476715,0.13807999789714814,0.13807999789714814,0.007658330538677587,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.03641600161790848,0.0198208000510931,0.01780799962580204,0.0057264128169845765,0.0,0.0,0.0,0.0,0.0,0.22099199891090393,0.23904000222682953,0.2293503984808922,0.2293503984808922,0.004861342907006028,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.035840000957250595,0.018908800091594458,0.015792000107467175,0.0064374817924757475,0.0,0.0,0.0,0.0,0.0,0.10134399682283401,0.12201599776744843,0.10896319895982742,0.10896319895982742,0.006336330809165179,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.03417599946260452,0.017846399918198586,0.014800000004470348,0.006495007539635255,0.0,0.0,0.0,0.0,0.0,0.13126400113105774,0.15561600029468536,0.1389280006289482,0.1389280006289482,0.008381472811075022,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03519999980926514,0.019168000388890504,0.01600000075995922,0.005995522477654695,0.0,0.0,0.0,0.0,0.0,0.21728000044822693,0.2343679964542389,0.2231455981731415,0.2231455981731415,0.004720730646739123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.03670400008559227,0.018441599886864425,0.015023999847471714,0.006596535162793127,0.0,0.0,0.0,0.0,0.0,0.39321601390838623,0.4524799883365631,0.4058080047369003,0.4058080047369003,0.01578349755088941,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.022143999114632607,0.016912000067532063,0.015168000012636185,0.0028156089295136347,0.0,0.0,0.0,0.0,0.0,0.15587200224399567,0.3079040050506592,0.17838079929351805,0.17838079929351805,0.04355865575265927,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.02112000063061714,0.015516800060868263,0.01488000014796853,0.001940870731593904,0.0,0.0,0.0,0.0,0.0,0.21587200462818146,0.23561599850654602,0.22250880002975468,0.22250880002975468,0.006181951170646666,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.039264000952243805,0.023609600123018028,0.02131200022995472,0.007236979625711548,0.0,0.0,0.0,0.0,0.0,0.408735990524292,0.470335990190506,0.4336863994598388,0.4336863994598388,0.01844662383160074,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.025407999753952026,0.016227199975401164,0.014960000291466713,0.0031617375441736185,0.0,0.0,0.0,0.0,0.0,0.7412800192832947,0.7627840042114258,0.7464000046253203,0.7464000046253203,0.006112167448837547,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.02812799997627735,0.020652799773961304,0.021359999664127827,0.004308706957613102,0.028383498565450627,0.039859687970646644,0.032492258074592426,0.032492258074592426,0.00453597266208597,0.029312501176103633,0.04116431058787463,0.03355574193327539,0.03355574193327539,0.004684436757701648,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.034143999218940735,0.023171199765056372,0.024255999363958836,0.005908614918096971,0.03333159243114438,0.038935341782478095,0.03580428402241854,0.03580428402241854,0.002082270297044095,0.03633240903369937,0.04244065945634484,0.03902771507087562,0.03902771507087562,0.0022697354261490147,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.02611199952661991,0.019635199941694735,0.018112000077962875,0.0038298050749257795,0.04189529417991216,0.057484239920526384,0.04744885718421094,0.04744885718421094,0.004779830748455743,0.051672703037266184,0.07089975418195164,0.05852234134479412,0.05852234134479412,0.005895334539786378,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.029311999678611755,0.018662399891763926,0.01673599984496832,0.0044017112162725355,0.04322973959325901,0.05049827064705393,0.04535414343408448,0.04535414343408448,0.0022944156000240697,0.08733025617719538,0.10201372836398578,0.09162185574241775,0.09162185574241775,0.004635047631846023,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.026367999613285065,0.016336000058799982,0.01515199989080429,0.003415353455946402,0.06031842775160765,0.06618323188375198,0.06274415549817247,0.06274415549817247,0.001985647289644255,0.14768157653992678,0.16204076249052324,0.15362064543185072,0.15362064543185072,0.004861590945216247,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.02454400062561035,0.017167999967932702,0.016191999427974224,0.0028685016454498436,0.09128700688359712,0.09689150775996329,0.09360396051475776,0.09360396051475776,0.0013477542744916764,0.2740889887523892,0.2909164873209846,0.2810456356995366,0.2810456356995366,0.004046628526808561,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01881599985063076,0.03097599931061268,0.024598400108516216,0.024848000146448612,0.0037539437391565975,0.1035249255866932,0.10603132147437412,0.10399797220840973,0.10399797220840973,0.0007025109429056814,0.5652750707893444,0.5789606938874117,0.5678580377517648,0.5678580377517648,0.003835906384194897,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018624000251293182,0.043487999588251114,0.024460799992084503,0.019600000232458115,0.008922081629781394,0.196169204945307,0.2076140047945799,0.2008444429250931,0.2008444429250931,0.00394350953292805,1.1196708009293268,1.1849940417370974,1.1463555573610091,1.1463555573610091,0.022508285530529783,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.027135999873280525,0.03667199984192848,0.02945920005440712,0.028256000019609928,0.0029446678408169553,0.37757279619664613,0.3898113624476372,0.38075520430942184,0.38075520430942184,0.0036600203831042254,0.2522831942132049,0.2604606493092597,0.25440958703617433,0.25440958703617433,0.0024455194930253126,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.0435199998319149,0.049695998430252075,0.04502719938755036,0.04395199939608574,0.002035785660169308,1.1048984388245497,1.1189053886476108,1.1112371236754128,1.1112371236754128,0.0050220696724624985,0.1395495077239122,0.14131859607071193,0.14035008841031085,0.14035008841031085,0.0006342911944856293,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018783999606966972,0.024288000538945198,0.020627199858427047,0.019567999988794327,0.0019457052717059358,0.09455999732017517,0.12185599654912949,0.10207359939813614,0.10207359939813614,0.00753346544014261,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027295999228954315,0.034752000123262405,0.02943360023200512,0.028528000228106976,0.0023159160681123767,0.14416000247001648,0.15904000401496887,0.14979200065135959,0.14979200065135959,0.00480624675866005,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.043455999344587326,0.05084799975156784,0.045500800386071204,0.04411200061440468,0.002490556062831049,0.24454399943351746,0.25865599513053894,0.2504959970712662,0.2504959970712662,0.003778494140301353,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.07673600316047668,0.08374399691820145,0.07970559895038605,0.07873599976301193,0.0022907976135004057,0.445248007774353,0.4758400022983551,0.45409599840641024,0.45409599840641024,0.009133230340383306,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.043296001851558685,0.051072001457214355,0.044972800090909,0.04399999976158142,0.002451216824441962,0.6147199869155884,0.6290879845619202,0.6204927921295166,0.6204927921295166,0.004344846739788608,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.07737600058317184,0.09123200178146362,0.08094720020890236,0.07980800047516823,0.004065264798368611,1.1744320392608643,1.1887680292129517,1.178323209285736,1.178323209285736,0.004395876469319665,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.02364799939095974,0.01809599995613098,0.016352000646293163,0.0035481127058959038,0.04822399839758873,0.08566399663686752,0.05961279980838299,0.05961279980838299,0.011445665413968877,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.020864000543951988,0.01668160008266568,0.015664000064134598,0.0025769063833097584,0.049695998430252075,0.08057600259780884,0.05882879942655563,0.05882879942655563,0.009515126108519331,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.028704000636935234,0.017430400010198355,0.015056000091135502,0.004349294613335555,0.049855999648571014,0.07366400212049484,0.05459520071744919,0.05459520071744919,0.0069610920757925574,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.03161599859595299,0.017788799852132796,0.015711999963968992,0.005005385723318659,0.06102399900555611,0.08179199695587158,0.06650560013949873,0.06650560013949873,0.006947995595801105,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.04416000097990036,0.019200000166893005,0.015488000120967627,0.008561241323364038,0.08054400235414505,0.09196799993515015,0.08607039973139763,0.08607039973139763,0.004145329035483754,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.023391999304294586,0.019667199812829494,0.018463999964296818,0.0022577669687832585,0.18729600310325623,0.20585599541664124,0.19546559900045393,0.19546559900045393,0.006339068824303663,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.026240000501275063,0.03446400165557861,0.028297600522637367,0.027088000439107418,0.0025148441522922374,0.5754240155220032,0.5889919996261597,0.5800191938877105,0.5800191938877105,0.003858869273829596,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04182400181889534,0.047200001776218414,0.043036799877882004,0.0423360001295805,0.0017073405772076728,2.063199996948242,2.0787200927734375,2.067151999473572,2.067151999473572,0.004959651271127835,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.05648000165820122,0.02270399993285537,0.017935999669134617,0.012377915150727689,0.049536000937223434,0.07196799665689468,0.056015999615192415,0.056015999615192415,0.0070476637552742884,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01408000010997057,0.021344000473618507,0.01611520005390048,0.014928000047802925,0.0025499884993961702,0.07072000205516815,0.2642880082130432,0.1588256008923054,0.1588256008923054,0.053220347086102376,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01360000018030405,0.03014400042593479,0.016975999902933837,0.015056000091135502,0.004715159697364111,0.08393599838018417,0.11036799848079681,0.09160000011324881,0.09160000011324881,0.007911669434472792,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.020767999812960625,0.016233599931001663,0.015392000321298838,0.002202615907995427,0.1998399943113327,0.22070400416851044,0.20855360180139543,0.20855360180139543,0.00689307200230667,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.033440001308918,0.018684800155460833,0.016048000194132328,0.00553991005639174,0.6043199896812439,0.635807991027832,0.6126143991947173,0.6126143991947173,0.008745933953408096,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.03359999880194664,0.017811199743300678,0.014944000169634819,0.005926140483565472,0.0,0.0,0.0,0.0,0.0,0.045471999794244766,0.07539200037717819,0.054758400097489356,0.054758400097489356,0.010253548506101549,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02409599907696247,0.01641279999166727,0.01473599998280406,0.003347158638713987,0.0,0.0,0.0,0.0,0.0,0.0461760014295578,0.07529599964618683,0.05460800044238568,0.05460800044238568,0.009748937798340135,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.028416000306606293,0.018611199874430894,0.01598400017246604,0.005209875093863647,0.0,0.0,0.0,0.0,0.0,0.048128001391887665,0.07897599786520004,0.061353600397706036,0.061353600397706036,0.010488153655157845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.025728000327944756,0.016092800162732603,0.01512000011280179,0.003300888123052605,0.0,0.0,0.0,0.0,0.0,0.04864000156521797,0.07648000121116638,0.05810560062527656,0.05810560062527656,0.009473544218404408,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.03551999852061272,0.018588799890130757,0.016064000315964222,0.006071057437992447,0.0,0.0,0.0,0.0,0.0,0.04822399839758873,0.07862400263547897,0.05660480037331582,0.05660480037331582,0.009565394213730401,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.028831999748945236,0.01699519995599985,0.015024000313133001,0.004285000744868188,0.0,0.0,0.0,0.0,0.0,0.04854400083422661,0.06719999760389328,0.05778240002691746,0.05778240002691746,0.006852125554679805,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.024927999824285507,0.0161183999851346,0.01521599991247058,0.0029774808524673907,0.0,0.0,0.0,0.0,0.0,0.05379199981689453,0.08966399729251862,0.06270079985260964,0.06270079985260964,0.009983591277092696,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03001599945127964,0.017648000083863736,0.01547200046479702,0.004585126309484848,0.0,0.0,0.0,0.0,0.0,0.061664000153541565,0.07692799717187881,0.06704320013523103,0.06704320013523103,0.005500191798490629,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.026367999613285065,0.016947199776768684,0.015039999969303608,0.0037951566103550205,0.0,0.0,0.0,0.0,0.0,0.08799999952316284,0.111455999314785,0.0964031994342804,0.0964031994342804,0.007397541615558088,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01369599997997284,0.021247999742627144,0.015359999984502793,0.014640000183135271,0.0020942770950814317,0.0,0.0,0.0,0.0,0.0,0.051711998879909515,0.07065600156784058,0.058054400235414506,0.058054400235414506,0.006633034815910025,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.0307839997112751,0.017148799914866685,0.015039999969303608,0.004882374686312284,0.0,0.0,0.0,0.0,0.0,0.061919998377561576,0.07843200117349625,0.06628479920327664,0.06628479920327664,0.004962852689801192,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013887999579310417,0.02969600073993206,0.017500799987465142,0.015232000034302473,0.00467273319705314,0.0,0.0,0.0,0.0,0.0,0.08819200098514557,0.11097600311040878,0.09493440166115762,0.09493440166115762,0.007509235042985577,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.022112000733613968,0.01751680001616478,0.016047999262809753,0.0030306621792915785,0.0,0.0,0.0,0.0,0.0,0.13065600395202637,0.15110400319099426,0.13857279866933822,0.13857279866933822,0.00750137841249771,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.04012800008058548,0.01892479993402958,0.01550400024279952,0.007459545267816961,0.0,0.0,0.0,0.0,0.0,0.06278400123119354,0.08259200304746628,0.0704512007534504,0.0704512007534504,0.005979055984382744,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.02236800082027912,0.016332800220698118,0.014928000047802925,0.002784044363186731,0.0,0.0,0.0,0.0,0.0,0.1003199964761734,0.1279360055923462,0.10921279862523078,0.10921279862523078,0.00862769617046716,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0225600004196167,0.016128000058233737,0.014800000004470348,0.002958953332547708,0.0,0.0,0.0,0.0,0.0,0.13116799294948578,0.14812800288200378,0.13783999979496003,0.13783999979496003,0.005696384361148053,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.029983999207615852,0.017468800116330386,0.01508800033479929,0.0047379550962483065,0.0,0.0,0.0,0.0,0.0,0.217631995677948,0.2447360008955002,0.22715839892625808,0.22715839892625808,0.008831828221847138,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02304000034928322,0.016512000095099212,0.014512000139802694,0.0035026774828624254,0.0,0.0,0.0,0.0,0.0,0.11020799726247787,0.12307199835777283,0.11600959971547126,0.11600959971547126,0.004667637266950902,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.042080000042915344,0.017510399967432023,0.014607999939471483,0.00823117883530167,0.0,0.0,0.0,0.0,0.0,0.15702399611473083,0.17587199807167053,0.16399359852075576,0.16399359852075576,0.006588299074676393,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0208320003002882,0.015299199987202883,0.01462399959564209,0.001916768086505386,0.0,0.0,0.0,0.0,0.0,0.21779200434684753,0.2415360063314438,0.2260768011212349,0.2260768011212349,0.007251352080685236,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014015999622642994,0.021088000386953354,0.015619200188666582,0.01473599998280406,0.002013227452302141,0.0,0.0,0.0,0.0,0.0,0.3959999978542328,0.4152640104293823,0.40332479774951924,0.40332479774951924,0.006942401431914052,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.03379200026392937,0.020595200080424547,0.019600000232458115,0.006548754248881344,0.026192623739694512,0.040006882507168426,0.029155180178492036,0.029155180178492036,0.00408828748438241,0.024591375524545756,0.037561119537986146,0.027372820355088746,0.027372820355088746,0.0038383559348575944,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.020896000787615776,0.015318400040268899,0.014431999996304512,0.0019640312960926966,0.029349018208693862,0.06236262941356679,0.03714313592014963,0.03714313592014963,0.009618313791872569,0.028826981462526914,0.06125337308649041,0.036482463672774496,0.036482463672774496,0.009447230957011863,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.032127998769283295,0.017286399938166143,0.014479999896138906,0.005536495446636193,0.03273085874558354,0.04394578491937082,0.03563837490653034,0.03563837490653034,0.003429601730256567,0.03488514202593899,0.04683821345079978,0.037984025407085474,0.037984025407085474,0.0036553316361902684,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.021503999829292297,0.015078400075435639,0.01425600005313754,0.002238435975478849,0.042100813549974185,0.051784144690147756,0.045378692890289625,0.045378692890289625,0.003184109940650555,0.05111518843748309,0.06287185663927425,0.05509490773570219,0.05509490773570219,0.0038658725544299132,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.02687999978661537,0.017667199857532977,0.01566399959847331,0.003864811846387173,0.048475323773821306,0.05599956978723733,0.05123499252968345,0.05123499252968345,0.002427379820423941,0.08429268135885387,0.09737642835214408,0.0890914090616175,0.0890914090616175,0.0042209177332077005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.03359999880194664,0.019804799742996693,0.016032000072300434,0.006750519908145474,0.07121508474579985,0.07292308367136396,0.07194410845270183,0.07194410845270183,0.0005500065915943844,0.14760091249712767,0.15114092373010238,0.14911189243564577,0.14911189243564577,0.0011399477384397005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.021856000646948814,0.016835200227797033,0.015263999812304974,0.00283854448975065,0.08146338272142935,0.08560865714384096,0.0834932625520734,0.0834932625520734,0.0012259635905347874,0.2782486219401307,0.29240733788179385,0.2851819365989658,0.2851819365989658,0.004187435731481665,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04495999962091446,0.02143679987639189,0.015856000129133463,0.009709987872683342,0.1194459208702178,0.12302524755181463,0.12053885718696096,0.12053885718696096,0.001005118119341339,0.5597220650459199,0.5764947443228802,0.5648435507167102,0.5648435507167102,0.004709970715400788,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018112000077962875,0.026496000587940216,0.019318400137126445,0.018432000651955605,0.0024249605112359905,0.19770253574610402,0.222811797868348,0.2054811520619412,0.2054811520619412,0.008000944254868043,0.13941746080159492,0.157124211777114,0.14490284788179197,0.14490284788179197,0.005642170080515992,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.02595200017094612,0.03155200183391571,0.02727359998971224,0.026575999334454536,0.0016260948764843166,0.6014684881116659,0.6095472988212,0.6046757700946376,0.6046757700946376,0.0023537435563177303,0.11325152264578045,0.11477269562841545,0.11385542721485654,0.11385542721485654,0.0004431903698023628,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.03359999880194664,0.01814719969406724,0.015375999733805656,0.005678506740662529,0.06774400174617767,0.07891199737787247,0.07312640026211739,0.07312640026211739,0.003826583051422731,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01788800023496151,0.028543999418616295,0.021340799890458582,0.019504000432789326,0.003718627037219618,0.09548799693584442,0.1327359974384308,0.10823359936475753,0.10823359936475753,0.013230635292043864,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.02627200074493885,0.030368000268936157,0.027184000052511693,0.02643200010061264,0.0013545679205210022,0.14364799857139587,0.1597760021686554,0.15008639842271806,0.15008639842271806,0.005367723415171222,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04150399938225746,0.04726399853825569,0.042950399965047834,0.04224000126123428,0.001720147501765378,0.2433920055627823,0.28963199257850647,0.2549152016639709,0.2549152016639709,0.013450991019687407,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.025887999683618546,0.03407999873161316,0.028303999826312064,0.026367999613285065,0.0028877495368841042,0.325439989566803,0.3441599905490875,0.33442879617214205,0.33442879617214205,0.005693597811441922,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04137599840760231,0.05084799975156784,0.044828799366950986,0.043087998405098915,0.003560363865797613,0.6110399961471558,0.6421759724617004,0.6204223990440368,0.6204223990440368,0.011214490370794758,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.022784000262618065,0.016435200069099664,0.015519999898970127,0.0023688301421469523,0.04879999905824661,0.09139200299978256,0.06054079942405224,0.06054079942405224,0.012152608702448775,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014816000126302242,0.02844800055027008,0.017008000146597625,0.015696000307798386,0.003941466294662679,0.047807998955249786,0.07356800138950348,0.05626560002565384,0.05626560002565384,0.00842179125412236,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.053408000618219376,0.019353600218892097,0.01532800029963255,0.01138302400841626,0.048448000103235245,0.0785600021481514,0.0556256003677845,0.0556256003677845,0.009348274502616908,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.03139200061559677,0.01823679991066456,0.016159999649971724,0.004860071238302512,0.055424001067876816,0.08505599945783615,0.0640383992344141,0.0640383992344141,0.00921448636178623,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.026528000831604004,0.017324799951165915,0.01536000007763505,0.0036004822686428305,0.07660800218582153,0.0942080020904541,0.08209280073642732,0.08209280073642732,0.00515895587669752,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.021247999742627144,0.016403199825435876,0.01532800029963255,0.001995897022647934,0.11395200341939926,0.15113599598407745,0.12431039959192276,0.12431039959192276,0.011164431123683732,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017184000462293625,0.028672000393271446,0.019865600019693376,0.017823999747633934,0.0035798491315929977,0.3171840012073517,0.3341119885444641,0.3261695951223373,0.3261695951223373,0.005111046452878918,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024639999493956566,0.030400000512599945,0.02656640000641346,0.02556800004094839,0.0020189846603237303,1.0648640394210815,1.0828479528427124,1.0715327858924866,1.0715327858924866,0.005558639263049851,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014976000413298607,0.021695999428629875,0.017305599898099898,0.01593599934130907,0.0025718046517268054,0.04956800118088722,0.07932800054550171,0.06228480041027069,0.06228480041027069,0.01027160349757827,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01539199985563755,0.03580800071358681,0.019686400331556796,0.017136000096797943,0.005918387811304003,0.05158400163054466,0.10713600367307663,0.06364160068333148,0.06364160068333148,0.015832957809696766,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02223999984562397,0.016672000009566545,0.015887999907135963,0.0020934945946034563,0.06521599739789963,0.08902399986982346,0.07520959973335266,0.07520959973335266,0.007913840684102929,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015231999568641186,0.04028800129890442,0.019971200078725816,0.01646399963647127,0.007351305886577286,0.14467200636863708,0.16844800114631653,0.15363519936800005,0.15363519936800005,0.008174435899956223,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015519999898970127,0.02304000034928322,0.01775679988786578,0.016784000210464,0.0025077243712082584,0.33740800619125366,0.35343998670578003,0.3445120006799698,0.3445120006799698,0.00445648463590358,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.02393599972128868,0.01809599995613098,0.01654400024563074,0.002998393102466254,0.0,0.0,0.0,0.0,0.0,0.0504320003092289,0.0843840017914772,0.060083200410008426,0.060083200410008426,0.00986959318572296,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.030047999694943428,0.019510399922728537,0.017680000513792038,0.004065185265963332,0.0,0.0,0.0,0.0,0.0,0.05004800111055374,0.07036799937486649,0.059315200522542,0.059315200522542,0.006537768821329647,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.02393599972128868,0.018035199772566558,0.016512000001966953,0.0030667689116777724,0.0,0.0,0.0,0.0,0.0,0.05100800096988678,0.06735999882221222,0.058387200161814694,0.058387200161814694,0.005832787739241381,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.026208000257611275,0.017500799987465142,0.01646399963647127,0.0030666103306165714,0.0,0.0,0.0,0.0,0.0,0.05023999884724617,0.07254400104284286,0.057254400476813315,0.057254400476813315,0.006890853653068151,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.027327999472618103,0.0179776000790298,0.01648000068962574,0.003783417447531392,0.0,0.0,0.0,0.0,0.0,0.04819199815392494,0.07100799679756165,0.05621119923889638,0.05621119923889638,0.007824391484124725,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.036448001861572266,0.019136000238358975,0.016704000532627106,0.006076530095624803,0.0,0.0,0.0,0.0,0.0,0.047488000243902206,0.08441600203514099,0.0588383998721838,0.0588383998721838,0.011275940726332522,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015135999768972397,0.033952001482248306,0.02119360016658902,0.018000000156462193,0.007136646614305304,0.0,0.0,0.0,0.0,0.0,0.04931199923157692,0.06992000341415405,0.05755840018391609,0.05755840018391609,0.007297587694315226,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.034272000193595886,0.0204415999352932,0.017311999574303627,0.00695404701803013,0.0,0.0,0.0,0.0,0.0,0.05215999856591225,0.0735040009021759,0.06117440015077591,0.06117440015077591,0.007381118436894922,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03743999823927879,0.02012479966506362,0.01775999926030636,0.006181045509079057,0.0,0.0,0.0,0.0,0.0,0.06355199962854385,0.08508799970149994,0.07019200026988984,0.07019200026988984,0.0062246580442117845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.032607998698949814,0.01959999995306134,0.017487999983131886,0.004882475068460749,0.0,0.0,0.0,0.0,0.0,0.04918399825692177,0.0865280032157898,0.05973760038614274,0.05973760038614274,0.011546847942113974,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02675200067460537,0.017430400010198355,0.016560000367462635,0.003243135094239819,0.0,0.0,0.0,0.0,0.0,0.052671998739242554,0.08367999643087387,0.06238719932734965,0.06238719932734965,0.011207824780630104,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.03612799942493439,0.019327999837696553,0.01688000001013279,0.006058251425153085,0.0,0.0,0.0,0.0,0.0,0.06364800035953522,0.07878399640321732,0.06935679838061332,0.06935679838061332,0.004515588328677643,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03711999952793121,0.019670399930328132,0.017152000218629837,0.006165994646522264,0.0,0.0,0.0,0.0,0.0,0.09071999788284302,0.11507199704647064,0.10252480059862136,0.10252480059862136,0.008782544051535657,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.026528000831604004,0.018396800104528665,0.01601599995046854,0.004279788048986283,0.0,0.0,0.0,0.0,0.0,0.05331199988722801,0.07977599650621414,0.06076480001211167,0.06076480001211167,0.008761579122069606,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.02643200010061264,0.01726400004699826,0.015855999663472176,0.003210008417242029,0.0,0.0,0.0,0.0,0.0,0.062431998550891876,0.08246400207281113,0.06970879957079888,0.06970879957079888,0.007016341676068233,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.02223999984562397,0.0166015999391675,0.015887999907135963,0.00204913969129354,0.0,0.0,0.0,0.0,0.0,0.10127999633550644,0.1141119971871376,0.10621120035648347,0.10621120035648347,0.004029318311754605,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03363199904561043,0.019305599946528675,0.016671999357640743,0.0055445582307981234,0.0,0.0,0.0,0.0,0.0,0.1319040060043335,0.15839999914169312,0.14040640145540234,0.14040640145540234,0.007998041073596942,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.04447999969124794,0.018908800091594458,0.0157279996201396,0.0086814937461721,0.0,0.0,0.0,0.0,0.0,0.06428799778223038,0.08982399851083755,0.07349760085344315,0.07349760085344315,0.008605284512197258,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.028672000393271446,0.01890560006722808,0.016080000437796116,0.004797584627815021,0.0,0.0,0.0,0.0,0.0,0.11123199760913849,0.14115199446678162,0.11942399889230729,0.11942399889230729,0.008513394706791027,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.024927999824285507,0.016915200091898442,0.015216000378131866,0.003374147499442635,0.0,0.0,0.0,0.0,0.0,0.15887999534606934,0.18111999332904816,0.16934399753808976,0.16934399753808976,0.007415181260761123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.026944000273942947,0.017737600207328796,0.015199999790638685,0.00415598714375255,0.0,0.0,0.0,0.0,0.0,0.2192319929599762,0.23472000658512115,0.22809920012950893,0.22809920012950893,0.004730841376327335,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.05052800104022026,0.020320000313222408,0.015520000364631414,0.010604522177924678,0.024270629882498958,0.03340247625954076,0.028446344104128624,0.028446344104128624,0.0032330369099793834,0.023697370291069768,0.03261352722995356,0.027774456325453972,0.027774456325453972,0.0031566742680923386,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.033663999289274216,0.019987199828028678,0.018240000121295452,0.005522929139253732,0.0259194055660947,0.03719755183990719,0.029916029687899703,0.029916029687899703,0.003966666590554318,0.027104595853449518,0.03889844644729374,0.03128396954022015,0.03128396954022015,0.004148046317967881,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.03299200162291527,0.019840000104159115,0.016207999549806118,0.006546343008726814,0.02587869595769926,0.03763167265431482,0.030174939058162802,0.030174939058162802,0.0037303581782277364,0.026473304070195713,0.03849632587654989,0.03086826083864964,0.03086826083864964,0.003816069654529222,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.021023999899625778,0.016304000187665223,0.015584000386297703,0.0023236165806545476,0.031810621525966996,0.04156949936878106,0.0346749350032807,0.0346749350032807,0.003130209181143985,0.035677378270900374,0.04662250161636451,0.038889864871740294,0.038889864871740294,0.0035107033960828727,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.02175999991595745,0.01569600012153387,0.015008000191301107,0.0020882666168036863,0.038211712107062853,0.07212229256520057,0.04364509673334097,0.04364509673334097,0.009671953074025085,0.0476442859917874,0.08992570455184198,0.05441890342616107,0.05441890342616107,0.01205947791784038,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.033824000507593155,0.017628800217062236,0.014864000026136637,0.005791495403857738,0.05214261250030033,0.06266261508706669,0.05677791295527661,0.05677791295527661,0.0030298223079651514,0.08648138506878382,0.10392938682791132,0.09416928531647478,0.09416928531647478,0.005025126612204489,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.031199999153614044,0.017900799959897996,0.015343999955803156,0.004974475661316249,0.06411958891421111,0.07405276123263956,0.06793047918211377,0.06793047918211377,0.0033918658556700006,0.14058441263169497,0.16236323591492058,0.1489399211274489,0.1489399211274489,0.00743678300375353,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.03855999931693077,0.018495999928563833,0.015647999942302704,0.006918530056761627,0.09872985549401277,0.10683454583043753,0.10185316839884552,0.10185316839884552,0.0020802254037042074,0.2743261390166379,0.2968454508984596,0.2830044295482752,0.2830044295482752,0.005780016596065092,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.030400000512599945,0.01977920001372695,0.01726400014013052,0.005350883464206169,0.13918872472233365,0.16279522855335207,0.1501912864839173,0.1501912864839173,0.008992185529011513,0.11892328861766266,0.13909276049083735,0.1283239123428725,0.1283239123428725,0.007682951884956939,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01727999933063984,0.02223999984562397,0.01819519978016615,0.017680000513792038,0.0014397298270620873,0.3728044181625443,0.38295504353701676,0.3761264483787333,0.3761264483787333,0.0033660272332490977,0.07967557017401883,0.08184495665371805,0.08038555277807365,0.08038555277807365,0.0007193856241088455,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.034591998904943466,0.01941439984366298,0.01756799966096878,0.005541380584373771,0.05951999872922897,0.08268799632787704,0.06715519949793816,0.06715519949793816,0.006657802116288364,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03283200040459633,0.019180799927562477,0.016543999314308167,0.0052187911233635975,0.07199999690055847,0.08675199747085571,0.07749439924955369,0.07749439924955369,0.004849874741156772,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01724799908697605,0.02284800074994564,0.01928640007972717,0.018400000408291817,0.0020641390157565697,0.09676799923181534,0.12310399860143663,0.10618879944086074,0.10618879944086074,0.008982738207839057,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024224000051617622,0.03315199911594391,0.027436799928545953,0.027328000403940678,0.002679154874546328,0.14716799557209015,0.16412800550460815,0.15470399856567385,0.15470399856567385,0.005423454433987419,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01833599992096424,0.04806400090456009,0.026252799853682517,0.022672000341117382,0.00878438440429033,0.1844799965620041,0.2072959989309311,0.19359359890222552,0.19359359890222552,0.007663539820967659,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024480000138282776,0.04979199916124344,0.029081599973142146,0.025679999962449074,0.007227422044689197,0.34147199988365173,0.37968000769615173,0.3583200007677078,0.3583200007677078,0.011804422100726231,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention_true_mixed_fused.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention_true_mixed_fused.csv new file mode 100644 index 0000000..2dd009a --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/attention_true_mixed_fused.csv @@ -0,0 +1,31 @@ +num_tensor_parallel_workers,batch_spec,num_prefill_seqs,num_decode_seqs,total_prefill_tokens,total_decode_tokens,decode_avg_kv_cache_size,attention_core_mean_ms,attention_core_mean_as_median_ms,kv_cache_update_median_ms,pure_prefill_reference_mean_ms,pure_decode_reference_mean_ms,projected_prefill_mean_ms,projected_decode_mean_ms,projection_sum_error_ms,representation +1,q64_8q1s512,1,8,64,8,512.0,0.06604800000786781,0.06604800000786781,0.021359999664127827,0.05810240097343922,0.0600041144660541,0.032492258074592426,0.03355574193327539,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_8q1s1k,1,8,128,8,1024.0,0.07483199909329416,0.07483199909329416,0.024255999363958836,0.05587520003318787,0.06090559959411622,0.03580428402241854,0.03902771507087562,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_16q1s1k,1,16,128,16,1024.0,0.10597119852900506,0.10597119852900506,0.018112000077962875,0.05587520003318787,0.06891520097851753,0.04744885718421094,0.05852234134479412,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_16q1s2k,1,16,256,16,2048.0,0.13697599917650224,0.13697599917650224,0.01673599984496832,0.056396800279617305,0.11392960175871848,0.04535414343408448,0.09162185574241775,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_32q1s2k,1,32,256,32,2048.0,0.2163648009300232,0.2163648009300232,0.01515199989080429,0.056396800279617305,0.13807999789714814,0.06274415549817247,0.15362064543185072,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_32q1s4k,1,32,512,32,4096.0,0.3746495962142944,0.3746495962142944,0.016191999427974224,0.07432000041007995,0.2231455981731415,0.09360396051475776,0.2810456356995366,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_64q1s4k,1,64,512,64,4096.0,0.6718560099601746,0.6718560099601746,0.024848000146448612,0.07432000041007995,0.4058080047369003,0.10399797220840973,0.5678580377517648,-1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q1k_64q1s8k,1,64,1024,64,8192.0,1.3472000002861022,1.3472000002861022,0.019600000232458115,0.1307712011039257,0.7464000046253203,0.2008444429250931,1.1463555573610091,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q2k_32q1s4k,1,32,2048,32,4096.0,0.6351647913455962,0.6351647913455962,0.028256000019609928,0.33396480381488797,0.2231455981731415,0.38075520430942184,0.25440958703617433,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q4k_16q1s4k,1,16,4096,16,4096.0,1.2515872120857237,1.2515872120857237,0.04395199939608574,1.0999775886535645,0.1389280006289482,1.1112371236754128,0.14035008841031085,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q64_8q1s512,1,8,64,8,512.0,0.05652800053358078,0.05652800053358078,0.019600000232458115,0.05961279980838299,0.055968457407185014,0.029155180178492036,0.027372820355088746,6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_8q1s1k,1,8,128,8,1024.0,0.07362559959292413,0.07362559959292413,0.014431999996304512,0.05882879942655563,0.05778240002691746,0.03714313592014963,0.036482463672774496,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_16q1s1k,1,16,128,16,1024.0,0.0736224003136158,0.0736224003136158,0.014479999896138906,0.05882879942655563,0.06270079985260964,0.03563837490653034,0.037984025407085474,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_16q1s2k,1,16,256,16,2048.0,0.10047360062599181,0.10047360062599181,0.01425600005313754,0.05459520071744919,0.06628479920327664,0.045378692890289625,0.05509490773570219,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_32q1s2k,1,32,256,32,2048.0,0.14032640159130094,0.14032640159130094,0.01566399959847331,0.05459520071744919,0.09493440166115762,0.05123499252968345,0.0890914090616175,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_32q1s4k,1,32,512,32,4096.0,0.22105600088834762,0.22105600088834762,0.016032000072300434,0.06650560013949873,0.13783999979496003,0.07194410845270183,0.14911189243564577,-2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_64q1s4k,1,64,512,64,4096.0,0.36867519915103913,0.36867519915103913,0.015263999812304974,0.06650560013949873,0.22715839892625808,0.0834932625520734,0.2851819365989658,5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q1k_64q1s8k,1,64,1024,64,8192.0,0.6853824079036711,0.6853824079036711,0.015856000129133463,0.08607039973139763,0.40332479774951924,0.12053885718696096,0.5648435507167102,1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q2k_32q1s4k,1,32,2048,32,4096.0,0.3503839999437332,0.3503839999437332,0.018432000651955605,0.19546559900045393,0.13783999979496003,0.2054811520619412,0.14490284788179197,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q4k_16q1s4k,1,16,4096,16,4096.0,0.7185311973094941,0.7185311973094941,0.026575999334454536,0.5800191938877105,0.10921279862523078,0.6046757700946376,0.11385542721485654,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q64_8q1s512,1,8,64,8,512.0,0.0562208004295826,0.0562208004295826,0.015520000364631414,0.06054079942405224,0.0591108573866742,0.028446344104128624,0.027774456325453972,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_8q1s1k,1,8,128,8,1024.0,0.061199999228119854,0.061199999228119854,0.018240000121295452,0.05626560002565384,0.0588383998721838,0.029916029687899703,0.03128396954022015,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_16q1s1k,1,16,128,16,1024.0,0.06104319989681244,0.06104319989681244,0.016207999549806118,0.05626560002565384,0.05755840018391609,0.030174939058162802,0.03086826083864964,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_16q1s2k,1,16,256,16,2048.0,0.07356479987502099,0.07356479987502099,0.015584000386297703,0.0556256003677845,0.06238719932734965,0.0346749350032807,0.038889864871740294,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_32q1s2k,1,32,256,32,2048.0,0.09806400015950203,0.09806400015950203,0.015008000191301107,0.0556256003677845,0.06935679838061332,0.04364509673334097,0.05441890342616107,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_32q1s4k,1,32,512,32,4096.0,0.1509471982717514,0.1509471982717514,0.014864000026136637,0.0640383992344141,0.10621120035648347,0.05677791295527661,0.09416928531647478,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_64q1s4k,1,64,512,64,4096.0,0.21687040030956264,0.21687040030956264,0.015343999955803156,0.0640383992344141,0.14040640145540234,0.06793047918211377,0.1489399211274489,2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q1k_64q1s8k,1,64,1024,64,8192.0,0.3848575979471207,0.3848575979471207,0.015647999942302704,0.08209280073642732,0.22809920012950893,0.10185316839884552,0.2830044295482752,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q2k_32q1s4k,1,32,2048,32,4096.0,0.2785151988267898,0.2785151988267898,0.01726400014013052,0.12431039959192276,0.10621120035648347,0.1501912864839173,0.1283239123428725,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q4k_16q1s4k,1,16,4096,16,4096.0,0.456512001156807,0.456512001156807,0.017680000513792038,0.3261695951223373,0.06970879957079888,0.3761264483787333,0.08038555277807365,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/linear_op.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/linear_op.csv new file mode 100644 index 0000000..2803f02 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/linear_op.csv @@ -0,0 +1,37 @@ +time_stats.emb.min,time_stats.emb.max,time_stats.emb.mean,time_stats.emb.median,time_stats.emb.std,time_stats.input_layernorm.min,time_stats.input_layernorm.max,time_stats.input_layernorm.mean,time_stats.input_layernorm.median,time_stats.input_layernorm.std,time_stats.attn_pre_proj.min,time_stats.attn_pre_proj.max,time_stats.attn_pre_proj.mean,time_stats.attn_pre_proj.median,time_stats.attn_pre_proj.std,time_stats.attn_rope.min,time_stats.attn_rope.max,time_stats.attn_rope.mean,time_stats.attn_rope.median,time_stats.attn_rope.std,time_stats.attn_post_proj.min,time_stats.attn_post_proj.max,time_stats.attn_post_proj.mean,time_stats.attn_post_proj.median,time_stats.attn_post_proj.std,time_stats.post_attention_layernorm.min,time_stats.post_attention_layernorm.max,time_stats.post_attention_layernorm.mean,time_stats.post_attention_layernorm.median,time_stats.post_attention_layernorm.std,n_head,n_kv_head,n_embd,n_expanded_embd,vocab_size,use_gated_mlp,use_qk_norm,attn_output_gate,num_tokens,num_tensor_parallel_workers,padded_n_embd,padded_n_expanded_embd,model_arch,is_step2_mini,share_expert_dim,share_q_dim,measurement_type,profiling_precision,quant_signature +0.029184000566601753,0.06780800223350525,0.03157280012965202,0.030736000277101994,0.005874173435341216,0.033215999603271484,0.04825599864125252,0.03443359974771738,0.0337119996547699,0.0031825678429048183,1.438431978225708,1.505568027496338,1.446228802204132,1.4429279565811157,0.014128607642643025,0.538752019405365,0.5440319776535034,0.5416463971138,0.5420799851417542,0.0016099306164432847,1.0073280334472656,1.0163840055465698,1.0098415970802308,1.0081279873847961,0.0032980457197329728,0.04022400081157684,0.041728001087903976,0.04091359991580248,0.04081599973142147,0.0004728505416767937,32,4,2048,768,151936,True,True,False,8192,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.01360000018030405,0.06784000247716904,0.017755200061947106,0.0179840000346303,0.00837681421401443,0.01836800016462803,0.03651199862360954,0.019606399815529585,0.018719999119639397,0.0038821958848767424,0.7512000203132629,0.8208960294723511,0.7566704005002975,0.75382399559021,0.014793092586577971,0.28995200991630554,0.29337599873542786,0.29135999977588656,0.29150401055812836,0.000998381071258815,0.5149760246276855,0.5169600248336792,0.5160208016633987,0.5158880054950714,0.0005038652783291977,0.02051199972629547,0.021503999829292297,0.021067200042307378,0.021104000508785248,0.00023542855306902367,32,4,2048,768,151936,True,True,False,4096,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.0080960001796484,0.04281599819660187,0.011092799971811474,0.011039999779313803,0.005489135752949302,0.012223999947309494,0.026335999369621277,0.013187199970707298,0.01247999956831336,0.0030236510562153375,0.3928639888763428,0.45372799038887024,0.3976895987987518,0.39528000354766846,0.012905176716136006,0.1547199934720993,0.15884800255298615,0.15712319910526276,0.1573439985513687,0.001165121945135037,0.26633599400520325,0.268095999956131,0.26719200164079665,0.2671840041875839,0.0004242740199415328,0.013024000450968742,0.013887999579310417,0.013489600038155913,0.013520000036805868,0.0002382817554057738,32,4,2048,768,151936,True,True,False,2048,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.00825599953532219,0.04755200073122978,0.02398160002194345,0.01961600035429001,0.00874683840888523,0.018880000337958336,0.03868800029158592,0.021590400114655496,0.0208320003002882,0.004057015751746236,0.24316799640655518,0.2710399925708771,0.2521967992186546,0.25065599381923676,0.007998418528894075,0.09644799679517746,0.19120000302791595,0.10407840013504029,0.09963199868798256,0.020043722414992166,0.13600000739097595,0.18892799317836761,0.15760480016469955,0.15760000050067902,0.0112223677907762,0.009664000011980534,0.010015999898314476,0.009836799977347255,0.009824000298976898,9.016971952948177e-05,32,4,2048,768,151936,True,True,False,1024,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017376000061631203,0.044704001396894455,0.026318399980664254,0.027312000282108784,0.007565344034680866,0.01836800016462803,0.03014400042593479,0.021276800055056812,0.020655999891459942,0.002632977855097951,0.1438719928264618,0.17132799327373505,0.152497598528862,0.15012799948453903,0.007947150319625347,0.1430719941854477,0.19305600225925446,0.16630879789590836,0.1685439944267273,0.014628024163894684,0.08899199962615967,0.10467199981212616,0.0943599995225668,0.09374399855732918,0.003472669494074113,0.007615999784320593,0.007935999892652035,0.007769599952735007,0.0077760000713169575,7.680004540222077e-05,32,4,2048,768,151936,True,True,False,512,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016992000862956047,0.0544000007212162,0.02573199989274144,0.026016000658273697,0.00803323401720434,0.018432000651955605,0.02348800003528595,0.020648000109940768,0.02062400057911873,0.0013939985047930988,0.10220800340175629,0.1361600011587143,0.11850560046732425,0.11684799939393997,0.010637754898360304,0.16710400581359863,0.21110400557518005,0.19078560024499894,0.19409599900245667,0.013714352646558832,0.06265600025653839,0.0740479975938797,0.06842879951000214,0.0690080001950264,0.0031292240446560557,0.00979200005531311,0.033440001308918,0.01864320016466081,0.017280000261962414,0.006957998188876383,32,4,2048,768,151936,True,True,False,256,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017152000218629837,0.04396799951791763,0.025907999789342284,0.02598400041460991,0.007714554737915267,0.018400000408291817,0.03667199984192848,0.024132800102233887,0.02112000063061714,0.006120348733354326,0.10678400099277496,0.1363839954137802,0.1193264003843069,0.11583999916911125,0.009838647443214228,0.17017599940299988,0.22748799622058868,0.18853759989142418,0.18433599919080734,0.015728078443174653,0.04569600149989128,0.06652799993753433,0.05192639995366335,0.05151999928057194,0.004516605694976449,0.021856000646948814,0.026335999369621277,0.02384479995816946,0.023599999956786633,0.0011301153013314744,32,4,2048,768,151936,True,True,False,128,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017343999817967415,1.0683200359344482,0.05748240072280168,0.029504000209271908,0.16366824814254827,0.018688000738620758,0.3317759931087494,0.03685439983382821,0.021151999942958355,0.06767422466731164,0.10255999863147736,0.9434880018234253,0.16412640027701855,0.11956800147891045,0.17964606281728834,0.1714559942483902,2.1306240558624268,0.3011296011507511,0.1926399990916252,0.42310127734378766,0.03574400022625923,0.6859520077705383,0.08389280084520578,0.04279999993741512,0.14321647071615612,0.020479999482631683,0.1831360012292862,0.033024000097066165,0.023856000043451786,0.03470987082429836,32,4,2048,768,151936,True,True,False,64,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016095999628305435,0.05142400041222572,0.026363200135529043,0.02676799986511469,0.008637725852473854,0.01849599927663803,0.03577600046992302,0.022193600237369538,0.020848000422120094,0.004615266181181815,0.10540799796581268,0.15014399588108063,0.12211520001292228,0.11896000057458878,0.012772013396624768,0.17315199971199036,0.21478399634361267,0.1881632000207901,0.1873439997434616,0.011761164657572015,0.03481600061058998,0.058079998940229416,0.042200000025331974,0.03969600051641464,0.006461281521769989,0.02143999934196472,0.038816001266241074,0.02466559996828437,0.023856000043451786,0.0035418033748569927,32,4,2048,768,151936,True,True,False,32,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015904000028967857,0.03753599897027016,0.023127999808639287,0.02527999971061945,0.0054230027890305385,0.018400000408291817,0.03001599945127964,0.020488000102341176,0.020096000283956528,0.002585688394137697,0.10355199873447418,0.1438400000333786,0.11536479964852334,0.11124800145626068,0.011136028294919255,0.16502399742603302,0.2072959989309311,0.18646399974822997,0.19075199961662292,0.013296437761247597,0.03142400085926056,0.05215999856591225,0.03888959977775812,0.03750399872660637,0.0057399302067536314,0.020128000527620316,0.04064000025391579,0.023937600292265417,0.023648000322282314,0.004144197400898248,32,4,2048,768,151936,True,True,False,16,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015200000256299973,0.037151999771595,0.02144480012357235,0.02195199951529503,0.005576364091933697,0.017952000722289085,0.0226879995316267,0.019934400077909233,0.019952000118792057,0.0011812499470458758,0.10063999891281128,0.13468800485134125,0.11595199964940547,0.1207519993185997,0.011974301621092394,0.16332800686359406,0.20748800039291382,0.18162400051951408,0.1767839938402176,0.01474120743720334,0.03222399950027466,0.043455999344587326,0.03829439990222454,0.03859200142323971,0.0031378131240041122,0.020959999412298203,0.03587200120091438,0.023795200139284135,0.023423999547958374,0.0031339489695198443,32,4,2048,768,151936,True,True,False,8,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.014944000169634819,0.04022400081157684,0.02162720002233982,0.022463999688625336,0.006017219317026815,0.018464000895619392,0.026528000831604004,0.021264000236988066,0.020896000787615776,0.001934095017586082,0.10156799852848053,0.1703999936580658,0.12565439902245998,0.1244799979031086,0.01641776722785409,0.1653759926557541,0.23865599930286407,0.1969360001385212,0.19223999977111816,0.02058903050274278,0.03254399821162224,0.06752000004053116,0.0443536002188921,0.041519999504089355,0.00984829071098989,0.020096000283956528,0.040031999349594116,0.026934400014579297,0.02478400059044361,0.00562071702648593,32,4,2048,768,151936,True,True,False,1,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.7531200051307678,0.8461440205574036,0.7618160009384155,0.7576479911804199,0.019464233617982506,0.2922559976577759,0.2985599935054779,0.2953856036067009,0.29576000571250916,0.001575901077689139,0.510047972202301,0.5140479803085327,0.5121696025133133,0.5123839974403381,0.0012263137313476844,,,,,,32,4,2048,768,151936,True,True,False,8192,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.39529600739479065,0.47279998660087585,0.40216960161924364,0.39825600385665894,0.0162749796240819,0.15625600516796112,0.16211199760437012,0.15959519892930984,0.15988799929618835,0.0011493418942396922,0.2635200023651123,0.2642880082130432,0.2639120012521744,0.26392000913619995,0.0001903593401384135,,,,,,32,4,2048,768,151936,True,True,False,4096,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2531839907169342,0.32419198751449585,0.26924319565296173,0.26049599051475525,0.01657194262368116,0.10051199793815613,0.2375359982252121,0.12411200068891048,0.10311999917030334,0.039466165428540506,0.15881599485874176,0.19289599359035492,0.16896959990262986,0.1640480011701584,0.009162416086418127,,,,,,32,4,2048,768,151936,True,True,False,2048,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15302400290966034,0.1780800074338913,0.16284480094909667,0.16113600134849548,0.00740355661356144,0.14521600306034088,0.20233599841594696,0.17807039842009545,0.1796799972653389,0.013908448560094403,0.0907519981265068,0.09750399738550186,0.09460479989647866,0.09478399902582169,0.0017965487667361475,,,,,,32,4,2048,768,151936,True,True,False,1024,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11740799993276596,0.17958399653434753,0.13262080028653145,0.12878400087356567,0.014146060532423056,0.17468799650669098,0.21161599457263947,0.1910431995987892,0.18966399878263474,0.01270903647700971,0.05926400050520897,0.07577600330114365,0.06530559975653887,0.06404799968004227,0.00479458462514625,,,,,,32,4,2048,768,151936,True,True,False,512,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11260800063610077,0.17209599912166595,0.13610880002379416,0.13809599727392197,0.017675922458993677,0.18729600310325623,0.26822400093078613,0.20815680101513861,0.2078079953789711,0.01700718593658771,0.04499199986457825,0.06537599861621857,0.0507551996037364,0.04787199944257736,0.005727123232692158,,,,,,32,4,2048,768,151936,True,True,False,256,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11257600039243698,0.23452800512313843,0.13866880126297473,0.13964799791574478,0.026590047697062115,0.1828799992799759,0.24751999974250793,0.20735519900918006,0.20670399814844131,0.01500438131586336,0.03654399886727333,0.05593600124120712,0.0408239996060729,0.03892800025641918,0.004864409450710354,,,,,,32,4,2048,768,151936,True,True,False,128,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11734399944543839,0.18726399540901184,0.13890240006148816,0.13308800011873245,0.019086167340006257,0.16991999745368958,0.2443840056657791,0.19185120090842248,0.19257599860429764,0.018017536159219673,0.03033600002527237,0.04956800118088722,0.038387199863791466,0.03750400058925152,0.00500897018702887,,,,,,32,4,2048,768,151936,True,True,False,64,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11420799791812897,0.18115200102329254,0.13938880078494548,0.1393439993262291,0.018141313962922536,0.1693439930677414,0.221343994140625,0.19187839925289155,0.19438399374485016,0.01430752121272052,0.03017600066959858,0.06019200012087822,0.03866560012102127,0.0364960003644228,0.0074199790031205266,,,,,,32,4,2048,768,151936,True,True,False,32,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11276800185441971,0.16223999857902527,0.1331360016018152,0.13305599987506866,0.01432493740801146,0.17187200486660004,0.23625600337982178,0.19287680014967917,0.1913280040025711,0.01754628831589704,0.029823999851942062,0.07574400305747986,0.03900959976017475,0.036927999928593636,0.009277465083962879,,,,,,32,4,2048,768,151936,True,True,False,16,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1130559965968132,0.19814400374889374,0.13379519879817964,0.12531199678778648,0.020074427302248836,0.16841599345207214,0.21139200031757355,0.1893615983426571,0.19092799723148346,0.012418720676762711,0.029503999277949333,0.07558400183916092,0.040144000016152856,0.03728000074625015,0.010223239196498205,,,,,,32,4,2048,768,151936,True,True,False,8,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.115167997777462,0.2640640139579773,0.13805920109152794,0.13118399679660797,0.03132849683515479,0.17103999853134155,0.2977280020713806,0.19899839907884598,0.1976960003376007,0.028928046763067948,0.0297279991209507,0.05990400165319443,0.03834720011800528,0.0363520011305809,0.0070885330713495905,,,,,,32,4,2048,768,151936,True,True,False,1,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.3940800130367279,0.46483200788497925,0.39923040121793746,0.3957759886980057,0.015108903906233569,0.16022400557994843,0.1634880006313324,0.16203359961509706,0.16228799521923065,0.0010220720912414007,0.26073598861694336,0.2627840042114258,0.26138080209493636,0.2613760083913803,0.00041640363494172775,,,,,,32,4,2048,768,151936,True,True,False,8192,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2536959946155548,0.27161601185798645,0.25960480123758317,0.2577280104160309,0.005142017404245015,0.09849599748849869,0.10281600058078766,0.10057279989123344,0.10063999891281128,0.0008836232237268523,0.13913600146770477,0.17606399953365326,0.15875840038061143,0.15988799929618835,0.010058952112389172,,,,,,32,4,2048,768,151936,True,True,False,4096,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15142400562763214,0.1780479997396469,0.16139679849147798,0.1602879986166954,0.0070656055731385115,0.14601600170135498,0.23343999683856964,0.17913119941949845,0.179967999458313,0.023065274309176566,0.09388799965381622,0.12108799815177917,0.10317599996924401,0.10073599964380264,0.007364345618470178,,,,,,32,4,2048,768,151936,True,True,False,2048,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12198399752378464,0.18406400084495544,0.13770400024950505,0.13232000172138214,0.01623164885687898,0.1773120015859604,0.20688000321388245,0.1870912007987499,0.18535999953746796,0.00825628058448234,0.05766399949789047,0.06739199906587601,0.06187200043350458,0.061216000467538834,0.003004312467037863,,,,,,32,4,2048,768,151936,True,True,False,1024,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11737599968910217,0.18614399433135986,0.13109439946711063,0.12771200388669968,0.01372168725934724,0.17587199807167053,0.2699519991874695,0.19926720038056372,0.19075199961662292,0.02327478982490538,0.041728001087903976,0.06224000081419945,0.05000480003654957,0.049375999718904495,0.006185850944273961,,,,,,32,4,2048,768,151936,True,True,False,512,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1159679964184761,0.1610880047082901,0.12988320142030715,0.12494400516152382,0.011865375783184065,0.1767680048942566,0.27379199862480164,0.21035519987344742,0.21275199949741364,0.023206964017550125,0.037087999284267426,0.062144000083208084,0.04439679980278015,0.04283200018107891,0.006155226392511492,,,,,,32,4,2048,768,151936,True,True,False,256,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12015999853610992,0.158720001578331,0.13573280088603495,0.1343199983239174,0.010666830836014414,0.1737920045852661,0.22127999365329742,0.20261440128087999,0.20321600139141083,0.011153965849067301,0.03587200120091438,0.04944000020623207,0.038265600241720675,0.037328001111745834,0.0030431580113575636,,,,,,32,4,2048,768,151936,True,True,False,128,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1212799996137619,0.15139199793338776,0.13183839991688728,0.13014400005340576,0.00851207547022104,0.17257599532604218,0.2250880002975464,0.1872655987739563,0.18193599581718445,0.013638284975248818,0.02969600073993206,0.0525440014898777,0.03840640028938651,0.03444799967110157,0.007598511754254174,,,,,,32,4,2048,768,151936,True,True,False,64,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11270400136709213,0.1624639928340912,0.13508000001311302,0.13180799782276154,0.014243564451606775,0.17052799463272095,0.23715199530124664,0.1953311987221241,0.1966560035943985,0.01500670718978398,0.030239999294281006,0.05270399898290634,0.03807039987295866,0.03742399998009205,0.004906165602021684,,,,,,32,4,2048,768,151936,True,True,False,32,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12319999933242798,0.17919999361038208,0.14059039913117885,0.14156799763441086,0.016217662982107223,0.1701119989156723,0.21987199783325195,0.18903039917349815,0.1870879977941513,0.014117854507841239,0.030368000268936157,0.06406400352716446,0.04045119984075427,0.03710399940609932,0.00878942205983628,,,,,,32,4,2048,768,151936,True,True,False,16,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11296000331640244,0.16761599481105804,0.13843199908733367,0.13814399391412735,0.014925286935582404,0.1711679995059967,0.21561600267887115,0.1906527981162071,0.1876479983329773,0.011803992622137974,0.0306560005992651,0.09216000139713287,0.041129599791020155,0.036847999319434166,0.013739721468154687,,,,,,32,4,2048,768,151936,True,True,False,8,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11929599940776825,0.2008959949016571,0.14535359852015972,0.13971199840307236,0.024769473061576282,0.16803200542926788,0.25123199820518494,0.19573760256171227,0.19366399943828583,0.023196011081705884,0.03049599938094616,0.08089599758386612,0.04739360017701984,0.0453919991850853,0.01286389846889463,,,,,,32,4,2048,768,151936,True,True,False,1,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/manifest.json b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/manifest.json new file mode 100644 index 0000000..511fc99 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/manifest.json @@ -0,0 +1,57 @@ +{ + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp1-20260717-v1-20260717T014550833983Z/artifacts/artifacts/attention-composition-tp1-v1/raw/flashattn-composition-tp1.json": "ce461be92a5059e2d168b2cbf21b55d9a55ee71c3b0586a3e63cd8288938168d", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp2-20260717-v1-20260717T014551831545Z/artifacts/artifacts/attention-composition-tp2-v1/raw/flashattn-composition-tp2.json": "997d67fb3a28b488c3aa4945ae7c6d4ecf8a08726ee3cb4652d16c221ff395ad", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp4-20260717-v1-20260717T014552897206Z/artifacts/artifacts/attention-composition-tp4-v1/raw/flashattn-composition-tp4.json": "046cffd4c11d8a36513167ec25d3e1ca016324715679ea688a58d4a71dfe62b2", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "03180eec28087a301c9d8c185874c50bdb28b31e5e9af35fcce5a6a5a9e62147", + "attention_true_mixed_fused.csv": "cccf76e6cb6a9359f188b07d4dc2942442568a6e8cf3ff3d77ab3d35a052cecb", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving-pure-prefill-batch-composition", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_pure_prefill_batch_composition": "Direct FA3 measurements for 2/4 requests at query length 2048 and 2/4/8/16 requests at query length 512 for each TP; included only when batch-composition attention inputs are supplied", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 150, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v3" +} diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/moe.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/moe.csv new file mode 100644 index 0000000..311097f --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final/moe.csv @@ -0,0 +1,73 @@ +time_stats.moe_gating_linear.min,time_stats.moe_gating_linear.max,time_stats.moe_gating_linear.mean,time_stats.moe_gating_linear.median,time_stats.moe_gating_linear.std,time_stats.moe_gating_routing_topk.min,time_stats.moe_gating_routing_topk.max,time_stats.moe_gating_routing_topk.mean,time_stats.moe_gating_routing_topk.median,time_stats.moe_gating_routing_topk.std,time_stats.moe_shuffling.min,time_stats.moe_shuffling.max,time_stats.moe_shuffling.mean,time_stats.moe_shuffling.median,time_stats.moe_shuffling.std,time_stats.moe_grouped_gemm.min,time_stats.moe_grouped_gemm.max,time_stats.moe_grouped_gemm.mean,time_stats.moe_grouped_gemm.median,time_stats.moe_grouped_gemm.std,num_tokens,num_experts,num_experts_per_device,expert_parallel_size,routing_runtime_path,routing_assignment_policy,routing_weight_policy,routing_uses_router_logits,gating_runtime_context,gating_runtime_context_impl,router_topk,hidden_dim,expert_hidden_dim,use_gated,num_tensor_parallel_workers,total_routed_tokens,model_expansion_ratio,tokens_per_expert_avg,tokens_to_experts_ratio,expert_utilization,min_load_ratio,load_imbalance_cv,max_load_ratio,load_entropy,load_gini_coefficient,load_distribution,seed,moe_grouped_gemm_backend,measurement_type,profiling_precision,model_arch,quant_signature,router_median_nonadditivity_ratio,projection_policy +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.33926400542259216,0.405023992061615,0.36780479848384856,0.36507199704647064,0.01690507644474779,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,1.1943039894104004,1.286784052848816,1.228384006023407,1.2273280024528503,0.02832547242381263,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.3984375,0.0,1.346291201783626,4.0,5.59375,0.661865234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,1.679744005203247,1.766144037246704,1.7095808148384095,1.7015680074691772,0.02438921262998535,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.625,0.0,1.015504800579495,5.0,6.15516433212955,0.529052734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,2.1576640605926514,2.2921600341796875,2.2097824096679686,2.188944101333618,0.045572321842012986,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.875,0.0,0.6343057228182637,2.5,6.64370748444639,0.35369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,2.422368049621582,2.5130879878997803,2.4516672134399413,2.434159994125366,0.03278900287381846,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.984375,0.0,0.4921254921257382,2.25,6.817190042344769,0.272369384765625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,2.2217600345611572,2.289599895477295,2.2571327924728393,2.263375997543335,0.021660416089449488,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,1.0,0.125,0.3486861500690843,1.875,6.908192310183997,0.197662353515625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,2.18668794631958,2.318079948425293,2.2218016147613526,2.211087942123413,0.035380897213135316,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,1.0,0.4375,0.2525504668006971,1.875,6.953347743053017,0.1410369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,2.2291839122772217,2.3929600715637207,2.2908096313476562,2.2804640531539917,0.04479348924786221,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,1.0,0.65625,0.15765965680164504,1.5625,6.98229848728205,0.08779525756835938,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,2.0678720474243164,2.1297600269317627,2.0837119817733765,2.0779199600219727,0.017880044357986735,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,1.0,0.625,0.12169081635504074,1.3125,6.9892029662356325,0.06879425048828125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,2.916032075881958,3.0819520950317383,2.9805248022079467,2.9656319618225098,0.05482195799019572,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,1.0,0.796875,0.07935434147688751,1.1796875,6.9954297964750305,0.044734954833984375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,4.421599864959717,4.535359859466553,4.486294317245483,4.497056007385254,0.036990243787549344,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,1.0,0.8203125,0.060849326483103046,1.17578125,6.9973188375859685,0.033740997314453125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,7.302591800689697,7.402751922607422,7.354758310317993,7.3464319705963135,0.032142662400335566,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,1.0,0.890625,0.0412323087266341,1.08984375,6.998772433185578,0.02334284782409668,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.35280001163482666,0.39692801237106323,0.37662720382213594,0.37196800112724304,0.013401318050665304,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.4692479968070984,0.5523840188980103,0.5134752035140991,0.5100640058517456,0.02291433464135784,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.34652799367904663,0.4119040071964264,0.3789471983909607,0.38550400733947754,0.02073945105803335,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.31462401151657104,0.7456960082054138,0.38617280423641204,0.34545600414276123,0.12230201266253077,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.32521599531173706,0.419871985912323,0.36325119733810424,0.34968000650405884,0.03161798848223672,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.289792001247406,0.4663360118865967,0.4091839998960495,0.41655999422073364,0.0446001986506615,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.3761279881000519,0.4416320025920868,0.40686399936676027,0.40540799498558044,0.02257778769899645,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.7172480225563049,0.8663039803504944,0.7723807990550995,0.7591840028762817,0.04164772379451242,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.0195519924163818,1.2216639518737793,1.1253888130187988,1.1453600525856018,0.06548005322243594,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7490559816360474,1.9644800424575806,1.8529024004936219,1.814303994178772,0.08042565617327288,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,3.2479360103607178,3.385279893875122,3.296070408821106,3.2800960540771484,0.04525529026442046,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,6.344799995422363,6.517856121063232,6.464438438415527,6.478623867034912,0.05116674443145098,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.2648000121116638,0.325439989566803,0.28852800130844114,0.28390398621559143,0.01933778377635077,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.7347840070724487,0.862496018409729,0.7769344031810761,0.769216001033783,0.03290485328796285,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.421875,0.0,1.346291201783626,6.0,5.652114648336087,0.636962890625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.9198399782180786,0.9646080136299133,0.9412063956260681,0.9411839842796326,0.014939365085478117,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.5703125,0.0,1.118033988749895,5.0,6.008641773518898,0.580810546875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,1.2796800136566162,1.3484159708023071,1.3006976008415223,1.2929120063781738,0.020807998177176254,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.8828125,0.0,0.6959705453537527,3.0,6.60872850615583,0.38055419921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,1.3630399703979492,1.4430400133132935,1.3909215927124023,1.3892319798469543,0.022335744492366926,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.984375,0.0,0.5201036555341637,3.0,6.798826509158851,0.28302001953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,1.27948796749115,1.3904000520706177,1.3176063895225525,1.309440016746521,0.038060887827312775,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,1.0,0.25,0.3511282039725661,1.875,6.91002266305238,0.1970977783203125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,1.264415979385376,1.3145920038223267,1.2791999936103822,1.2753440141677856,0.014130605249568332,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,1.0,0.375,0.24692938483248605,1.6875,6.955481130775285,0.13909912109375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,1.3081920146942139,1.347648024559021,1.3292255997657776,1.329967975616455,0.014558863679016933,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,1.0,0.625,0.17143053326165383,1.5625,6.9786675275754035,0.09520339965820312,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.242751955986023,1.3112000226974487,1.2747935891151427,1.266207993030548,0.021093073517695057,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,1.0,0.78125,0.11000099875256815,1.296875,6.991308871213679,0.062183380126953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7388160228729248,1.8077759742736816,1.772764801979065,1.772704005241394,0.021056644077284283,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,1.0,0.78125,0.0864630150197678,1.1796875,6.994552526394139,0.048796653747558594,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,2.6563520431518555,2.7063679695129395,2.6785055875778196,2.6791679859161377,0.01639963463052944,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,1.0,0.8671875,0.06127686514721937,1.16015625,6.997291583027146,0.03497934341430664,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,4.386879920959473,4.452256202697754,4.4108480453491214,4.406303882598877,0.019768161791937636,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,1.0,0.884765625,0.041723768525324195,1.1171875,6.998746434318934,0.02298593521118164,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.24208000302314758,0.4028480052947998,0.3041536003351212,0.277103990316391,0.05660721484881584,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.2447039932012558,0.30502399802207947,0.26446720361709597,0.26265600323677063,0.016744548364435372,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.2337920069694519,0.2881599962711334,0.26074880361557007,0.264384001493454,0.016469850143940968,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.23369599878787994,0.28591999411582947,0.25465920120477675,0.25385600328445435,0.01593204652619194,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.2295999974012375,0.26556798815727234,0.24674240052700042,0.2497600018978119,0.010345732066199003,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.21721599996089935,0.29020801186561584,0.2394208014011383,0.2346400022506714,0.018747330357768585,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2717759907245636,0.305184006690979,0.28813759982585907,0.28809599578380585,0.01183422600545854,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.3917759954929352,0.43772798776626587,0.41130879521369934,0.4131519943475723,0.012992473640805227,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.6176639795303345,0.7009919881820679,0.642767995595932,0.6330719888210297,0.024074084919938756,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.0820800065994263,1.1674879789352417,1.1034304022789,1.0977439880371094,0.0233067292981566,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.9809919595718384,2.0415360927581787,2.003715181350708,1.992751955986023,0.022066076434645737,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,3.790112018585205,3.8651199340820312,3.829139161109924,3.8230879306793213,0.025177464160110564,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.212351992726326,0.24383999407291412,0.22760000079870224,0.22723200172185898,0.01050568575837594,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.47494399547576904,0.5184000134468079,0.4920704007148743,0.49169600009918213,0.011991064701471855,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.3984375,0.0,1.3919410907075054,6.0,5.570159765557392,0.667236328125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.6360960006713867,0.7004479765892029,0.6608384013175964,0.6572319865226746,0.020416877242438597,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.625,0.0,1.0307764064044151,4.0,6.138251855282827,0.5382080078125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.780896008014679,0.8301439881324768,0.80346559882164,0.8030399978160858,0.016801230312128875,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.859375,0.0,0.6903350635742038,3.0,6.5943747091218174,0.38067626953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.8607040047645569,0.9195200204849243,0.8783008038997651,0.8751039803028107,0.01719059253115595,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.9765625,0.0,0.49410588440130926,2.75,6.814452474347134,0.271270751953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.833952009677887,0.894752025604248,0.8619967997074127,0.863215982913971,0.018716378851797198,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,1.0,0.25,0.33693529145074724,2.125,6.9186075263155535,0.1867218017578125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.834879994392395,0.8871039748191833,0.8651552021503448,0.8638879954814911,0.015262430894400969,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,1.0,0.375,0.25567294018677456,1.8125,6.952441049154937,0.14349365234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.8518080115318298,0.9097599983215332,0.8810272097587586,0.8751040101051331,0.017005819633271906,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,1.0,0.625,0.1747801353218523,1.53125,6.978069554482723,0.09820938110351562,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.8470079898834229,0.9010239839553833,0.8694015920162201,0.8716959953308105,0.016138881171190216,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,1.0,0.65625,0.1158122428154187,1.3125,6.9901908183358845,0.06445503234863281,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.1698240041732788,1.2311359643936157,1.1888479948043824,1.1890720129013062,0.017978797794492758,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,1.0,0.78125,0.08347181893108634,1.1796875,6.994921772573154,0.046871185302734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.7702720165252686,1.8097599744796753,1.7919103980064393,1.7956640124320984,0.012641295676021557,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,1.0,0.78125,0.06866734477822484,1.20703125,6.996602562938728,0.03801727294921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.968672037124634,3.0278079509735107,2.9899007797241213,2.9824799299240112,0.01816414122631667,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,1.0,0.8984375,0.04399546833977376,1.126953125,6.998607314922362,0.024699926376342773,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.19574399292469025,0.2512960135936737,0.21939200013875962,0.2199999988079071,0.017532156418212565,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.20483200252056122,0.24006399512290955,0.22215040028095245,0.22433599829673767,0.00969639786132892,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.20559999346733093,0.24726399779319763,0.22126719802618028,0.22207999974489212,0.01328093478485499,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.20003199577331543,0.2301120012998581,0.21453119963407516,0.21598400175571442,0.010402239855151332,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.19551999866962433,0.22972799837589264,0.21238719969987868,0.21488000452518463,0.010706095835489097,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.19420799612998962,0.2903999984264374,0.2211231991648674,0.21476799994707108,0.025955008886196004,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2290560007095337,0.289247989654541,0.2543327987194061,0.24939200282096863,0.01663236753503115,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.30831998586654663,0.36953601241111755,0.3324000000953674,0.3288639932870865,0.018617999572156707,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.462911993265152,0.5497919917106628,0.4893856018781662,0.4816960096359253,0.023348152887178286,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,0.7662079930305481,0.8717759847640991,0.788454395532608,0.7744799852371216,0.03174746482604812,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.363935947418213,1.4143040180206299,1.3812703967094422,1.3798720240592957,0.014770450075530007,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.5507519245147705,2.680704116821289,2.579859209060669,2.566223978996277,0.03673283558365955,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/allreduce.json b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/allreduce.json new file mode 100644 index 0000000..aae1f0a --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/allreduce.json @@ -0,0 +1,824 @@ +{ + "environment": [ + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + } + ], + "frontier_consumption": "diagnostic_only_in_base_profile_only_run; measured lookup requires a separate CC-backend injection ablation", + "rows": [ + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08288000151515007, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.4872959852218628, + "mean": 0.1310015980154276, + "median": 0.07679999992251396, + "min": 0.06217600032687187, + "std": 0.12790721677293843 + }, + { + "max": 0.4402880072593689, + "mean": 0.12842560112476348, + "median": 0.08288000151515007, + "min": 0.0655680000782013, + "std": 0.11220176524616535 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0793600007891655, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.12716799974441528, + "mean": 0.07871360033750534, + "median": 0.0759200006723404, + "min": 0.06032000109553337, + "std": 0.019331314939874535 + }, + { + "max": 0.12380799651145935, + "mean": 0.08059840016067028, + "median": 0.0793600007891655, + "min": 0.06217600032687187, + "std": 0.01687088356254092 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0713919997215271, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12697599828243256, + "mean": 0.0767391998320818, + "median": 0.07078400254249573, + "min": 0.05910399928689003, + "std": 0.018775178979463278 + }, + { + "max": 0.11430399864912033, + "mean": 0.07594559974968433, + "median": 0.0713919997215271, + "min": 0.06124800071120262, + "std": 0.015719922150631036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08056000247597694, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.1037760004401207, + "mean": 0.07954559996724128, + "median": 0.08056000247597694, + "min": 0.05955199897289276, + "std": 0.0135697420393132 + }, + { + "max": 0.10608000308275223, + "mean": 0.07971520014107228, + "median": 0.07593599706888199, + "min": 0.06028800085186958, + "std": 0.015319849772775456 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0865279994904995, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.14470399916172028, + "mean": 0.09121599942445754, + "median": 0.0865279994904995, + "min": 0.06441599875688553, + "std": 0.024893837894277456 + }, + { + "max": 0.12438400089740753, + "mean": 0.08531199917197227, + "median": 0.08031999692320824, + "min": 0.06364800035953522, + "std": 0.01878029830059533 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07135999947786331, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.11753600090742111, + "mean": 0.07606079950928687, + "median": 0.07135999947786331, + "min": 0.05843200162053108, + "std": 0.01755519771639284 + }, + { + "max": 0.1103999987244606, + "mean": 0.07607359997928143, + "median": 0.07073600217700005, + "min": 0.05721599981188774, + "std": 0.016904445220949783 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07321599870920181, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.11740799993276596, + "mean": 0.07749119997024537, + "median": 0.07203200086951256, + "min": 0.05862399935722351, + "std": 0.017701381594822835 + }, + { + "max": 0.11382400244474411, + "mean": 0.07733759954571724, + "median": 0.07321599870920181, + "min": 0.059039998799562454, + "std": 0.017188890882557036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09025600180029869, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.13980799913406372, + "mean": 0.0950367994606495, + "median": 0.09025600180029869, + "min": 0.06815999746322632, + "std": 0.022829835127539378 + }, + { + "max": 0.14764800667762756, + "mean": 0.09710080176591873, + "median": 0.08720000088214874, + "min": 0.07152000069618225, + "std": 0.02500574954063789 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08083200082182884, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.15574400126934052, + "mean": 0.08760640025138855, + "median": 0.07846399769186974, + "min": 0.07097599655389786, + "std": 0.024487311423551025 + }, + { + "max": 0.16284799575805664, + "mean": 0.08963519930839539, + "median": 0.08083200082182884, + "min": 0.07100799679756165, + "std": 0.026047320562445356 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.10891199856996536, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.1582079976797104, + "mean": 0.11541439890861512, + "median": 0.10836799815297127, + "min": 0.09062399715185165, + "std": 0.018285136316576037 + }, + { + "max": 0.1578879952430725, + "mean": 0.11537599861621857, + "median": 0.10891199856996536, + "min": 0.0960640013217926, + "std": 0.018246538626977286 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1703840047121048, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.19327999651432037, + "mean": 0.1707327976822853, + "median": 0.1703840047121048, + "min": 0.14815999567508698, + "std": 0.014211056022719618 + }, + { + "max": 0.19276799261569977, + "mean": 0.1658592015504837, + "median": 0.16379200667142868, + "min": 0.14560000598430634, + "std": 0.013840249648693638 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.25539200007915497, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.2807359993457794, + "mean": 0.25750079900026324, + "median": 0.25539200007915497, + "min": 0.24624000489711761, + "std": 0.008925204570802302 + }, + { + "max": 0.2863999903202057, + "mean": 0.2585055992007256, + "median": 0.255280002951622, + "min": 0.24371199309825897, + "std": 0.012059738582656496 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1021759994328022, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.9770879745483398, + "mean": 0.1974783968180418, + "median": 0.10099200159311295, + "min": 0.05913599953055382, + "std": 0.2660581738745569 + }, + { + "max": 0.892799973487854, + "mean": 0.18164799660444259, + "median": 0.1021759994328022, + "min": 0.06435199826955795, + "std": 0.23947520188197013 + }, + { + "max": 0.6467199921607971, + "mean": 0.15839359983801843, + "median": 0.10100800171494484, + "min": 0.06800000369548798, + "std": 0.16617013141866102 + }, + { + "max": 0.6725760102272034, + "mean": 0.15686400160193442, + "median": 0.10044800117611885, + "min": 0.06063999980688095, + "std": 0.17523222161300497 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12694399803876877, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.6659520268440247, + "mean": 0.1927648030221462, + "median": 0.12694399803876877, + "min": 0.07609599828720093, + "std": 0.16697784531907736 + }, + { + "max": 0.695360004901886, + "mean": 0.19356480240821838, + "median": 0.11726400256156921, + "min": 0.0796160027384758, + "std": 0.17588096678867862 + }, + { + "max": 0.5939840078353882, + "mean": 0.1868800014257431, + "median": 0.12379200011491776, + "min": 0.07427199929952621, + "std": 0.1463231714943902 + }, + { + "max": 0.6635839939117432, + "mean": 0.1874335989356041, + "median": 0.12014400213956833, + "min": 0.07526399940252304, + "std": 0.16680959677760304 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09161599725484848, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12665599584579468, + "mean": 0.09443839862942696, + "median": 0.09124799817800522, + "min": 0.06431999802589417, + "std": 0.0205690155775906 + }, + { + "max": 0.1303039938211441, + "mean": 0.09712959825992584, + "median": 0.09161599725484848, + "min": 0.07648000121116638, + "std": 0.019072048129173236 + }, + { + "max": 0.13836799561977386, + "mean": 0.09821119979023933, + "median": 0.09148800000548363, + "min": 0.0727040022611618, + "std": 0.021314066545189116 + }, + { + "max": 0.12992000579833984, + "mean": 0.09272959977388381, + "median": 0.08931199833750725, + "min": 0.06406400352716446, + "std": 0.021360803830354765 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08580800145864487, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.13449600338935852, + "mean": 0.08960640132427215, + "median": 0.08299200236797333, + "min": 0.06796800345182419, + "std": 0.020633597898445998 + }, + { + "max": 0.14735999703407288, + "mean": 0.09248319901525974, + "median": 0.08580800145864487, + "min": 0.05913599953055382, + "std": 0.025960234928829564 + }, + { + "max": 0.13705599308013916, + "mean": 0.08947199806571007, + "median": 0.08460799977183342, + "min": 0.0634239986538887, + "std": 0.02122838946992339 + }, + { + "max": 0.13846400380134583, + "mean": 0.0859104000031948, + "median": 0.08308799937367439, + "min": 0.05974400043487549, + "std": 0.022373631424433445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09867199882864952, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.1279039978981018, + "mean": 0.09748799875378608, + "median": 0.09678399935364723, + "min": 0.0655359998345375, + "std": 0.02315719144614622 + }, + { + "max": 0.1356479972600937, + "mean": 0.10018239840865135, + "median": 0.09532799944281578, + "min": 0.06185600161552429, + "std": 0.02330700253650042 + }, + { + "max": 0.13142399489879608, + "mean": 0.09778879955410957, + "median": 0.09492799639701843, + "min": 0.06560000032186508, + "std": 0.02307579212430045 + }, + { + "max": 0.1276479959487915, + "mean": 0.09611519873142242, + "median": 0.09867199882864952, + "min": 0.0642239972949028, + "std": 0.02273612181973431 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09646400064229965, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.14060799777507782, + "mean": 0.09770880043506622, + "median": 0.09115200117230415, + "min": 0.06950400024652481, + "std": 0.024752645089849798 + }, + { + "max": 0.14377599954605103, + "mean": 0.09824960008263588, + "median": 0.08999999985098839, + "min": 0.07103999704122543, + "std": 0.025403407389046027 + }, + { + "max": 0.13680000603199005, + "mean": 0.09993600100278854, + "median": 0.09646400064229965, + "min": 0.06790400296449661, + "std": 0.022297424273985882 + }, + { + "max": 0.1391039937734604, + "mean": 0.09769919961690902, + "median": 0.09601600095629692, + "min": 0.06835199892520905, + "std": 0.023899922100804445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08377600088715553, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.1464959979057312, + "mean": 0.09276160076260567, + "median": 0.08377600088715553, + "min": 0.06777600198984146, + "std": 0.024751086465295658 + }, + { + "max": 0.14319999516010284, + "mean": 0.09080640003085136, + "median": 0.080400001257658, + "min": 0.06796800345182419, + "std": 0.023365718881708488 + }, + { + "max": 0.1382399946451187, + "mean": 0.09063360020518303, + "median": 0.08193599805235863, + "min": 0.06627199798822403, + "std": 0.02372618650854302 + }, + { + "max": 0.14313599467277527, + "mean": 0.09044799953699112, + "median": 0.08128000050783157, + "min": 0.06652799993753433, + "std": 0.02462486862033686 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1128000020980835, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.1600320041179657, + "mean": 0.11426239982247352, + "median": 0.1128000020980835, + "min": 0.07657600194215775, + "std": 0.030240087702350687 + }, + { + "max": 0.15881599485874176, + "mean": 0.11206399947404862, + "median": 0.10628800094127655, + "min": 0.0772159993648529, + "std": 0.029933135345483627 + }, + { + "max": 0.15612800419330597, + "mean": 0.10761600062251091, + "median": 0.09860799834132195, + "min": 0.07689599692821503, + "std": 0.02634237020678647 + }, + { + "max": 0.15865600109100342, + "mean": 0.11094079986214637, + "median": 0.10979199782013893, + "min": 0.07583999633789062, + "std": 0.028619217028542445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08755199983716011, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.11999999731779099, + "mean": 0.0843871995806694, + "median": 0.08755199983716011, + "min": 0.06332799792289734, + "std": 0.016638543890716024 + }, + { + "max": 0.1218239963054657, + "mean": 0.08518079966306687, + "median": 0.08032000064849854, + "min": 0.06393600255250931, + "std": 0.019900899429956945 + }, + { + "max": 0.11849600076675415, + "mean": 0.0843968003988266, + "median": 0.08702399954199791, + "min": 0.06297600269317627, + "std": 0.017031908773433545 + }, + { + "max": 0.12300799787044525, + "mean": 0.0846304003149271, + "median": 0.08139199763536453, + "min": 0.06195199862122536, + "std": 0.020106523698622265 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12361599877476692, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.17155200242996216, + "mean": 0.12168959975242614, + "median": 0.12055999785661697, + "min": 0.09609600156545639, + "std": 0.022565485532483928 + }, + { + "max": 0.9246399998664856, + "mean": 0.19978560134768486, + "median": 0.12361599877476692, + "min": 0.09715200215578079, + "std": 0.24230694305662265 + }, + { + "max": 0.9317439794540405, + "mean": 0.20037759989500045, + "median": 0.12327999994158745, + "min": 0.09603200107812881, + "std": 0.24450903278000383 + }, + { + "max": 0.9321280121803284, + "mean": 0.19875840097665787, + "median": 0.12230399996042252, + "min": 0.0950080007314682, + "std": 0.24519252220785093 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.20030399411916733, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.3261440098285675, + "mean": 0.20826880186796187, + "median": 0.19974400103092194, + "min": 0.1409280002117157, + "std": 0.051397264855718945 + }, + { + "max": 0.3248000144958496, + "mean": 0.20548800230026246, + "median": 0.1979840025305748, + "min": 0.141184002161026, + "std": 0.04980002399656717 + }, + { + "max": 0.32547199726104736, + "mean": 0.21280319690704347, + "median": 0.20030399411916733, + "min": 0.14127999544143677, + "std": 0.05304243085685509 + }, + { + "max": 0.26047998666763306, + "mean": 0.1969312012195587, + "median": 0.18079999834299088, + "min": 0.14057600498199463, + "std": 0.042174123638424224 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.2924960106611252, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.45052799582481384, + "mean": 0.3013375997543335, + "median": 0.28273600339889526, + "min": 0.21334399282932281, + "std": 0.07214223764527236 + }, + { + "max": 0.4466240108013153, + "mean": 0.2968191936612129, + "median": 0.27796798944473267, + "min": 0.21241599321365356, + "std": 0.07400678240849741 + }, + { + "max": 0.3830080032348633, + "mean": 0.2946112036705017, + "median": 0.2924960106611252, + "min": 0.21084800362586975, + "std": 0.05323627615746332 + }, + { + "max": 0.4609600007534027, + "mean": 0.3054272010922432, + "median": 0.289792001247406, + "min": 0.21062399446964264, + "std": 0.07608482904865672 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + } + ], + "schema_version": "qwen30_vllm020_allreduce_frozen.v1" +} diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention.csv new file mode 100644 index 0000000..835d3e5 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention.csv @@ -0,0 +1,214 @@ +time_stats.attn_input_reshape.min,time_stats.attn_input_reshape.max,time_stats.attn_input_reshape.mean,time_stats.attn_input_reshape.median,time_stats.attn_input_reshape.std,time_stats.attn_kv_cache_save.min,time_stats.attn_kv_cache_save.max,time_stats.attn_kv_cache_save.mean,time_stats.attn_kv_cache_save.median,time_stats.attn_kv_cache_save.std,time_stats.attn_prefill.min,time_stats.attn_prefill.max,time_stats.attn_prefill.mean,time_stats.attn_prefill.median,time_stats.attn_prefill.std,time_stats.attn_decode.min,time_stats.attn_decode.max,time_stats.attn_decode.mean,time_stats.attn_decode.median,time_stats.attn_decode.std,time_stats.attn_output_reshape.min,time_stats.attn_output_reshape.max,time_stats.attn_output_reshape.mean,time_stats.attn_output_reshape.median,time_stats.attn_output_reshape.std,n_embd,n_q_head,n_kv_head,block_size,num_tensor_parallel_workers,max_model_len,batch_size,prefill_chunk_size,kv_cache_size,is_prefill,attention_backend,is_mixed_batch,mode,seq_lens,total_tokens,max_seq_len,min_seq_len,avg_seq_len,equal_seq_len,seq_len_variance,seq_len_std,seq_len_cv,is_chunked_prefill_sample,chunk_start_token,chunk_end_token,total_prefill_tokens,profiling_precision,model_arch,quant_signature,measurement_type,is_true_mixed_batch,prefill_seq_lens,prefill_kv_cache_sizes,decode_kv_cache_sizes,num_prefill_seqs,num_decode_seqs,decode_batch_size,total_batch_size,total_decode_tokens,decode_avg_kv_cache_size,batch_composition_ratio,batch_spec,projection_policy +0.0,0.0,0.0,0.0,0.0,0.01414399966597557,0.028863999992609024,0.019705599918961526,0.01771199982613325,0.005157200849836681,0.047968000173568726,0.07046400010585785,0.05810240097343922,0.05810240097343922,0.007477463486041561,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04947200044989586,0.020412799902260303,0.01635199971497059,0.010107497379722417,0.046560000628232956,0.08323200047016144,0.05587520003318787,0.05587520003318787,0.011126758739503428,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.022207999601960182,0.017033600155264138,0.015312000177800655,0.002819235991970241,0.05104000121355057,0.07692799717187881,0.056396800279617305,0.056396800279617305,0.007481982178637539,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.022272000089287758,0.01706880023702979,0.01616000011563301,0.002460889579319197,0.06931199878454208,0.0838719978928566,0.07432000041007995,0.07432000041007995,0.004777766433175866,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.018592000007629395,0.028543999418616295,0.02095999978482723,0.019183999858796597,0.003198175496053494,0.12179200351238251,0.15408000349998474,0.1307712011039257,0.1307712011039257,0.00858807797538298,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027775999158620834,0.03385600075125694,0.030131200328469276,0.029680000618100166,0.0021152558103575215,0.32678401470184326,0.3450239896774292,0.33396480381488797,0.33396480381488797,0.0045872424917606375,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04438399896025658,0.05084799975156784,0.046540799736976626,0.04531199857592583,0.002277905811237223,1.0959680080413818,1.1151360273361206,1.0999775886535645,1.0999775886535645,0.005694403246120485,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.078015998005867,0.08691199868917465,0.08114239946007729,0.08019199967384338,0.00292795706334475,4.070400238037109,4.113152027130127,4.087088012695312,4.087088012695312,0.013660567012509554,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.05167999863624573,0.022115200012922286,0.017583999782800674,0.010340094822340818,0.05196800082921982,0.09011200070381165,0.06328320093452933,0.06328320093452933,0.012557341255467452,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01583999954164028,0.026623999699950218,0.018927999772131443,0.017376000061631203,0.003514650316260619,0.06681600213050842,0.07993599772453308,0.0725280001759529,0.0725280001759529,0.004343502558613716,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01648000068962574,0.030880000442266464,0.01945280022919178,0.017967999912798405,0.004096211183007485,0.1311360001564026,0.1546880006790161,0.13908160030841826,0.13908160030841826,0.007511906874366178,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.03558399900794029,0.020851199887692927,0.018559999763965607,0.005235911594130716,0.32950401306152344,0.350271999835968,0.33912960588932034,0.33912960588932034,0.006027400986663648,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.019328000023961067,0.040608000010252,0.022790400311350822,0.020704000256955624,0.006113051965778337,1.1415679454803467,1.1518720388412476,1.144483208656311,1.144483208656311,0.0032332311374389127,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015807999297976494,0.030688000842928886,0.019596799835562707,0.01774400006979704,0.004343384771033462,0.0,0.0,0.0,0.0,0.0,0.049056001007556915,0.07580800354480743,0.05948160067200661,0.05948160067200661,0.009031541471446955,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.02486399933695793,0.01923839971423149,0.018079999834299088,0.0032282528537266424,0.0,0.0,0.0,0.0,0.0,0.05142400041222572,0.07353600114583969,0.059328000620007516,0.059328000620007516,0.0073307807735143084,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016543999314308167,0.03977600112557411,0.021379199624061585,0.018511999398469925,0.006593576176246171,0.0,0.0,0.0,0.0,0.0,0.0488319993019104,0.06435199826955795,0.05479039996862411,0.05479039996862411,0.005672522998491864,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01635199971497059,0.02844800055027008,0.019267200119793416,0.017952000722289085,0.0035068687666949577,0.0,0.0,0.0,0.0,0.0,0.049855999648571014,0.07798399776220322,0.05986879989504815,0.05986879989504815,0.01043914754878828,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016383999958634377,0.026079999282956123,0.01923519968986511,0.017791999503970146,0.0032161974331284568,0.0,0.0,0.0,0.0,0.0,0.058111999183893204,0.1045759990811348,0.06708480007946492,0.06708480007946492,0.013479022462646494,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.04057599976658821,0.019100800156593323,0.015455999877303839,0.007512281243011577,0.0,0.0,0.0,0.0,0.0,0.05363199859857559,0.07782399654388428,0.06090559959411622,0.06090559959411622,0.007544620176348091,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.02191999927163124,0.016684799920767546,0.01563199982047081,0.0024293118621811216,0.0,0.0,0.0,0.0,0.0,0.0629120022058487,0.07891199737787247,0.06891520097851753,0.06891520097851753,0.005472695665695425,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.038943998515605927,0.018313600029796363,0.01561600062996149,0.007127270260115769,0.0,0.0,0.0,0.0,0.0,0.08675199747085571,0.10662399977445602,0.09391999915242194,0.09391999915242194,0.006988099589086635,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.054687999188899994,0.021439999900758268,0.01539199985563755,0.012052764849597775,0.0,0.0,0.0,0.0,0.0,0.13488000631332397,0.1528639942407608,0.1431359991431236,0.1431359991431236,0.005436271464033599,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.041919998824596405,0.01899839974939823,0.015343999955803156,0.007989843526623287,0.0,0.0,0.0,0.0,0.0,0.06176000088453293,0.08374399691820145,0.06747519969940186,0.06747519969940186,0.0066067747128778,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016256000846624374,0.11353600025177002,0.042761600017547606,0.028960000723600388,0.029104301538020762,0.0,0.0,0.0,0.0,0.0,0.09734400361776352,0.14422400295734406,0.11392960175871848,0.11392960175871848,0.013198594600417867,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.034143999218940735,0.018918400071561335,0.01643200032413006,0.005500943993080684,0.0,0.0,0.0,0.0,0.0,0.12918399274349213,0.15087999403476715,0.13807999789714814,0.13807999789714814,0.007658330538677587,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.03641600161790848,0.0198208000510931,0.01780799962580204,0.0057264128169845765,0.0,0.0,0.0,0.0,0.0,0.22099199891090393,0.23904000222682953,0.2293503984808922,0.2293503984808922,0.004861342907006028,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.035840000957250595,0.018908800091594458,0.015792000107467175,0.0064374817924757475,0.0,0.0,0.0,0.0,0.0,0.10134399682283401,0.12201599776744843,0.10896319895982742,0.10896319895982742,0.006336330809165179,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.03417599946260452,0.017846399918198586,0.014800000004470348,0.006495007539635255,0.0,0.0,0.0,0.0,0.0,0.13126400113105774,0.15561600029468536,0.1389280006289482,0.1389280006289482,0.008381472811075022,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03519999980926514,0.019168000388890504,0.01600000075995922,0.005995522477654695,0.0,0.0,0.0,0.0,0.0,0.21728000044822693,0.2343679964542389,0.2231455981731415,0.2231455981731415,0.004720730646739123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.03670400008559227,0.018441599886864425,0.015023999847471714,0.006596535162793127,0.0,0.0,0.0,0.0,0.0,0.39321601390838623,0.4524799883365631,0.4058080047369003,0.4058080047369003,0.01578349755088941,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.022143999114632607,0.016912000067532063,0.015168000012636185,0.0028156089295136347,0.0,0.0,0.0,0.0,0.0,0.15587200224399567,0.3079040050506592,0.17838079929351805,0.17838079929351805,0.04355865575265927,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.02112000063061714,0.015516800060868263,0.01488000014796853,0.001940870731593904,0.0,0.0,0.0,0.0,0.0,0.21587200462818146,0.23561599850654602,0.22250880002975468,0.22250880002975468,0.006181951170646666,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.039264000952243805,0.023609600123018028,0.02131200022995472,0.007236979625711548,0.0,0.0,0.0,0.0,0.0,0.408735990524292,0.470335990190506,0.4336863994598388,0.4336863994598388,0.01844662383160074,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.025407999753952026,0.016227199975401164,0.014960000291466713,0.0031617375441736185,0.0,0.0,0.0,0.0,0.0,0.7412800192832947,0.7627840042114258,0.7464000046253203,0.7464000046253203,0.006112167448837547,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.02812799997627735,0.020652799773961304,0.021359999664127827,0.004308706957613102,0.028383498565450627,0.039859687970646644,0.032492258074592426,0.032492258074592426,0.00453597266208597,0.029312501176103633,0.04116431058787463,0.03355574193327539,0.03355574193327539,0.004684436757701648,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.034143999218940735,0.023171199765056372,0.024255999363958836,0.005908614918096971,0.03333159243114438,0.038935341782478095,0.03580428402241854,0.03580428402241854,0.002082270297044095,0.03633240903369937,0.04244065945634484,0.03902771507087562,0.03902771507087562,0.0022697354261490147,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.02611199952661991,0.019635199941694735,0.018112000077962875,0.0038298050749257795,0.04189529417991216,0.057484239920526384,0.04744885718421094,0.04744885718421094,0.004779830748455743,0.051672703037266184,0.07089975418195164,0.05852234134479412,0.05852234134479412,0.005895334539786378,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.029311999678611755,0.018662399891763926,0.01673599984496832,0.0044017112162725355,0.04322973959325901,0.05049827064705393,0.04535414343408448,0.04535414343408448,0.0022944156000240697,0.08733025617719538,0.10201372836398578,0.09162185574241775,0.09162185574241775,0.004635047631846023,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.026367999613285065,0.016336000058799982,0.01515199989080429,0.003415353455946402,0.06031842775160765,0.06618323188375198,0.06274415549817247,0.06274415549817247,0.001985647289644255,0.14768157653992678,0.16204076249052324,0.15362064543185072,0.15362064543185072,0.004861590945216247,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.02454400062561035,0.017167999967932702,0.016191999427974224,0.0028685016454498436,0.09128700688359712,0.09689150775996329,0.09360396051475776,0.09360396051475776,0.0013477542744916764,0.2740889887523892,0.2909164873209846,0.2810456356995366,0.2810456356995366,0.004046628526808561,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01881599985063076,0.03097599931061268,0.024598400108516216,0.024848000146448612,0.0037539437391565975,0.1035249255866932,0.10603132147437412,0.10399797220840973,0.10399797220840973,0.0007025109429056814,0.5652750707893444,0.5789606938874117,0.5678580377517648,0.5678580377517648,0.003835906384194897,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018624000251293182,0.043487999588251114,0.024460799992084503,0.019600000232458115,0.008922081629781394,0.196169204945307,0.2076140047945799,0.2008444429250931,0.2008444429250931,0.00394350953292805,1.1196708009293268,1.1849940417370974,1.1463555573610091,1.1463555573610091,0.022508285530529783,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.027135999873280525,0.03667199984192848,0.02945920005440712,0.028256000019609928,0.0029446678408169553,0.37757279619664613,0.3898113624476372,0.38075520430942184,0.38075520430942184,0.0036600203831042254,0.2522831942132049,0.2604606493092597,0.25440958703617433,0.25440958703617433,0.0024455194930253126,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.0435199998319149,0.049695998430252075,0.04502719938755036,0.04395199939608574,0.002035785660169308,1.1048984388245497,1.1189053886476108,1.1112371236754128,1.1112371236754128,0.0050220696724624985,0.1395495077239122,0.14131859607071193,0.14035008841031085,0.14035008841031085,0.0006342911944856293,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018783999606966972,0.024288000538945198,0.020627199858427047,0.019567999988794327,0.0019457052717059358,0.09455999732017517,0.12185599654912949,0.10207359939813614,0.10207359939813614,0.00753346544014261,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027295999228954315,0.034752000123262405,0.02943360023200512,0.028528000228106976,0.0023159160681123767,0.14416000247001648,0.15904000401496887,0.14979200065135959,0.14979200065135959,0.00480624675866005,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.043455999344587326,0.05084799975156784,0.045500800386071204,0.04411200061440468,0.002490556062831049,0.24454399943351746,0.25865599513053894,0.2504959970712662,0.2504959970712662,0.003778494140301353,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.07673600316047668,0.08374399691820145,0.07970559895038605,0.07873599976301193,0.0022907976135004057,0.445248007774353,0.4758400022983551,0.45409599840641024,0.45409599840641024,0.009133230340383306,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.043296001851558685,0.051072001457214355,0.044972800090909,0.04399999976158142,0.002451216824441962,0.6147199869155884,0.6290879845619202,0.6204927921295166,0.6204927921295166,0.004344846739788608,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.07737600058317184,0.09123200178146362,0.08094720020890236,0.07980800047516823,0.004065264798368611,1.1744320392608643,1.1887680292129517,1.178323209285736,1.178323209285736,0.004395876469319665,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.03254399821162224,0.01810879958793521,0.01611199975013733,0.005058478889747962,0.0,0.0,0.0,0.0,0.0,0.06339199841022491,0.0841279998421669,0.07019519805908202,0.07019519805908202,0.006070126182471763,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.030271999537944794,0.018124799989163876,0.015711999498307705,0.004696400814632773,0.0,0.0,0.0,0.0,0.0,0.2739199995994568,0.2922239899635315,0.28281279802322384,0.28281279802322384,0.00633037978599564,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.022431999444961548,0.016844799742102623,0.015471999999135733,0.0027624803153841917,0.0,0.0,0.0,0.0,0.0,0.3909119963645935,0.4160960018634796,0.39996159672737125,0.39996159672737125,0.007037174636804898,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014240000396966934,0.035071998834609985,0.018764800019562246,0.015583999920636415,0.006305594700523817,0.0,0.0,0.0,0.0,0.0,0.7383679747581482,0.7597119808197021,0.7430047929286957,0.7430047929286957,0.006192645960911466,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.028991999104619026,0.019401599932461978,0.01665600063279271,0.005434953793780546,0.0,0.0,0.0,0.0,0.0,1.427008032798767,1.4517120122909546,1.4361984014511109,1.4361984014511109,0.008802755821028187,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.02844800055027008,0.01820160010829568,0.014944000169634819,0.004942002036909087,0.0,0.0,0.0,0.0,0.0,0.08361600339412689,0.11027199774980545,0.09058240056037903,0.09058240056037903,0.00872938604423622,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.023104000836610794,0.01648960020393133,0.015632000286132097,0.0027628828340349578,0.0,0.0,0.0,0.0,0.0,0.5063040256500244,0.5497599840164185,0.5168287932872773,0.5168287932872773,0.013557229606313293,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01616000011563301,0.029055999591946602,0.018780799955129622,0.017215999774634838,0.003794434947431941,0.0,0.0,0.0,0.0,0.0,0.7439360022544861,0.7719680070877075,0.7502080142498017,0.7502080142498017,0.00839205598262567,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015968000516295433,0.03046399913728237,0.018822400271892546,0.01726400014013052,0.004212536831927695,0.0,0.0,0.0,0.0,0.0,1.4256000518798828,1.449504017829895,1.4325888037681578,1.4325888037681578,0.0065448028108711165,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.027712000533938408,0.01633920017629862,0.014944000169634819,0.0038488220071983326,0.0,0.0,0.0,0.0,0.0,2.798719882965088,3.0184640884399414,2.8293471813201903,2.8293471813201903,0.06375662767704417,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.021407999098300934,0.01637439979240298,0.014992000069469213,0.0024463594774459265,0.0,0.0,0.0,0.0,0.0,0.09347199648618698,0.10473600029945374,0.09820479974150656,0.09820479974150656,0.004061668684999473,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.04790399968624115,0.020179199893027543,0.016447999514639378,0.009635010283506967,0.0,0.0,0.0,0.0,0.0,0.6221439838409424,0.6444799900054932,0.6277984082698822,0.6277984082698822,0.007234140179397069,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.03276799991726875,0.01921279989182949,0.015232000034302473,0.0065506148511265076,0.0,0.0,0.0,0.0,0.0,0.9111359715461731,0.9307839870452881,0.9151648044586183,0.9151648044586183,0.005528712062927265,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.02707199938595295,0.016512000095099212,0.014688000082969666,0.0038989080209321414,0.0,0.0,0.0,0.0,0.0,1.7645119428634644,1.7849279642105103,1.7684095859527589,1.7684095859527589,0.005876963340184193,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014240000396966934,0.029184000566601753,0.016825600154697896,0.014752000104635954,0.00457910514148248,0.0,0.0,0.0,0.0,0.0,3.4781761169433594,3.5388801097869873,3.490892815589905,3.490892815589905,0.01818910600692522,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014976000413298607,0.03139200061559677,0.01898880014196038,0.016848000697791576,0.005083715524393418,0.13752702814163098,0.14117629917511842,0.13848196486571557,0.13848196486571557,0.0009880564964713527,0.523336967410432,0.5372236808930884,0.5269708253945994,0.5269708253945994,0.00375988994658546,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,520,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,8,8,9,8,16384.0,0.1111111111111111,q512_8q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01833599992096424,0.02707199938595295,0.020995199866592883,0.01976000051945448,0.0028392656352050185,0.17817885890237703,0.18622915300900206,0.18109068484526028,0.18109068484526028,0.0027040006080457624,0.5449571488834201,0.5695788427872015,0.5538629212834322,0.5538629212834322,0.008270141985514729,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1040,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,16,16,17,16,16384.0,0.058823529411764705,q1k_16q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.02703999914228916,0.03308799862861633,0.028883199393749236,0.02759999968111515,0.002285758578932753,0.4688266550410815,0.48301665772804186,0.4727750380198454,0.4727750380198454,0.004396396389258256,1.0430453981052807,1.0746153117238624,1.051829759343436,1.051829759343436,0.009781101336187197,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,32,32,33,32,16384.0,0.030303030303030304,q2k_32q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.04368000105023384,0.059328000992536545,0.04809600040316582,0.04617599956691265,0.005158792915002645,1.3509461459747514,1.3759066409666496,1.3589365122072,1.3589365122072,0.008008877716183384,0.9213738861449998,0.9383974724214122,0.9268234851606594,0.9268234851606594,0.005462224239661061,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,16,16,17,16,32768.0,0.058823529411764705,q4k_16q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.026944000273942947,0.040511999279260635,0.030291200056672095,0.02817599941045046,0.004386386489305873,0.5071830964059985,0.5142792136001758,0.5096392092471257,0.5096392092471257,0.0022059272685869546,2.175632932188972,2.2060727208328075,2.186168772243963,2.186168772243963,0.009462633998570079,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,32,32,33,32,32768.0,0.030303030303030304,q2k_32q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018303999677300453,0.03017600066959858,0.02039040010422468,0.018943999893963337,0.0034612051983613614,0.21341429693945616,0.21552776300295579,0.21393496609766816,0.21393496609766816,0.0005736773262009222,4.617401495144284,4.663128147488988,4.628666619290515,4.628666619290515,0.012412001359412127,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,64,64,65,64,32768.0,0.015384615384615385,q1k_64q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.02364799939095974,0.01809599995613098,0.016352000646293163,0.0035481127058959038,0.04822399839758873,0.08566399663686752,0.05961279980838299,0.05961279980838299,0.011445665413968877,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.020864000543951988,0.01668160008266568,0.015664000064134598,0.0025769063833097584,0.049695998430252075,0.08057600259780884,0.05882879942655563,0.05882879942655563,0.009515126108519331,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.028704000636935234,0.017430400010198355,0.015056000091135502,0.004349294613335555,0.049855999648571014,0.07366400212049484,0.05459520071744919,0.05459520071744919,0.0069610920757925574,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.03161599859595299,0.017788799852132796,0.015711999963968992,0.005005385723318659,0.06102399900555611,0.08179199695587158,0.06650560013949873,0.06650560013949873,0.006947995595801105,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.04416000097990036,0.019200000166893005,0.015488000120967627,0.008561241323364038,0.08054400235414505,0.09196799993515015,0.08607039973139763,0.08607039973139763,0.004145329035483754,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.023391999304294586,0.019667199812829494,0.018463999964296818,0.0022577669687832585,0.18729600310325623,0.20585599541664124,0.19546559900045393,0.19546559900045393,0.006339068824303663,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.026240000501275063,0.03446400165557861,0.028297600522637367,0.027088000439107418,0.0025148441522922374,0.5754240155220032,0.5889919996261597,0.5800191938877105,0.5800191938877105,0.003858869273829596,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04182400181889534,0.047200001776218414,0.043036799877882004,0.0423360001295805,0.0017073405772076728,2.063199996948242,2.0787200927734375,2.067151999473572,2.067151999473572,0.004959651271127835,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.05648000165820122,0.02270399993285537,0.017935999669134617,0.012377915150727689,0.049536000937223434,0.07196799665689468,0.056015999615192415,0.056015999615192415,0.0070476637552742884,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01408000010997057,0.021344000473618507,0.01611520005390048,0.014928000047802925,0.0025499884993961702,0.07072000205516815,0.2642880082130432,0.1588256008923054,0.1588256008923054,0.053220347086102376,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01360000018030405,0.03014400042593479,0.016975999902933837,0.015056000091135502,0.004715159697364111,0.08393599838018417,0.11036799848079681,0.09160000011324881,0.09160000011324881,0.007911669434472792,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.020767999812960625,0.016233599931001663,0.015392000321298838,0.002202615907995427,0.1998399943113327,0.22070400416851044,0.20855360180139543,0.20855360180139543,0.00689307200230667,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.033440001308918,0.018684800155460833,0.016048000194132328,0.00553991005639174,0.6043199896812439,0.635807991027832,0.6126143991947173,0.6126143991947173,0.008745933953408096,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.03359999880194664,0.017811199743300678,0.014944000169634819,0.005926140483565472,0.0,0.0,0.0,0.0,0.0,0.045471999794244766,0.07539200037717819,0.054758400097489356,0.054758400097489356,0.010253548506101549,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02409599907696247,0.01641279999166727,0.01473599998280406,0.003347158638713987,0.0,0.0,0.0,0.0,0.0,0.0461760014295578,0.07529599964618683,0.05460800044238568,0.05460800044238568,0.009748937798340135,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.028416000306606293,0.018611199874430894,0.01598400017246604,0.005209875093863647,0.0,0.0,0.0,0.0,0.0,0.048128001391887665,0.07897599786520004,0.061353600397706036,0.061353600397706036,0.010488153655157845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.025728000327944756,0.016092800162732603,0.01512000011280179,0.003300888123052605,0.0,0.0,0.0,0.0,0.0,0.04864000156521797,0.07648000121116638,0.05810560062527656,0.05810560062527656,0.009473544218404408,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.03551999852061272,0.018588799890130757,0.016064000315964222,0.006071057437992447,0.0,0.0,0.0,0.0,0.0,0.04822399839758873,0.07862400263547897,0.05660480037331582,0.05660480037331582,0.009565394213730401,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.028831999748945236,0.01699519995599985,0.015024000313133001,0.004285000744868188,0.0,0.0,0.0,0.0,0.0,0.04854400083422661,0.06719999760389328,0.05778240002691746,0.05778240002691746,0.006852125554679805,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.024927999824285507,0.0161183999851346,0.01521599991247058,0.0029774808524673907,0.0,0.0,0.0,0.0,0.0,0.05379199981689453,0.08966399729251862,0.06270079985260964,0.06270079985260964,0.009983591277092696,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03001599945127964,0.017648000083863736,0.01547200046479702,0.004585126309484848,0.0,0.0,0.0,0.0,0.0,0.061664000153541565,0.07692799717187881,0.06704320013523103,0.06704320013523103,0.005500191798490629,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.026367999613285065,0.016947199776768684,0.015039999969303608,0.0037951566103550205,0.0,0.0,0.0,0.0,0.0,0.08799999952316284,0.111455999314785,0.0964031994342804,0.0964031994342804,0.007397541615558088,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01369599997997284,0.021247999742627144,0.015359999984502793,0.014640000183135271,0.0020942770950814317,0.0,0.0,0.0,0.0,0.0,0.051711998879909515,0.07065600156784058,0.058054400235414506,0.058054400235414506,0.006633034815910025,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.0307839997112751,0.017148799914866685,0.015039999969303608,0.004882374686312284,0.0,0.0,0.0,0.0,0.0,0.061919998377561576,0.07843200117349625,0.06628479920327664,0.06628479920327664,0.004962852689801192,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013887999579310417,0.02969600073993206,0.017500799987465142,0.015232000034302473,0.00467273319705314,0.0,0.0,0.0,0.0,0.0,0.08819200098514557,0.11097600311040878,0.09493440166115762,0.09493440166115762,0.007509235042985577,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.022112000733613968,0.01751680001616478,0.016047999262809753,0.0030306621792915785,0.0,0.0,0.0,0.0,0.0,0.13065600395202637,0.15110400319099426,0.13857279866933822,0.13857279866933822,0.00750137841249771,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.04012800008058548,0.01892479993402958,0.01550400024279952,0.007459545267816961,0.0,0.0,0.0,0.0,0.0,0.06278400123119354,0.08259200304746628,0.0704512007534504,0.0704512007534504,0.005979055984382744,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.02236800082027912,0.016332800220698118,0.014928000047802925,0.002784044363186731,0.0,0.0,0.0,0.0,0.0,0.1003199964761734,0.1279360055923462,0.10921279862523078,0.10921279862523078,0.00862769617046716,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0225600004196167,0.016128000058233737,0.014800000004470348,0.002958953332547708,0.0,0.0,0.0,0.0,0.0,0.13116799294948578,0.14812800288200378,0.13783999979496003,0.13783999979496003,0.005696384361148053,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.029983999207615852,0.017468800116330386,0.01508800033479929,0.0047379550962483065,0.0,0.0,0.0,0.0,0.0,0.217631995677948,0.2447360008955002,0.22715839892625808,0.22715839892625808,0.008831828221847138,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02304000034928322,0.016512000095099212,0.014512000139802694,0.0035026774828624254,0.0,0.0,0.0,0.0,0.0,0.11020799726247787,0.12307199835777283,0.11600959971547126,0.11600959971547126,0.004667637266950902,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.042080000042915344,0.017510399967432023,0.014607999939471483,0.00823117883530167,0.0,0.0,0.0,0.0,0.0,0.15702399611473083,0.17587199807167053,0.16399359852075576,0.16399359852075576,0.006588299074676393,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0208320003002882,0.015299199987202883,0.01462399959564209,0.001916768086505386,0.0,0.0,0.0,0.0,0.0,0.21779200434684753,0.2415360063314438,0.2260768011212349,0.2260768011212349,0.007251352080685236,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014015999622642994,0.021088000386953354,0.015619200188666582,0.01473599998280406,0.002013227452302141,0.0,0.0,0.0,0.0,0.0,0.3959999978542328,0.4152640104293823,0.40332479774951924,0.40332479774951924,0.006942401431914052,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.03379200026392937,0.020595200080424547,0.019600000232458115,0.006548754248881344,0.026192623739694512,0.040006882507168426,0.029155180178492036,0.029155180178492036,0.00408828748438241,0.024591375524545756,0.037561119537986146,0.027372820355088746,0.027372820355088746,0.0038383559348575944,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.020896000787615776,0.015318400040268899,0.014431999996304512,0.0019640312960926966,0.029349018208693862,0.06236262941356679,0.03714313592014963,0.03714313592014963,0.009618313791872569,0.028826981462526914,0.06125337308649041,0.036482463672774496,0.036482463672774496,0.009447230957011863,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.032127998769283295,0.017286399938166143,0.014479999896138906,0.005536495446636193,0.03273085874558354,0.04394578491937082,0.03563837490653034,0.03563837490653034,0.003429601730256567,0.03488514202593899,0.04683821345079978,0.037984025407085474,0.037984025407085474,0.0036553316361902684,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.021503999829292297,0.015078400075435639,0.01425600005313754,0.002238435975478849,0.042100813549974185,0.051784144690147756,0.045378692890289625,0.045378692890289625,0.003184109940650555,0.05111518843748309,0.06287185663927425,0.05509490773570219,0.05509490773570219,0.0038658725544299132,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.02687999978661537,0.017667199857532977,0.01566399959847331,0.003864811846387173,0.048475323773821306,0.05599956978723733,0.05123499252968345,0.05123499252968345,0.002427379820423941,0.08429268135885387,0.09737642835214408,0.0890914090616175,0.0890914090616175,0.0042209177332077005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.03359999880194664,0.019804799742996693,0.016032000072300434,0.006750519908145474,0.07121508474579985,0.07292308367136396,0.07194410845270183,0.07194410845270183,0.0005500065915943844,0.14760091249712767,0.15114092373010238,0.14911189243564577,0.14911189243564577,0.0011399477384397005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.021856000646948814,0.016835200227797033,0.015263999812304974,0.00283854448975065,0.08146338272142935,0.08560865714384096,0.0834932625520734,0.0834932625520734,0.0012259635905347874,0.2782486219401307,0.29240733788179385,0.2851819365989658,0.2851819365989658,0.004187435731481665,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04495999962091446,0.02143679987639189,0.015856000129133463,0.009709987872683342,0.1194459208702178,0.12302524755181463,0.12053885718696096,0.12053885718696096,0.001005118119341339,0.5597220650459199,0.5764947443228802,0.5648435507167102,0.5648435507167102,0.004709970715400788,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018112000077962875,0.026496000587940216,0.019318400137126445,0.018432000651955605,0.0024249605112359905,0.19770253574610402,0.222811797868348,0.2054811520619412,0.2054811520619412,0.008000944254868043,0.13941746080159492,0.157124211777114,0.14490284788179197,0.14490284788179197,0.005642170080515992,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.02595200017094612,0.03155200183391571,0.02727359998971224,0.026575999334454536,0.0016260948764843166,0.6014684881116659,0.6095472988212,0.6046757700946376,0.6046757700946376,0.0023537435563177303,0.11325152264578045,0.11477269562841545,0.11385542721485654,0.11385542721485654,0.0004431903698023628,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.03359999880194664,0.01814719969406724,0.015375999733805656,0.005678506740662529,0.06774400174617767,0.07891199737787247,0.07312640026211739,0.07312640026211739,0.003826583051422731,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01788800023496151,0.028543999418616295,0.021340799890458582,0.019504000432789326,0.003718627037219618,0.09548799693584442,0.1327359974384308,0.10823359936475753,0.10823359936475753,0.013230635292043864,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.02627200074493885,0.030368000268936157,0.027184000052511693,0.02643200010061264,0.0013545679205210022,0.14364799857139587,0.1597760021686554,0.15008639842271806,0.15008639842271806,0.005367723415171222,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04150399938225746,0.04726399853825569,0.042950399965047834,0.04224000126123428,0.001720147501765378,0.2433920055627823,0.28963199257850647,0.2549152016639709,0.2549152016639709,0.013450991019687407,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.025887999683618546,0.03407999873161316,0.028303999826312064,0.026367999613285065,0.0028877495368841042,0.325439989566803,0.3441599905490875,0.33442879617214205,0.33442879617214205,0.005693597811441922,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04137599840760231,0.05084799975156784,0.044828799366950986,0.043087998405098915,0.003560363865797613,0.6110399961471558,0.6421759724617004,0.6204223990440368,0.6204223990440368,0.011214490370794758,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014816000126302242,0.02364799939095974,0.016668799985200166,0.01592000015079975,0.0025515238179941247,0.0,0.0,0.0,0.0,0.0,0.0544000007212162,0.09014400094747543,0.06511679962277411,0.06511679962277411,0.013005726711295521,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.0208320003002882,0.016604799963533878,0.01566399959847331,0.0021353594266203244,0.0,0.0,0.0,0.0,0.0,0.17871999740600586,0.1961279958486557,0.18568639904260634,0.18568639904260634,0.004566142885274747,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.02006400004029274,0.015619200002402068,0.01508800033479929,0.001572984783715635,0.0,0.0,0.0,0.0,0.0,0.2730880081653595,0.29721599817276,0.2815328001976013,0.2815328001976013,0.007016534727616923,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.02755199931561947,0.018678399827331306,0.015919999685138464,0.004323555372382858,0.0,0.0,0.0,0.0,0.0,0.3893119990825653,0.44041600823402405,0.40225600004196166,0.40225600004196166,0.013442998165884852,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.02659199945628643,0.016748800035566093,0.015343999955803156,0.0036412341868394082,0.0,0.0,0.0,0.0,0.0,0.7404800057411194,0.9689919948577881,0.8451807916164398,0.8451807916164398,0.08637455920036795,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.02160000056028366,0.016752000153064727,0.01521599991247058,0.0025167650350367246,0.0,0.0,0.0,0.0,0.0,0.06412799656391144,0.08956799656152725,0.07242240011692047,0.07242240011692047,0.009797120588901621,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.022911999374628067,0.016137600038200618,0.015343999955803156,0.002330911555470513,0.0,0.0,0.0,0.0,0.0,0.3128319978713989,0.3282879889011383,0.3203647971153259,0.3203647971153259,0.004598568238907996,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.03574400022625923,0.01919359974563122,0.01583999954164028,0.006642521913489215,0.0,0.0,0.0,0.0,0.0,0.5050879716873169,0.5311999917030334,0.511932796239853,0.511932796239853,0.007150929647317677,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.02595200017094612,0.016992000117897987,0.015456000342965126,0.003390731937723684,0.0,0.0,0.0,0.0,0.0,0.7385600209236145,0.7681919932365417,0.7483008027076722,0.7483008027076722,0.010214373356208012,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.038975998759269714,0.01952639976516366,0.015327999833971262,0.007695927285621061,0.0,0.0,0.0,0.0,0.0,1.4228800535202026,1.5237760543823242,1.4435008168220522,1.4435008168220522,0.0342955291725742,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.03907199949026108,0.01793599994853139,0.015248000156134367,0.007167103363234667,0.0,0.0,0.0,0.0,0.0,0.06947200000286102,0.08857599645853043,0.07469440028071403,0.07469440028071403,0.00583249952929747,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.02175999991595745,0.016710399929434062,0.015327999833971262,0.002705383977079183,0.0,0.0,0.0,0.0,0.0,0.388480007648468,0.4073280096054077,0.3966591984033584,0.3966591984033584,0.004487315516095964,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.022048000246286392,0.017011200170964004,0.015392000321298838,0.002776899649845722,0.0,0.0,0.0,0.0,0.0,0.622048020362854,0.6347839832305908,0.6262047946453094,0.6262047946453094,0.004593360122958751,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.02937600016593933,0.01847040019929409,0.015344000421464443,0.005043809142122341,0.0,0.0,0.0,0.0,0.0,0.9105280041694641,0.9304640293121338,0.914108806848526,0.914108806848526,0.00563919268816644,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.039903998374938965,0.01791359977796674,0.015440000221133232,0.007367547649297217,0.0,0.0,0.0,0.0,0.0,1.764799952507019,1.7965760231018066,1.770739197731018,1.770739197731018,0.009082122770320404,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.03187200054526329,0.018553599901497363,0.016127999871969223,0.0050628774268006264,0.16963527081512533,0.1737871320906266,0.17088926838108623,0.17088926838108623,0.0011703114258121128,0.47362872483230506,0.48522089403242025,0.47712993814280913,0.47712993814280913,0.0032675581298664976,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,520,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,8,8,9,8,16384.0,0.1111111111111111,q512_8q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.03868800029158592,0.01921919994056225,0.016959999687969685,0.006691309533030663,0.1589600576212269,0.16117782913137854,0.15969057520605917,0.15969057520605917,0.0006212984448748182,0.5199519263456005,0.5272061673552852,0.5223414198520004,0.5223414198520004,0.002032242112153396,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1040,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,16,16,17,16,16384.0,0.058823529411764705,q1k_16q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018400000408291817,0.024320000782608986,0.019910399988293647,0.0191040001809597,0.001841532763984352,0.25650752966102763,0.27066608538463055,0.259736894547936,0.259736894547936,0.00462927315947332,0.5278764825612624,0.5570139063374764,0.5345223138928448,0.5345223138928448,0.009526755161797178,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,32,32,33,32,16384.0,0.030303030303030304,q2k_32q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.026048000901937485,0.03728000074625015,0.028636799938976765,0.02705600019544363,0.003313953649011259,0.9270006318443208,0.9712965120641314,0.93480767601568,0.93480767601568,0.012612551450284608,0.8181833128578277,0.8572794566782395,0.8250739157811953,0.8250739157811953,0.01113200873299586,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,16,16,17,16,32768.0,0.058823529411764705,q4k_16q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018079999834299088,0.037696000188589096,0.022092800214886667,0.019024000503122807,0.005986278786633071,0.2839857165542317,0.2876110048757805,0.2849506939696605,0.2849506939696605,0.0009995178425916847,1.0871823008331585,1.101060989333509,1.0908765231323903,1.0908765231323903,0.0038264533900426207,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,32,32,33,32,32768.0,0.030303030303030304,q2k_32q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.025567999109625816,0.01726400014013052,0.015728000551462173,0.0031324465940990903,0.137270464802061,0.13799613818579368,0.13751499486424038,0.13751499486424038,0.00025221761530472904,2.3021855212208884,2.314355908780759,2.3062865750744197,2.3062865750744197,0.004229983070201486,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,64,64,65,64,32768.0,0.015384615384615385,q1k_64q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.022784000262618065,0.016435200069099664,0.015519999898970127,0.0023688301421469523,0.04879999905824661,0.09139200299978256,0.06054079942405224,0.06054079942405224,0.012152608702448775,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014816000126302242,0.02844800055027008,0.017008000146597625,0.015696000307798386,0.003941466294662679,0.047807998955249786,0.07356800138950348,0.05626560002565384,0.05626560002565384,0.00842179125412236,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.053408000618219376,0.019353600218892097,0.01532800029963255,0.01138302400841626,0.048448000103235245,0.0785600021481514,0.0556256003677845,0.0556256003677845,0.009348274502616908,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.03139200061559677,0.01823679991066456,0.016159999649971724,0.004860071238302512,0.055424001067876816,0.08505599945783615,0.0640383992344141,0.0640383992344141,0.00921448636178623,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.026528000831604004,0.017324799951165915,0.01536000007763505,0.0036004822686428305,0.07660800218582153,0.0942080020904541,0.08209280073642732,0.08209280073642732,0.00515895587669752,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.021247999742627144,0.016403199825435876,0.01532800029963255,0.001995897022647934,0.11395200341939926,0.15113599598407745,0.12431039959192276,0.12431039959192276,0.011164431123683732,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017184000462293625,0.028672000393271446,0.019865600019693376,0.017823999747633934,0.0035798491315929977,0.3171840012073517,0.3341119885444641,0.3261695951223373,0.3261695951223373,0.005111046452878918,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024639999493956566,0.030400000512599945,0.02656640000641346,0.02556800004094839,0.0020189846603237303,1.0648640394210815,1.0828479528427124,1.0715327858924866,1.0715327858924866,0.005558639263049851,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014976000413298607,0.021695999428629875,0.017305599898099898,0.01593599934130907,0.0025718046517268054,0.04956800118088722,0.07932800054550171,0.06228480041027069,0.06228480041027069,0.01027160349757827,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01539199985563755,0.03580800071358681,0.019686400331556796,0.017136000096797943,0.005918387811304003,0.05158400163054466,0.10713600367307663,0.06364160068333148,0.06364160068333148,0.015832957809696766,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02223999984562397,0.016672000009566545,0.015887999907135963,0.0020934945946034563,0.06521599739789963,0.08902399986982346,0.07520959973335266,0.07520959973335266,0.007913840684102929,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015231999568641186,0.04028800129890442,0.019971200078725816,0.01646399963647127,0.007351305886577286,0.14467200636863708,0.16844800114631653,0.15363519936800005,0.15363519936800005,0.008174435899956223,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015519999898970127,0.02304000034928322,0.01775679988786578,0.016784000210464,0.0025077243712082584,0.33740800619125366,0.35343998670578003,0.3445120006799698,0.3445120006799698,0.00445648463590358,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.02393599972128868,0.01809599995613098,0.01654400024563074,0.002998393102466254,0.0,0.0,0.0,0.0,0.0,0.0504320003092289,0.0843840017914772,0.060083200410008426,0.060083200410008426,0.00986959318572296,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.030047999694943428,0.019510399922728537,0.017680000513792038,0.004065185265963332,0.0,0.0,0.0,0.0,0.0,0.05004800111055374,0.07036799937486649,0.059315200522542,0.059315200522542,0.006537768821329647,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.02393599972128868,0.018035199772566558,0.016512000001966953,0.0030667689116777724,0.0,0.0,0.0,0.0,0.0,0.05100800096988678,0.06735999882221222,0.058387200161814694,0.058387200161814694,0.005832787739241381,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.026208000257611275,0.017500799987465142,0.01646399963647127,0.0030666103306165714,0.0,0.0,0.0,0.0,0.0,0.05023999884724617,0.07254400104284286,0.057254400476813315,0.057254400476813315,0.006890853653068151,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.027327999472618103,0.0179776000790298,0.01648000068962574,0.003783417447531392,0.0,0.0,0.0,0.0,0.0,0.04819199815392494,0.07100799679756165,0.05621119923889638,0.05621119923889638,0.007824391484124725,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.036448001861572266,0.019136000238358975,0.016704000532627106,0.006076530095624803,0.0,0.0,0.0,0.0,0.0,0.047488000243902206,0.08441600203514099,0.0588383998721838,0.0588383998721838,0.011275940726332522,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015135999768972397,0.033952001482248306,0.02119360016658902,0.018000000156462193,0.007136646614305304,0.0,0.0,0.0,0.0,0.0,0.04931199923157692,0.06992000341415405,0.05755840018391609,0.05755840018391609,0.007297587694315226,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.034272000193595886,0.0204415999352932,0.017311999574303627,0.00695404701803013,0.0,0.0,0.0,0.0,0.0,0.05215999856591225,0.0735040009021759,0.06117440015077591,0.06117440015077591,0.007381118436894922,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03743999823927879,0.02012479966506362,0.01775999926030636,0.006181045509079057,0.0,0.0,0.0,0.0,0.0,0.06355199962854385,0.08508799970149994,0.07019200026988984,0.07019200026988984,0.0062246580442117845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.032607998698949814,0.01959999995306134,0.017487999983131886,0.004882475068460749,0.0,0.0,0.0,0.0,0.0,0.04918399825692177,0.0865280032157898,0.05973760038614274,0.05973760038614274,0.011546847942113974,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02675200067460537,0.017430400010198355,0.016560000367462635,0.003243135094239819,0.0,0.0,0.0,0.0,0.0,0.052671998739242554,0.08367999643087387,0.06238719932734965,0.06238719932734965,0.011207824780630104,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.03612799942493439,0.019327999837696553,0.01688000001013279,0.006058251425153085,0.0,0.0,0.0,0.0,0.0,0.06364800035953522,0.07878399640321732,0.06935679838061332,0.06935679838061332,0.004515588328677643,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03711999952793121,0.019670399930328132,0.017152000218629837,0.006165994646522264,0.0,0.0,0.0,0.0,0.0,0.09071999788284302,0.11507199704647064,0.10252480059862136,0.10252480059862136,0.008782544051535657,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.026528000831604004,0.018396800104528665,0.01601599995046854,0.004279788048986283,0.0,0.0,0.0,0.0,0.0,0.05331199988722801,0.07977599650621414,0.06076480001211167,0.06076480001211167,0.008761579122069606,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.02643200010061264,0.01726400004699826,0.015855999663472176,0.003210008417242029,0.0,0.0,0.0,0.0,0.0,0.062431998550891876,0.08246400207281113,0.06970879957079888,0.06970879957079888,0.007016341676068233,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.02223999984562397,0.0166015999391675,0.015887999907135963,0.00204913969129354,0.0,0.0,0.0,0.0,0.0,0.10127999633550644,0.1141119971871376,0.10621120035648347,0.10621120035648347,0.004029318311754605,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03363199904561043,0.019305599946528675,0.016671999357640743,0.0055445582307981234,0.0,0.0,0.0,0.0,0.0,0.1319040060043335,0.15839999914169312,0.14040640145540234,0.14040640145540234,0.007998041073596942,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.04447999969124794,0.018908800091594458,0.0157279996201396,0.0086814937461721,0.0,0.0,0.0,0.0,0.0,0.06428799778223038,0.08982399851083755,0.07349760085344315,0.07349760085344315,0.008605284512197258,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.028672000393271446,0.01890560006722808,0.016080000437796116,0.004797584627815021,0.0,0.0,0.0,0.0,0.0,0.11123199760913849,0.14115199446678162,0.11942399889230729,0.11942399889230729,0.008513394706791027,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.024927999824285507,0.016915200091898442,0.015216000378131866,0.003374147499442635,0.0,0.0,0.0,0.0,0.0,0.15887999534606934,0.18111999332904816,0.16934399753808976,0.16934399753808976,0.007415181260761123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.026944000273942947,0.017737600207328796,0.015199999790638685,0.00415598714375255,0.0,0.0,0.0,0.0,0.0,0.2192319929599762,0.23472000658512115,0.22809920012950893,0.22809920012950893,0.004730841376327335,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.05052800104022026,0.020320000313222408,0.015520000364631414,0.010604522177924678,0.024270629882498958,0.03340247625954076,0.028446344104128624,0.028446344104128624,0.0032330369099793834,0.023697370291069768,0.03261352722995356,0.027774456325453972,0.027774456325453972,0.0031566742680923386,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.033663999289274216,0.019987199828028678,0.018240000121295452,0.005522929139253732,0.0259194055660947,0.03719755183990719,0.029916029687899703,0.029916029687899703,0.003966666590554318,0.027104595853449518,0.03889844644729374,0.03128396954022015,0.03128396954022015,0.004148046317967881,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.03299200162291527,0.019840000104159115,0.016207999549806118,0.006546343008726814,0.02587869595769926,0.03763167265431482,0.030174939058162802,0.030174939058162802,0.0037303581782277364,0.026473304070195713,0.03849632587654989,0.03086826083864964,0.03086826083864964,0.003816069654529222,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.021023999899625778,0.016304000187665223,0.015584000386297703,0.0023236165806545476,0.031810621525966996,0.04156949936878106,0.0346749350032807,0.0346749350032807,0.003130209181143985,0.035677378270900374,0.04662250161636451,0.038889864871740294,0.038889864871740294,0.0035107033960828727,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.02175999991595745,0.01569600012153387,0.015008000191301107,0.0020882666168036863,0.038211712107062853,0.07212229256520057,0.04364509673334097,0.04364509673334097,0.009671953074025085,0.0476442859917874,0.08992570455184198,0.05441890342616107,0.05441890342616107,0.01205947791784038,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.033824000507593155,0.017628800217062236,0.014864000026136637,0.005791495403857738,0.05214261250030033,0.06266261508706669,0.05677791295527661,0.05677791295527661,0.0030298223079651514,0.08648138506878382,0.10392938682791132,0.09416928531647478,0.09416928531647478,0.005025126612204489,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.031199999153614044,0.017900799959897996,0.015343999955803156,0.004974475661316249,0.06411958891421111,0.07405276123263956,0.06793047918211377,0.06793047918211377,0.0033918658556700006,0.14058441263169497,0.16236323591492058,0.1489399211274489,0.1489399211274489,0.00743678300375353,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.03855999931693077,0.018495999928563833,0.015647999942302704,0.006918530056761627,0.09872985549401277,0.10683454583043753,0.10185316839884552,0.10185316839884552,0.0020802254037042074,0.2743261390166379,0.2968454508984596,0.2830044295482752,0.2830044295482752,0.005780016596065092,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.030400000512599945,0.01977920001372695,0.01726400014013052,0.005350883464206169,0.13918872472233365,0.16279522855335207,0.1501912864839173,0.1501912864839173,0.008992185529011513,0.11892328861766266,0.13909276049083735,0.1283239123428725,0.1283239123428725,0.007682951884956939,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01727999933063984,0.02223999984562397,0.01819519978016615,0.017680000513792038,0.0014397298270620873,0.3728044181625443,0.38295504353701676,0.3761264483787333,0.3761264483787333,0.0033660272332490977,0.07967557017401883,0.08184495665371805,0.08038555277807365,0.08038555277807365,0.0007193856241088455,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.034591998904943466,0.01941439984366298,0.01756799966096878,0.005541380584373771,0.05951999872922897,0.08268799632787704,0.06715519949793816,0.06715519949793816,0.006657802116288364,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,2,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512]",1024,512,512,512.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03283200040459633,0.019180799927562477,0.016543999314308167,0.0052187911233635975,0.07199999690055847,0.08675199747085571,0.07749439924955369,0.07749439924955369,0.004849874741156772,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,4,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512]",2048,512,512,512.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01724799908697605,0.02284800074994564,0.01928640007972717,0.018400000408291817,0.0020641390157565697,0.09676799923181534,0.12310399860143663,0.10618879944086074,0.10618879944086074,0.008982738207839057,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512]",4096,512,512,512.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024224000051617622,0.03315199911594391,0.027436799928545953,0.027328000403940678,0.002679154874546328,0.14716799557209015,0.16412800550460815,0.15470399856567385,0.15470399856567385,0.005423454433987419,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512, 512]",8192,512,512,512.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01833599992096424,0.04806400090456009,0.026252799853682517,0.022672000341117382,0.00878438440429033,0.1844799965620041,0.2072959989309311,0.19359359890222552,0.19359359890222552,0.007663539820967659,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,2,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048]",4096,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,2q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024480000138282776,0.04979199916124344,0.029081599973142146,0.025679999962449074,0.007227422044689197,0.34147199988365173,0.37968000769615173,0.3583200007677078,0.3583200007677078,0.011804422100726231,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,4,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,"[2048, 2048, 2048, 2048]",8192,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,4q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.023744000121951103,0.017753600236028434,0.016368000768125057,0.003061040281616705,0.0,0.0,0.0,0.0,0.0,0.05686400085687637,0.1090880036354065,0.06629760004580021,0.06629760004580021,0.014803727026812366,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.021088000386953354,0.01647359998896718,0.015840000472962856,0.0017964402433206,0.0,0.0,0.0,0.0,0.0,0.08540800213813782,0.09961599856615067,0.09146559983491898,0.09146559983491898,0.00447423706371901,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.04156799986958504,0.018512000143527985,0.015584000386297703,0.007849701681877904,0.0,0.0,0.0,0.0,0.0,0.17948800325393677,0.194815993309021,0.1866239994764328,0.1866239994764328,0.004156328241374654,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.029632000252604485,0.019449600111693145,0.016736000776290894,0.005126811425136928,0.0,0.0,0.0,0.0,0.0,0.27529600262641907,0.2898879945278168,0.2825664013624191,0.2825664013624191,0.004542801609674093,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.031328000128269196,0.019759999960660933,0.016367999836802483,0.006266303740589241,0.0,0.0,0.0,0.0,0.0,0.39190399646759033,0.4079039990901947,0.3987520009279252,0.3987520009279252,0.0037367727509362725,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,16384.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s16k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01539199985563755,0.040031999349594116,0.0212032001465559,0.01649599988013506,0.008068547381446682,0.0,0.0,0.0,0.0,0.0,0.06588800251483917,0.07897599786520004,0.0703904002904892,0.0703904002904892,0.004214826869769361,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.022495999932289124,0.016006399970501663,0.01515199989080429,0.002265581884342885,0.0,0.0,0.0,0.0,0.0,0.1231679990887642,0.13526399433612823,0.1288223996758461,0.1288223996758461,0.004328976434897094,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.023231999948620796,0.01705600004643202,0.016048000194132328,0.002810904312746391,0.0,0.0,0.0,0.0,0.0,0.3158079981803894,0.33129599690437317,0.32348800003528594,0.32348800003528594,0.00417599076136664,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.021888000890612602,0.01708160014823079,0.01609600055962801,0.0025289234621475062,0.0,0.0,0.0,0.0,0.0,0.5103679895401001,0.5200319886207581,0.5132320046424866,0.5132320046424866,0.0035759433157749533,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.032416000962257385,0.019353600032627583,0.016176000237464905,0.0061458798960684425,0.0,0.0,0.0,0.0,0.0,0.7400320172309875,0.7516480088233948,0.7449311971664427,0.7449311971664427,0.004573461776168607,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,32768.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s32k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.02006400004029274,0.01573119992390275,0.014992000069469213,0.001654446341262991,0.0,0.0,0.0,0.0,0.0,0.07097599655389786,0.09932799637317657,0.07828159928321837,0.07828159928321837,0.009464602895232642,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.02319999970495701,0.016883199848234654,0.015775999519973993,0.0026693906156048403,0.0,0.0,0.0,0.0,0.0,0.14176000654697418,0.1598079949617386,0.15063679963350293,0.15063679963350293,0.005741662969679433,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.03203200176358223,0.022224000189453363,0.02247999981045723,0.005996176183124981,0.0,0.0,0.0,0.0,0.0,0.39180800318717957,0.41046398878097534,0.4003200054168701,0.4003200054168701,0.005552433764307601,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.028192000463604927,0.017152000125497578,0.015536000020802021,0.003847486121602065,0.0,0.0,0.0,0.0,0.0,0.6239359974861145,0.6367359757423401,0.6285343945026398,0.6285343945026398,0.004006084286606002,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.026688000187277794,0.01718079997226596,0.015488000120967627,0.0037132536813398722,0.0,0.0,0.0,0.0,0.0,0.9141119718551636,0.9721279740333557,0.927455997467041,0.927455997467041,0.02148767779232235,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,40960.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s40k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.047520000487565994,0.028710400220006704,0.0266720000654459,0.010774688401598903,0.06094816381288764,0.06816969726785131,0.0644060660218149,0.0644060660218149,0.002203106589073183,0.087051838094461,0.09736630405680231,0.09199073574791852,0.09199073574791852,0.0031466818046499644,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,520,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,8,8,9,8,16384.0,0.1111111111111111,q512_8q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.02489599958062172,0.017427200078964235,0.01595200039446354,0.003071906489592143,0.19733789497223914,0.2016295616652644,0.19860388994013833,0.19860388994013833,0.001512389485439305,0.44861409134062713,0.45837046456077934,0.45149211526120153,0.45149211526120153,0.0034381598874302305,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1040,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,16,16,17,16,16384.0,0.058823529411764705,q1k_16q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.03046399913728237,0.020643199887126686,0.016944000497460365,0.006320148675595729,0.2157924314537054,0.22215709640166995,0.21769400765743155,0.21769400765743155,0.0019679921844645526,0.4905115822753901,0.5049789194903589,0.4948340005651485,0.4948340005651485,0.0044733865493072344,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384]",1,32,32,33,32,16384.0,0.030303030303030304,q2k_32q1s16k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.017152000218629837,0.02425600029528141,0.01865920014679432,0.018112000077962875,0.0019454553198986453,0.7430866512973927,0.7578834738720781,0.7455351005236347,0.7455351005236347,0.00416692487556653,0.7369773831645824,0.7516525540362471,0.7394057025273603,0.7394057025273603,0.004132666607961175,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,16,16,17,16,32768.0,0.058823529411764705,q4k_16q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015263999812304974,0.02470399998128414,0.018009600043296815,0.01657600048929453,0.0031544532774534346,0.25251173919752334,0.25566890792461106,0.2536729054481981,0.2536729054481981,0.0010279562245342983,1.0425282722084215,1.0555630628624373,1.0473223013846877,1.0473223013846877,0.004244053880724073,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,32,32,33,32,32768.0,0.030303030303030304,q2k_32q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.027807999402284622,0.01912960009649396,0.01643200032413006,0.004688164695934997,0.1251379565220268,0.14341821167748953,0.12742497577885914,0.12742497577885914,0.005353812167343766,1.1355340167064276,1.3014137556763312,1.1562870179154463,1.1562870179154463,0.048581869194943325,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768, 32768]",1,64,64,65,64,32768.0,0.015384615384615385,q1k_64q1s32k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention_true_mixed_fused.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention_true_mixed_fused.csv new file mode 100644 index 0000000..1bcebdc --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/attention_true_mixed_fused.csv @@ -0,0 +1,49 @@ +num_tensor_parallel_workers,batch_spec,num_prefill_seqs,num_decode_seqs,total_prefill_tokens,total_decode_tokens,decode_avg_kv_cache_size,attention_core_mean_ms,attention_core_mean_as_median_ms,kv_cache_update_median_ms,pure_prefill_reference_mean_ms,pure_decode_reference_mean_ms,projected_prefill_mean_ms,projected_decode_mean_ms,projection_sum_error_ms,representation +1,q64_8q1s512,1,8,64,8,512.0,0.06604800000786781,0.06604800000786781,0.021359999664127827,0.05810240097343922,0.0600041144660541,0.032492258074592426,0.03355574193327539,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_8q1s1k,1,8,128,8,1024.0,0.07483199909329416,0.07483199909329416,0.024255999363958836,0.05587520003318787,0.06090559959411622,0.03580428402241854,0.03902771507087562,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_16q1s1k,1,16,128,16,1024.0,0.10597119852900506,0.10597119852900506,0.018112000077962875,0.05587520003318787,0.06891520097851753,0.04744885718421094,0.05852234134479412,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_16q1s2k,1,16,256,16,2048.0,0.13697599917650224,0.13697599917650224,0.01673599984496832,0.056396800279617305,0.11392960175871848,0.04535414343408448,0.09162185574241775,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_32q1s2k,1,32,256,32,2048.0,0.2163648009300232,0.2163648009300232,0.01515199989080429,0.056396800279617305,0.13807999789714814,0.06274415549817247,0.15362064543185072,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_32q1s4k,1,32,512,32,4096.0,0.3746495962142944,0.3746495962142944,0.016191999427974224,0.07432000041007995,0.2231455981731415,0.09360396051475776,0.2810456356995366,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_64q1s4k,1,64,512,64,4096.0,0.6718560099601746,0.6718560099601746,0.024848000146448612,0.07432000041007995,0.4058080047369003,0.10399797220840973,0.5678580377517648,-1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q1k_64q1s8k,1,64,1024,64,8192.0,1.3472000002861022,1.3472000002861022,0.019600000232458115,0.1307712011039257,0.7464000046253203,0.2008444429250931,1.1463555573610091,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q2k_32q1s4k,1,32,2048,32,4096.0,0.6351647913455962,0.6351647913455962,0.028256000019609928,0.33396480381488797,0.2231455981731415,0.38075520430942184,0.25440958703617433,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q4k_16q1s4k,1,16,4096,16,4096.0,1.2515872120857237,1.2515872120857237,0.04395199939608574,1.0999775886535645,0.1389280006289482,1.1112371236754128,0.14035008841031085,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_8q1s16k,1,8,512,8,16384.0,0.6654527902603149,0.6654527902603149,0.016848000697791576,0.07432000041007995,0.28281279802322384,0.13848196486571557,0.5269708253945994,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q1k_16q1s16k,1,16,1024,16,16384.0,0.7349536061286925,0.7349536061286925,0.01976000051945448,0.1307712011039257,0.39996159672737125,0.18109068484526028,0.5538629212834322,-1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q2k_32q1s16k,1,32,2048,32,16384.0,1.5246047973632815,1.5246047973632815,0.02759999968111515,0.33396480381488797,0.7430047929286957,0.4727750380198454,1.051829759343436,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q4k_16q1s32k,1,16,4096,16,32768.0,2.285759997367859,2.285759997367859,0.04617599956691265,1.0999775886535645,0.7502080142498017,1.3589365122072,0.9268234851606594,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q2k_32q1s32k,1,32,2048,32,32768.0,2.6958079814910887,2.6958079814910887,0.02817599941045046,0.33396480381488797,1.4325888037681578,0.5096392092471257,2.186168772243963,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q1k_64q1s32k,1,64,1024,64,32768.0,4.842601585388183,4.842601585388183,0.018943999893963337,0.1307712011039257,2.8293471813201903,0.21393496609766816,4.628666619290515,-8.881784197001252e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q64_8q1s512,1,8,64,8,512.0,0.05652800053358078,0.05652800053358078,0.019600000232458115,0.05961279980838299,0.055968457407185014,0.029155180178492036,0.027372820355088746,6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_8q1s1k,1,8,128,8,1024.0,0.07362559959292413,0.07362559959292413,0.014431999996304512,0.05882879942655563,0.05778240002691746,0.03714313592014963,0.036482463672774496,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_16q1s1k,1,16,128,16,1024.0,0.0736224003136158,0.0736224003136158,0.014479999896138906,0.05882879942655563,0.06270079985260964,0.03563837490653034,0.037984025407085474,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_16q1s2k,1,16,256,16,2048.0,0.10047360062599181,0.10047360062599181,0.01425600005313754,0.05459520071744919,0.06628479920327664,0.045378692890289625,0.05509490773570219,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_32q1s2k,1,32,256,32,2048.0,0.14032640159130094,0.14032640159130094,0.01566399959847331,0.05459520071744919,0.09493440166115762,0.05123499252968345,0.0890914090616175,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_32q1s4k,1,32,512,32,4096.0,0.22105600088834762,0.22105600088834762,0.016032000072300434,0.06650560013949873,0.13783999979496003,0.07194410845270183,0.14911189243564577,-2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_64q1s4k,1,64,512,64,4096.0,0.36867519915103913,0.36867519915103913,0.015263999812304974,0.06650560013949873,0.22715839892625808,0.0834932625520734,0.2851819365989658,5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q1k_64q1s8k,1,64,1024,64,8192.0,0.6853824079036711,0.6853824079036711,0.015856000129133463,0.08607039973139763,0.40332479774951924,0.12053885718696096,0.5648435507167102,1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q2k_32q1s4k,1,32,2048,32,4096.0,0.3503839999437332,0.3503839999437332,0.018432000651955605,0.19546559900045393,0.13783999979496003,0.2054811520619412,0.14490284788179197,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q4k_16q1s4k,1,16,4096,16,4096.0,0.7185311973094941,0.7185311973094941,0.026575999334454536,0.5800191938877105,0.10921279862523078,0.6046757700946376,0.11385542721485654,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_8q1s16k,1,8,512,8,16384.0,0.6480192065238953,0.6480192065238953,0.016127999871969223,0.06650560013949873,0.18568639904260634,0.17088926838108623,0.47712993814280913,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q1k_16q1s16k,1,16,1024,16,16384.0,0.6820319950580597,0.6820319950580597,0.016959999687969685,0.08607039973139763,0.2815328001976013,0.15969057520605917,0.5223414198520004,-1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q2k_32q1s16k,1,32,2048,32,16384.0,0.7942592084407808,0.7942592084407808,0.0191040001809597,0.19546559900045393,0.40225600004196166,0.259736894547936,0.5345223138928448,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q4k_16q1s32k,1,16,4096,16,32768.0,1.759881591796875,1.759881591796875,0.02705600019544363,0.5800191938877105,0.511932796239853,0.93480767601568,0.8250739157811953,2.220446049250313e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q2k_32q1s32k,1,32,2048,32,32768.0,1.3758272171020507,1.3758272171020507,0.019024000503122807,0.19546559900045393,0.7483008027076722,0.2849506939696605,1.0908765231323903,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q1k_64q1s32k,1,64,1024,64,32768.0,2.44380156993866,2.44380156993866,0.015728000551462173,0.08607039973139763,1.4435008168220522,0.13751499486424038,2.3062865750744197,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q64_8q1s512,1,8,64,8,512.0,0.0562208004295826,0.0562208004295826,0.015520000364631414,0.06054079942405224,0.0591108573866742,0.028446344104128624,0.027774456325453972,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_8q1s1k,1,8,128,8,1024.0,0.061199999228119854,0.061199999228119854,0.018240000121295452,0.05626560002565384,0.0588383998721838,0.029916029687899703,0.03128396954022015,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_16q1s1k,1,16,128,16,1024.0,0.06104319989681244,0.06104319989681244,0.016207999549806118,0.05626560002565384,0.05755840018391609,0.030174939058162802,0.03086826083864964,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_16q1s2k,1,16,256,16,2048.0,0.07356479987502099,0.07356479987502099,0.015584000386297703,0.0556256003677845,0.06238719932734965,0.0346749350032807,0.038889864871740294,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_32q1s2k,1,32,256,32,2048.0,0.09806400015950203,0.09806400015950203,0.015008000191301107,0.0556256003677845,0.06935679838061332,0.04364509673334097,0.05441890342616107,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_32q1s4k,1,32,512,32,4096.0,0.1509471982717514,0.1509471982717514,0.014864000026136637,0.0640383992344141,0.10621120035648347,0.05677791295527661,0.09416928531647478,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_64q1s4k,1,64,512,64,4096.0,0.21687040030956264,0.21687040030956264,0.015343999955803156,0.0640383992344141,0.14040640145540234,0.06793047918211377,0.1489399211274489,2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q1k_64q1s8k,1,64,1024,64,8192.0,0.3848575979471207,0.3848575979471207,0.015647999942302704,0.08209280073642732,0.22809920012950893,0.10185316839884552,0.2830044295482752,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q2k_32q1s4k,1,32,2048,32,4096.0,0.2785151988267898,0.2785151988267898,0.01726400014013052,0.12431039959192276,0.10621120035648347,0.1501912864839173,0.1283239123428725,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q4k_16q1s4k,1,16,4096,16,4096.0,0.456512001156807,0.456512001156807,0.017680000513792038,0.3261695951223373,0.06970879957079888,0.3761264483787333,0.08038555277807365,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_8q1s16k,1,8,512,8,16384.0,0.15639680176973342,0.15639680176973342,0.0266720000654459,0.0640383992344141,0.09146559983491898,0.0644060660218149,0.09199073574791852,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q1k_16q1s16k,1,16,1024,16,16384.0,0.6500960052013398,0.6500960052013398,0.01595200039446354,0.08209280073642732,0.1866239994764328,0.19860388994013833,0.45149211526120153,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q2k_32q1s16k,1,32,2048,32,16384.0,0.71252800822258,0.71252800822258,0.016944000497460365,0.12431039959192276,0.2825664013624191,0.21769400765743155,0.4948340005651485,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q4k_16q1s32k,1,16,4096,16,32768.0,1.484940803050995,1.484940803050995,0.018112000077962875,0.3261695951223373,0.32348800003528594,0.7455351005236347,0.7394057025273603,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q2k_32q1s32k,1,32,2048,32,32768.0,1.3009952068328858,1.3009952068328858,0.01657600048929453,0.12431039959192276,0.5132320046424866,0.2536729054481981,1.0473223013846877,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q1k_64q1s32k,1,64,1024,64,32768.0,1.2837119936943056,1.2837119936943056,0.01643200032413006,0.08209280073642732,0.7449311971664427,0.12742497577885914,1.1562870179154463,-2.220446049250313e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/linear_op.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/linear_op.csv new file mode 100644 index 0000000..2803f02 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/linear_op.csv @@ -0,0 +1,37 @@ +time_stats.emb.min,time_stats.emb.max,time_stats.emb.mean,time_stats.emb.median,time_stats.emb.std,time_stats.input_layernorm.min,time_stats.input_layernorm.max,time_stats.input_layernorm.mean,time_stats.input_layernorm.median,time_stats.input_layernorm.std,time_stats.attn_pre_proj.min,time_stats.attn_pre_proj.max,time_stats.attn_pre_proj.mean,time_stats.attn_pre_proj.median,time_stats.attn_pre_proj.std,time_stats.attn_rope.min,time_stats.attn_rope.max,time_stats.attn_rope.mean,time_stats.attn_rope.median,time_stats.attn_rope.std,time_stats.attn_post_proj.min,time_stats.attn_post_proj.max,time_stats.attn_post_proj.mean,time_stats.attn_post_proj.median,time_stats.attn_post_proj.std,time_stats.post_attention_layernorm.min,time_stats.post_attention_layernorm.max,time_stats.post_attention_layernorm.mean,time_stats.post_attention_layernorm.median,time_stats.post_attention_layernorm.std,n_head,n_kv_head,n_embd,n_expanded_embd,vocab_size,use_gated_mlp,use_qk_norm,attn_output_gate,num_tokens,num_tensor_parallel_workers,padded_n_embd,padded_n_expanded_embd,model_arch,is_step2_mini,share_expert_dim,share_q_dim,measurement_type,profiling_precision,quant_signature +0.029184000566601753,0.06780800223350525,0.03157280012965202,0.030736000277101994,0.005874173435341216,0.033215999603271484,0.04825599864125252,0.03443359974771738,0.0337119996547699,0.0031825678429048183,1.438431978225708,1.505568027496338,1.446228802204132,1.4429279565811157,0.014128607642643025,0.538752019405365,0.5440319776535034,0.5416463971138,0.5420799851417542,0.0016099306164432847,1.0073280334472656,1.0163840055465698,1.0098415970802308,1.0081279873847961,0.0032980457197329728,0.04022400081157684,0.041728001087903976,0.04091359991580248,0.04081599973142147,0.0004728505416767937,32,4,2048,768,151936,True,True,False,8192,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.01360000018030405,0.06784000247716904,0.017755200061947106,0.0179840000346303,0.00837681421401443,0.01836800016462803,0.03651199862360954,0.019606399815529585,0.018719999119639397,0.0038821958848767424,0.7512000203132629,0.8208960294723511,0.7566704005002975,0.75382399559021,0.014793092586577971,0.28995200991630554,0.29337599873542786,0.29135999977588656,0.29150401055812836,0.000998381071258815,0.5149760246276855,0.5169600248336792,0.5160208016633987,0.5158880054950714,0.0005038652783291977,0.02051199972629547,0.021503999829292297,0.021067200042307378,0.021104000508785248,0.00023542855306902367,32,4,2048,768,151936,True,True,False,4096,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.0080960001796484,0.04281599819660187,0.011092799971811474,0.011039999779313803,0.005489135752949302,0.012223999947309494,0.026335999369621277,0.013187199970707298,0.01247999956831336,0.0030236510562153375,0.3928639888763428,0.45372799038887024,0.3976895987987518,0.39528000354766846,0.012905176716136006,0.1547199934720993,0.15884800255298615,0.15712319910526276,0.1573439985513687,0.001165121945135037,0.26633599400520325,0.268095999956131,0.26719200164079665,0.2671840041875839,0.0004242740199415328,0.013024000450968742,0.013887999579310417,0.013489600038155913,0.013520000036805868,0.0002382817554057738,32,4,2048,768,151936,True,True,False,2048,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.00825599953532219,0.04755200073122978,0.02398160002194345,0.01961600035429001,0.00874683840888523,0.018880000337958336,0.03868800029158592,0.021590400114655496,0.0208320003002882,0.004057015751746236,0.24316799640655518,0.2710399925708771,0.2521967992186546,0.25065599381923676,0.007998418528894075,0.09644799679517746,0.19120000302791595,0.10407840013504029,0.09963199868798256,0.020043722414992166,0.13600000739097595,0.18892799317836761,0.15760480016469955,0.15760000050067902,0.0112223677907762,0.009664000011980534,0.010015999898314476,0.009836799977347255,0.009824000298976898,9.016971952948177e-05,32,4,2048,768,151936,True,True,False,1024,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017376000061631203,0.044704001396894455,0.026318399980664254,0.027312000282108784,0.007565344034680866,0.01836800016462803,0.03014400042593479,0.021276800055056812,0.020655999891459942,0.002632977855097951,0.1438719928264618,0.17132799327373505,0.152497598528862,0.15012799948453903,0.007947150319625347,0.1430719941854477,0.19305600225925446,0.16630879789590836,0.1685439944267273,0.014628024163894684,0.08899199962615967,0.10467199981212616,0.0943599995225668,0.09374399855732918,0.003472669494074113,0.007615999784320593,0.007935999892652035,0.007769599952735007,0.0077760000713169575,7.680004540222077e-05,32,4,2048,768,151936,True,True,False,512,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016992000862956047,0.0544000007212162,0.02573199989274144,0.026016000658273697,0.00803323401720434,0.018432000651955605,0.02348800003528595,0.020648000109940768,0.02062400057911873,0.0013939985047930988,0.10220800340175629,0.1361600011587143,0.11850560046732425,0.11684799939393997,0.010637754898360304,0.16710400581359863,0.21110400557518005,0.19078560024499894,0.19409599900245667,0.013714352646558832,0.06265600025653839,0.0740479975938797,0.06842879951000214,0.0690080001950264,0.0031292240446560557,0.00979200005531311,0.033440001308918,0.01864320016466081,0.017280000261962414,0.006957998188876383,32,4,2048,768,151936,True,True,False,256,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017152000218629837,0.04396799951791763,0.025907999789342284,0.02598400041460991,0.007714554737915267,0.018400000408291817,0.03667199984192848,0.024132800102233887,0.02112000063061714,0.006120348733354326,0.10678400099277496,0.1363839954137802,0.1193264003843069,0.11583999916911125,0.009838647443214228,0.17017599940299988,0.22748799622058868,0.18853759989142418,0.18433599919080734,0.015728078443174653,0.04569600149989128,0.06652799993753433,0.05192639995366335,0.05151999928057194,0.004516605694976449,0.021856000646948814,0.026335999369621277,0.02384479995816946,0.023599999956786633,0.0011301153013314744,32,4,2048,768,151936,True,True,False,128,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017343999817967415,1.0683200359344482,0.05748240072280168,0.029504000209271908,0.16366824814254827,0.018688000738620758,0.3317759931087494,0.03685439983382821,0.021151999942958355,0.06767422466731164,0.10255999863147736,0.9434880018234253,0.16412640027701855,0.11956800147891045,0.17964606281728834,0.1714559942483902,2.1306240558624268,0.3011296011507511,0.1926399990916252,0.42310127734378766,0.03574400022625923,0.6859520077705383,0.08389280084520578,0.04279999993741512,0.14321647071615612,0.020479999482631683,0.1831360012292862,0.033024000097066165,0.023856000043451786,0.03470987082429836,32,4,2048,768,151936,True,True,False,64,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016095999628305435,0.05142400041222572,0.026363200135529043,0.02676799986511469,0.008637725852473854,0.01849599927663803,0.03577600046992302,0.022193600237369538,0.020848000422120094,0.004615266181181815,0.10540799796581268,0.15014399588108063,0.12211520001292228,0.11896000057458878,0.012772013396624768,0.17315199971199036,0.21478399634361267,0.1881632000207901,0.1873439997434616,0.011761164657572015,0.03481600061058998,0.058079998940229416,0.042200000025331974,0.03969600051641464,0.006461281521769989,0.02143999934196472,0.038816001266241074,0.02466559996828437,0.023856000043451786,0.0035418033748569927,32,4,2048,768,151936,True,True,False,32,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015904000028967857,0.03753599897027016,0.023127999808639287,0.02527999971061945,0.0054230027890305385,0.018400000408291817,0.03001599945127964,0.020488000102341176,0.020096000283956528,0.002585688394137697,0.10355199873447418,0.1438400000333786,0.11536479964852334,0.11124800145626068,0.011136028294919255,0.16502399742603302,0.2072959989309311,0.18646399974822997,0.19075199961662292,0.013296437761247597,0.03142400085926056,0.05215999856591225,0.03888959977775812,0.03750399872660637,0.0057399302067536314,0.020128000527620316,0.04064000025391579,0.023937600292265417,0.023648000322282314,0.004144197400898248,32,4,2048,768,151936,True,True,False,16,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015200000256299973,0.037151999771595,0.02144480012357235,0.02195199951529503,0.005576364091933697,0.017952000722289085,0.0226879995316267,0.019934400077909233,0.019952000118792057,0.0011812499470458758,0.10063999891281128,0.13468800485134125,0.11595199964940547,0.1207519993185997,0.011974301621092394,0.16332800686359406,0.20748800039291382,0.18162400051951408,0.1767839938402176,0.01474120743720334,0.03222399950027466,0.043455999344587326,0.03829439990222454,0.03859200142323971,0.0031378131240041122,0.020959999412298203,0.03587200120091438,0.023795200139284135,0.023423999547958374,0.0031339489695198443,32,4,2048,768,151936,True,True,False,8,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.014944000169634819,0.04022400081157684,0.02162720002233982,0.022463999688625336,0.006017219317026815,0.018464000895619392,0.026528000831604004,0.021264000236988066,0.020896000787615776,0.001934095017586082,0.10156799852848053,0.1703999936580658,0.12565439902245998,0.1244799979031086,0.01641776722785409,0.1653759926557541,0.23865599930286407,0.1969360001385212,0.19223999977111816,0.02058903050274278,0.03254399821162224,0.06752000004053116,0.0443536002188921,0.041519999504089355,0.00984829071098989,0.020096000283956528,0.040031999349594116,0.026934400014579297,0.02478400059044361,0.00562071702648593,32,4,2048,768,151936,True,True,False,1,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.7531200051307678,0.8461440205574036,0.7618160009384155,0.7576479911804199,0.019464233617982506,0.2922559976577759,0.2985599935054779,0.2953856036067009,0.29576000571250916,0.001575901077689139,0.510047972202301,0.5140479803085327,0.5121696025133133,0.5123839974403381,0.0012263137313476844,,,,,,32,4,2048,768,151936,True,True,False,8192,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.39529600739479065,0.47279998660087585,0.40216960161924364,0.39825600385665894,0.0162749796240819,0.15625600516796112,0.16211199760437012,0.15959519892930984,0.15988799929618835,0.0011493418942396922,0.2635200023651123,0.2642880082130432,0.2639120012521744,0.26392000913619995,0.0001903593401384135,,,,,,32,4,2048,768,151936,True,True,False,4096,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2531839907169342,0.32419198751449585,0.26924319565296173,0.26049599051475525,0.01657194262368116,0.10051199793815613,0.2375359982252121,0.12411200068891048,0.10311999917030334,0.039466165428540506,0.15881599485874176,0.19289599359035492,0.16896959990262986,0.1640480011701584,0.009162416086418127,,,,,,32,4,2048,768,151936,True,True,False,2048,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15302400290966034,0.1780800074338913,0.16284480094909667,0.16113600134849548,0.00740355661356144,0.14521600306034088,0.20233599841594696,0.17807039842009545,0.1796799972653389,0.013908448560094403,0.0907519981265068,0.09750399738550186,0.09460479989647866,0.09478399902582169,0.0017965487667361475,,,,,,32,4,2048,768,151936,True,True,False,1024,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11740799993276596,0.17958399653434753,0.13262080028653145,0.12878400087356567,0.014146060532423056,0.17468799650669098,0.21161599457263947,0.1910431995987892,0.18966399878263474,0.01270903647700971,0.05926400050520897,0.07577600330114365,0.06530559975653887,0.06404799968004227,0.00479458462514625,,,,,,32,4,2048,768,151936,True,True,False,512,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11260800063610077,0.17209599912166595,0.13610880002379416,0.13809599727392197,0.017675922458993677,0.18729600310325623,0.26822400093078613,0.20815680101513861,0.2078079953789711,0.01700718593658771,0.04499199986457825,0.06537599861621857,0.0507551996037364,0.04787199944257736,0.005727123232692158,,,,,,32,4,2048,768,151936,True,True,False,256,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11257600039243698,0.23452800512313843,0.13866880126297473,0.13964799791574478,0.026590047697062115,0.1828799992799759,0.24751999974250793,0.20735519900918006,0.20670399814844131,0.01500438131586336,0.03654399886727333,0.05593600124120712,0.0408239996060729,0.03892800025641918,0.004864409450710354,,,,,,32,4,2048,768,151936,True,True,False,128,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11734399944543839,0.18726399540901184,0.13890240006148816,0.13308800011873245,0.019086167340006257,0.16991999745368958,0.2443840056657791,0.19185120090842248,0.19257599860429764,0.018017536159219673,0.03033600002527237,0.04956800118088722,0.038387199863791466,0.03750400058925152,0.00500897018702887,,,,,,32,4,2048,768,151936,True,True,False,64,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11420799791812897,0.18115200102329254,0.13938880078494548,0.1393439993262291,0.018141313962922536,0.1693439930677414,0.221343994140625,0.19187839925289155,0.19438399374485016,0.01430752121272052,0.03017600066959858,0.06019200012087822,0.03866560012102127,0.0364960003644228,0.0074199790031205266,,,,,,32,4,2048,768,151936,True,True,False,32,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11276800185441971,0.16223999857902527,0.1331360016018152,0.13305599987506866,0.01432493740801146,0.17187200486660004,0.23625600337982178,0.19287680014967917,0.1913280040025711,0.01754628831589704,0.029823999851942062,0.07574400305747986,0.03900959976017475,0.036927999928593636,0.009277465083962879,,,,,,32,4,2048,768,151936,True,True,False,16,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1130559965968132,0.19814400374889374,0.13379519879817964,0.12531199678778648,0.020074427302248836,0.16841599345207214,0.21139200031757355,0.1893615983426571,0.19092799723148346,0.012418720676762711,0.029503999277949333,0.07558400183916092,0.040144000016152856,0.03728000074625015,0.010223239196498205,,,,,,32,4,2048,768,151936,True,True,False,8,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.115167997777462,0.2640640139579773,0.13805920109152794,0.13118399679660797,0.03132849683515479,0.17103999853134155,0.2977280020713806,0.19899839907884598,0.1976960003376007,0.028928046763067948,0.0297279991209507,0.05990400165319443,0.03834720011800528,0.0363520011305809,0.0070885330713495905,,,,,,32,4,2048,768,151936,True,True,False,1,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.3940800130367279,0.46483200788497925,0.39923040121793746,0.3957759886980057,0.015108903906233569,0.16022400557994843,0.1634880006313324,0.16203359961509706,0.16228799521923065,0.0010220720912414007,0.26073598861694336,0.2627840042114258,0.26138080209493636,0.2613760083913803,0.00041640363494172775,,,,,,32,4,2048,768,151936,True,True,False,8192,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2536959946155548,0.27161601185798645,0.25960480123758317,0.2577280104160309,0.005142017404245015,0.09849599748849869,0.10281600058078766,0.10057279989123344,0.10063999891281128,0.0008836232237268523,0.13913600146770477,0.17606399953365326,0.15875840038061143,0.15988799929618835,0.010058952112389172,,,,,,32,4,2048,768,151936,True,True,False,4096,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15142400562763214,0.1780479997396469,0.16139679849147798,0.1602879986166954,0.0070656055731385115,0.14601600170135498,0.23343999683856964,0.17913119941949845,0.179967999458313,0.023065274309176566,0.09388799965381622,0.12108799815177917,0.10317599996924401,0.10073599964380264,0.007364345618470178,,,,,,32,4,2048,768,151936,True,True,False,2048,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12198399752378464,0.18406400084495544,0.13770400024950505,0.13232000172138214,0.01623164885687898,0.1773120015859604,0.20688000321388245,0.1870912007987499,0.18535999953746796,0.00825628058448234,0.05766399949789047,0.06739199906587601,0.06187200043350458,0.061216000467538834,0.003004312467037863,,,,,,32,4,2048,768,151936,True,True,False,1024,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11737599968910217,0.18614399433135986,0.13109439946711063,0.12771200388669968,0.01372168725934724,0.17587199807167053,0.2699519991874695,0.19926720038056372,0.19075199961662292,0.02327478982490538,0.041728001087903976,0.06224000081419945,0.05000480003654957,0.049375999718904495,0.006185850944273961,,,,,,32,4,2048,768,151936,True,True,False,512,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1159679964184761,0.1610880047082901,0.12988320142030715,0.12494400516152382,0.011865375783184065,0.1767680048942566,0.27379199862480164,0.21035519987344742,0.21275199949741364,0.023206964017550125,0.037087999284267426,0.062144000083208084,0.04439679980278015,0.04283200018107891,0.006155226392511492,,,,,,32,4,2048,768,151936,True,True,False,256,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12015999853610992,0.158720001578331,0.13573280088603495,0.1343199983239174,0.010666830836014414,0.1737920045852661,0.22127999365329742,0.20261440128087999,0.20321600139141083,0.011153965849067301,0.03587200120091438,0.04944000020623207,0.038265600241720675,0.037328001111745834,0.0030431580113575636,,,,,,32,4,2048,768,151936,True,True,False,128,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1212799996137619,0.15139199793338776,0.13183839991688728,0.13014400005340576,0.00851207547022104,0.17257599532604218,0.2250880002975464,0.1872655987739563,0.18193599581718445,0.013638284975248818,0.02969600073993206,0.0525440014898777,0.03840640028938651,0.03444799967110157,0.007598511754254174,,,,,,32,4,2048,768,151936,True,True,False,64,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11270400136709213,0.1624639928340912,0.13508000001311302,0.13180799782276154,0.014243564451606775,0.17052799463272095,0.23715199530124664,0.1953311987221241,0.1966560035943985,0.01500670718978398,0.030239999294281006,0.05270399898290634,0.03807039987295866,0.03742399998009205,0.004906165602021684,,,,,,32,4,2048,768,151936,True,True,False,32,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12319999933242798,0.17919999361038208,0.14059039913117885,0.14156799763441086,0.016217662982107223,0.1701119989156723,0.21987199783325195,0.18903039917349815,0.1870879977941513,0.014117854507841239,0.030368000268936157,0.06406400352716446,0.04045119984075427,0.03710399940609932,0.00878942205983628,,,,,,32,4,2048,768,151936,True,True,False,16,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11296000331640244,0.16761599481105804,0.13843199908733367,0.13814399391412735,0.014925286935582404,0.1711679995059967,0.21561600267887115,0.1906527981162071,0.1876479983329773,0.011803992622137974,0.0306560005992651,0.09216000139713287,0.041129599791020155,0.036847999319434166,0.013739721468154687,,,,,,32,4,2048,768,151936,True,True,False,8,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11929599940776825,0.2008959949016571,0.14535359852015972,0.13971199840307236,0.024769473061576282,0.16803200542926788,0.25123199820518494,0.19573760256171227,0.19366399943828583,0.023196011081705884,0.03049599938094616,0.08089599758386612,0.04739360017701984,0.0453919991850853,0.01286389846889463,,,,,,32,4,2048,768,151936,True,True,False,1,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/manifest.json b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/manifest.json new file mode 100644 index 0000000..a4f773a --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/manifest.json @@ -0,0 +1,129 @@ +{ + "attention_long_context_coverage": { + "by_tp": { + "1": { + "decode_kv_lengths": [ + 128, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768, + 40960 + ], + "true_mixed_decode_avg_kv_lengths": [ + 512, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768 + ] + }, + "2": { + "decode_kv_lengths": [ + 128, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768, + 40960 + ], + "true_mixed_decode_avg_kv_lengths": [ + 512, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768 + ] + }, + "4": { + "decode_kv_lengths": [ + 128, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768, + 40960 + ], + "true_mixed_decode_avg_kv_lengths": [ + 512, + 1024, + 2048, + 4096, + 8192, + 16384, + 32768 + ] + } + }, + "included": true + }, + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp1-20260717-v1-20260717T014550833983Z/artifacts/artifacts/attention-composition-tp1-v1/raw/flashattn-composition-tp1.json": "ce461be92a5059e2d168b2cbf21b55d9a55ee71c3b0586a3e63cd8288938168d", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp2-20260717-v1-20260717T014551831545Z/artifacts/artifacts/attention-composition-tp2-v1/raw/flashattn-composition-tp2.json": "997d67fb3a28b488c3aa4945ae7c6d4ecf8a08726ee3cb4652d16c221ff395ad", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp4-20260717-v1-20260717T014552897206Z/artifacts/artifacts/attention-composition-tp4-v1/raw/flashattn-composition-tp4.json": "046cffd4c11d8a36513167ec25d3e1ca016324715679ea688a58d4a71dfe62b2", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-long-context-tp1-20260717-v1-20260717T022700381447Z/artifacts/artifacts/attention-long-context-tp1-v1/raw/flashattn-long-context-tp1.json": "da3b5adaa0d06f6a0ab7394d3c85d8d0524f0bd571613785f1a6183f70eb457f", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-long-context-tp2-20260717-v1-20260717T022701357659Z/artifacts/artifacts/attention-long-context-tp2-v1/raw/flashattn-long-context-tp2.json": "d503acb959f91b1713ff02d3251c781c240c4b057208033f4fd4a637af76e2cb", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-long-context-tp4-20260717-v1-20260717T022702416725Z/artifacts/artifacts/attention-long-context-tp4-v1/raw/flashattn-long-context-tp4.json": "a24acd44ab4075a3ac0aad00b91e676b325b2427dbe14282260cae6769481cf7", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "4f0c42d2c47f52880d81b8d2b2d7d52bd62d2c34e0f3a7f94c4c6cbd64d751c8", + "attention_true_mixed_fused.csv": "c84c129d466aec97b42f1255f3b7363fe7f73948dc27ed65b6ce599d6fb9af6d", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving-pure-prefill-batch-composition-long-context-decode-mixed", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_long_context": "Direct FA3 measurements at decode KV lengths 16K/32K/40K, plus true-mixed prefill/decode measurements at 16K/32K, for TP1/2/4", + "attention_pure_prefill_batch_composition": "Direct FA3 measurements for 2/4 requests at query length 2048 and 2/4/8/16 requests at query length 512 for each TP; included only when batch-composition attention inputs are supplied", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 213, + "attention_true_mixed_fused_diagnostic": 48, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v4" +} diff --git a/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/moe.csv b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/moe.csv new file mode 100644 index 0000000..311097f --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final/moe.csv @@ -0,0 +1,73 @@ +time_stats.moe_gating_linear.min,time_stats.moe_gating_linear.max,time_stats.moe_gating_linear.mean,time_stats.moe_gating_linear.median,time_stats.moe_gating_linear.std,time_stats.moe_gating_routing_topk.min,time_stats.moe_gating_routing_topk.max,time_stats.moe_gating_routing_topk.mean,time_stats.moe_gating_routing_topk.median,time_stats.moe_gating_routing_topk.std,time_stats.moe_shuffling.min,time_stats.moe_shuffling.max,time_stats.moe_shuffling.mean,time_stats.moe_shuffling.median,time_stats.moe_shuffling.std,time_stats.moe_grouped_gemm.min,time_stats.moe_grouped_gemm.max,time_stats.moe_grouped_gemm.mean,time_stats.moe_grouped_gemm.median,time_stats.moe_grouped_gemm.std,num_tokens,num_experts,num_experts_per_device,expert_parallel_size,routing_runtime_path,routing_assignment_policy,routing_weight_policy,routing_uses_router_logits,gating_runtime_context,gating_runtime_context_impl,router_topk,hidden_dim,expert_hidden_dim,use_gated,num_tensor_parallel_workers,total_routed_tokens,model_expansion_ratio,tokens_per_expert_avg,tokens_to_experts_ratio,expert_utilization,min_load_ratio,load_imbalance_cv,max_load_ratio,load_entropy,load_gini_coefficient,load_distribution,seed,moe_grouped_gemm_backend,measurement_type,profiling_precision,model_arch,quant_signature,router_median_nonadditivity_ratio,projection_policy +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.33926400542259216,0.405023992061615,0.36780479848384856,0.36507199704647064,0.01690507644474779,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,1.1943039894104004,1.286784052848816,1.228384006023407,1.2273280024528503,0.02832547242381263,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.3984375,0.0,1.346291201783626,4.0,5.59375,0.661865234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,1.679744005203247,1.766144037246704,1.7095808148384095,1.7015680074691772,0.02438921262998535,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.625,0.0,1.015504800579495,5.0,6.15516433212955,0.529052734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,2.1576640605926514,2.2921600341796875,2.2097824096679686,2.188944101333618,0.045572321842012986,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.875,0.0,0.6343057228182637,2.5,6.64370748444639,0.35369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,2.422368049621582,2.5130879878997803,2.4516672134399413,2.434159994125366,0.03278900287381846,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.984375,0.0,0.4921254921257382,2.25,6.817190042344769,0.272369384765625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,2.2217600345611572,2.289599895477295,2.2571327924728393,2.263375997543335,0.021660416089449488,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,1.0,0.125,0.3486861500690843,1.875,6.908192310183997,0.197662353515625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,2.18668794631958,2.318079948425293,2.2218016147613526,2.211087942123413,0.035380897213135316,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,1.0,0.4375,0.2525504668006971,1.875,6.953347743053017,0.1410369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,2.2291839122772217,2.3929600715637207,2.2908096313476562,2.2804640531539917,0.04479348924786221,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,1.0,0.65625,0.15765965680164504,1.5625,6.98229848728205,0.08779525756835938,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,2.0678720474243164,2.1297600269317627,2.0837119817733765,2.0779199600219727,0.017880044357986735,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,1.0,0.625,0.12169081635504074,1.3125,6.9892029662356325,0.06879425048828125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,2.916032075881958,3.0819520950317383,2.9805248022079467,2.9656319618225098,0.05482195799019572,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,1.0,0.796875,0.07935434147688751,1.1796875,6.9954297964750305,0.044734954833984375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,4.421599864959717,4.535359859466553,4.486294317245483,4.497056007385254,0.036990243787549344,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,1.0,0.8203125,0.060849326483103046,1.17578125,6.9973188375859685,0.033740997314453125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,7.302591800689697,7.402751922607422,7.354758310317993,7.3464319705963135,0.032142662400335566,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,1.0,0.890625,0.0412323087266341,1.08984375,6.998772433185578,0.02334284782409668,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.35280001163482666,0.39692801237106323,0.37662720382213594,0.37196800112724304,0.013401318050665304,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.4692479968070984,0.5523840188980103,0.5134752035140991,0.5100640058517456,0.02291433464135784,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.34652799367904663,0.4119040071964264,0.3789471983909607,0.38550400733947754,0.02073945105803335,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.31462401151657104,0.7456960082054138,0.38617280423641204,0.34545600414276123,0.12230201266253077,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.32521599531173706,0.419871985912323,0.36325119733810424,0.34968000650405884,0.03161798848223672,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.289792001247406,0.4663360118865967,0.4091839998960495,0.41655999422073364,0.0446001986506615,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.3761279881000519,0.4416320025920868,0.40686399936676027,0.40540799498558044,0.02257778769899645,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.7172480225563049,0.8663039803504944,0.7723807990550995,0.7591840028762817,0.04164772379451242,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.0195519924163818,1.2216639518737793,1.1253888130187988,1.1453600525856018,0.06548005322243594,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7490559816360474,1.9644800424575806,1.8529024004936219,1.814303994178772,0.08042565617327288,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,3.2479360103607178,3.385279893875122,3.296070408821106,3.2800960540771484,0.04525529026442046,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,6.344799995422363,6.517856121063232,6.464438438415527,6.478623867034912,0.05116674443145098,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.2648000121116638,0.325439989566803,0.28852800130844114,0.28390398621559143,0.01933778377635077,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.7347840070724487,0.862496018409729,0.7769344031810761,0.769216001033783,0.03290485328796285,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.421875,0.0,1.346291201783626,6.0,5.652114648336087,0.636962890625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.9198399782180786,0.9646080136299133,0.9412063956260681,0.9411839842796326,0.014939365085478117,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.5703125,0.0,1.118033988749895,5.0,6.008641773518898,0.580810546875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,1.2796800136566162,1.3484159708023071,1.3006976008415223,1.2929120063781738,0.020807998177176254,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.8828125,0.0,0.6959705453537527,3.0,6.60872850615583,0.38055419921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,1.3630399703979492,1.4430400133132935,1.3909215927124023,1.3892319798469543,0.022335744492366926,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.984375,0.0,0.5201036555341637,3.0,6.798826509158851,0.28302001953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,1.27948796749115,1.3904000520706177,1.3176063895225525,1.309440016746521,0.038060887827312775,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,1.0,0.25,0.3511282039725661,1.875,6.91002266305238,0.1970977783203125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,1.264415979385376,1.3145920038223267,1.2791999936103822,1.2753440141677856,0.014130605249568332,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,1.0,0.375,0.24692938483248605,1.6875,6.955481130775285,0.13909912109375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,1.3081920146942139,1.347648024559021,1.3292255997657776,1.329967975616455,0.014558863679016933,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,1.0,0.625,0.17143053326165383,1.5625,6.9786675275754035,0.09520339965820312,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.242751955986023,1.3112000226974487,1.2747935891151427,1.266207993030548,0.021093073517695057,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,1.0,0.78125,0.11000099875256815,1.296875,6.991308871213679,0.062183380126953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7388160228729248,1.8077759742736816,1.772764801979065,1.772704005241394,0.021056644077284283,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,1.0,0.78125,0.0864630150197678,1.1796875,6.994552526394139,0.048796653747558594,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,2.6563520431518555,2.7063679695129395,2.6785055875778196,2.6791679859161377,0.01639963463052944,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,1.0,0.8671875,0.06127686514721937,1.16015625,6.997291583027146,0.03497934341430664,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,4.386879920959473,4.452256202697754,4.4108480453491214,4.406303882598877,0.019768161791937636,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,1.0,0.884765625,0.041723768525324195,1.1171875,6.998746434318934,0.02298593521118164,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.24208000302314758,0.4028480052947998,0.3041536003351212,0.277103990316391,0.05660721484881584,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.2447039932012558,0.30502399802207947,0.26446720361709597,0.26265600323677063,0.016744548364435372,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.2337920069694519,0.2881599962711334,0.26074880361557007,0.264384001493454,0.016469850143940968,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.23369599878787994,0.28591999411582947,0.25465920120477675,0.25385600328445435,0.01593204652619194,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.2295999974012375,0.26556798815727234,0.24674240052700042,0.2497600018978119,0.010345732066199003,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.21721599996089935,0.29020801186561584,0.2394208014011383,0.2346400022506714,0.018747330357768585,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2717759907245636,0.305184006690979,0.28813759982585907,0.28809599578380585,0.01183422600545854,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.3917759954929352,0.43772798776626587,0.41130879521369934,0.4131519943475723,0.012992473640805227,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.6176639795303345,0.7009919881820679,0.642767995595932,0.6330719888210297,0.024074084919938756,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.0820800065994263,1.1674879789352417,1.1034304022789,1.0977439880371094,0.0233067292981566,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.9809919595718384,2.0415360927581787,2.003715181350708,1.992751955986023,0.022066076434645737,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,3.790112018585205,3.8651199340820312,3.829139161109924,3.8230879306793213,0.025177464160110564,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.212351992726326,0.24383999407291412,0.22760000079870224,0.22723200172185898,0.01050568575837594,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.47494399547576904,0.5184000134468079,0.4920704007148743,0.49169600009918213,0.011991064701471855,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.3984375,0.0,1.3919410907075054,6.0,5.570159765557392,0.667236328125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.6360960006713867,0.7004479765892029,0.6608384013175964,0.6572319865226746,0.020416877242438597,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.625,0.0,1.0307764064044151,4.0,6.138251855282827,0.5382080078125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.780896008014679,0.8301439881324768,0.80346559882164,0.8030399978160858,0.016801230312128875,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.859375,0.0,0.6903350635742038,3.0,6.5943747091218174,0.38067626953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.8607040047645569,0.9195200204849243,0.8783008038997651,0.8751039803028107,0.01719059253115595,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.9765625,0.0,0.49410588440130926,2.75,6.814452474347134,0.271270751953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.833952009677887,0.894752025604248,0.8619967997074127,0.863215982913971,0.018716378851797198,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,1.0,0.25,0.33693529145074724,2.125,6.9186075263155535,0.1867218017578125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.834879994392395,0.8871039748191833,0.8651552021503448,0.8638879954814911,0.015262430894400969,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,1.0,0.375,0.25567294018677456,1.8125,6.952441049154937,0.14349365234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.8518080115318298,0.9097599983215332,0.8810272097587586,0.8751040101051331,0.017005819633271906,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,1.0,0.625,0.1747801353218523,1.53125,6.978069554482723,0.09820938110351562,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.8470079898834229,0.9010239839553833,0.8694015920162201,0.8716959953308105,0.016138881171190216,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,1.0,0.65625,0.1158122428154187,1.3125,6.9901908183358845,0.06445503234863281,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.1698240041732788,1.2311359643936157,1.1888479948043824,1.1890720129013062,0.017978797794492758,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,1.0,0.78125,0.08347181893108634,1.1796875,6.994921772573154,0.046871185302734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.7702720165252686,1.8097599744796753,1.7919103980064393,1.7956640124320984,0.012641295676021557,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,1.0,0.78125,0.06866734477822484,1.20703125,6.996602562938728,0.03801727294921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.968672037124634,3.0278079509735107,2.9899007797241213,2.9824799299240112,0.01816414122631667,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,1.0,0.8984375,0.04399546833977376,1.126953125,6.998607314922362,0.024699926376342773,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.19574399292469025,0.2512960135936737,0.21939200013875962,0.2199999988079071,0.017532156418212565,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.20483200252056122,0.24006399512290955,0.22215040028095245,0.22433599829673767,0.00969639786132892,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.20559999346733093,0.24726399779319763,0.22126719802618028,0.22207999974489212,0.01328093478485499,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.20003199577331543,0.2301120012998581,0.21453119963407516,0.21598400175571442,0.010402239855151332,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.19551999866962433,0.22972799837589264,0.21238719969987868,0.21488000452518463,0.010706095835489097,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.19420799612998962,0.2903999984264374,0.2211231991648674,0.21476799994707108,0.025955008886196004,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2290560007095337,0.289247989654541,0.2543327987194061,0.24939200282096863,0.01663236753503115,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.30831998586654663,0.36953601241111755,0.3324000000953674,0.3288639932870865,0.018617999572156707,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.462911993265152,0.5497919917106628,0.4893856018781662,0.4816960096359253,0.023348152887178286,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,0.7662079930305481,0.8717759847640991,0.788454395532608,0.7744799852371216,0.03174746482604812,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.363935947418213,1.4143040180206299,1.3812703967094422,1.3798720240592957,0.014770450075530007,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.5507519245147705,2.680704116821289,2.579859209060669,2.566223978996277,0.03673283558365955,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize diff --git a/runs/frontier-fidelity-envelope-v1/qwen235-v020-portability-card.md b/runs/frontier-fidelity-envelope-v1/qwen235-v020-portability-card.md new file mode 100644 index 0000000..159e7de --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/qwen235-v020-portability-card.md @@ -0,0 +1,41 @@ +# EXP-SIMFID-Q235-V020-PORTABILITY: large FP8 MoE boundary case + +> Status: compatibility gate prepared. No Qwen235 vLLM 0.20/deadc4a profile or +> real/simulator latency result exists yet; historical vLLM 0.10.2 SLO data is +> explicitly excluded. + +## Scope + +This is the requested Qwen3-235B-A22B-FP8 case, kept separate from the Qwen30 +surface. Its purpose is first to establish whether the current Frontier stack +can even represent the same large-FP8-MoE runtime; it is not evidence that a +Qwen30 profile generalizes. + +| Item | Frozen proposal | +|---|---| +| model | Qwen3-235B-A22B-FP8: FP8 e4m3/dynamic-block checkpoint; BF16 compute/KV contract recorded at runtime | +| engine | community vLLM 0.20.0 (`88d34c640…`) | +| simulator | Frontier `deadc4a…`, with `frontier.__file__` recorded by each profiler | +| candidate plans | TP4/EP1 and TP8/EP8, each MNS 64/128, MBT=8192 | +| first workload | Fixed-P: ISL=2048, OSL=1, prefix off, provisional global QPS=1.6 | +| objective | mean/p90 TTFT and E2E; TPOT=N/A | + +TP1/TP2 do not fit the 239-GB checkpoint. TP8/EP1 is not assumed legal: only +the two plans above have existing historical memory/topology evidence. + +## Gates + +1. A fresh vLLM 0.20 TP4 and TP8 server must load the checkpoint and return an + exact 2048→1 response without FP8 fallback. This is a compatibility smoke, + not a latency sample. +2. New model-specific vLLM 0.20/deadc4a profiles must cover linear, router, + FP8 MoE, prefill attention, and TP collectives for TP4/TP8. Old vLLM 0.10.2 + CUDA-event CSVs are forbidden inputs. +3. The profile manifest binds model config/quantization, vLLM, exact Frontier + import path/commit, hardware, and finite shape coverage. CPU replay must + emit legal request metrics before a real surface is launched. +4. Only then run the four-plan real/simulator Fixed-P surface with three fresh + trials; a missing simulator cell is coverage failure, not a poor config. + +This ordering prevents a profile-runtime mismatch from being misreported as a +simulator tuning failure. diff --git a/runs/frontier-fidelity-envelope-v1/qwen30-fixed-pressure-surface-card.md b/runs/frontier-fidelity-envelope-v1/qwen30-fixed-pressure-surface-card.md new file mode 100644 index 0000000..365943c --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/qwen30-fixed-pressure-surface-card.md @@ -0,0 +1,49 @@ +# EXP-SIMFID-Q30-FIXED-PRESSURE-SURFACE: high-pressure Fixed-PD / Fixed-PO + +> Status: approved and launching (2026-07-19). + +## Question + +When uniform Fixed-PD is moved from the near-singleton workload to +`4096 -> 256` at `1.125` request/s/GPU, does Frontier still choose a different +latency-optimal configuration from real community vLLM 0.20? Does removing +decode (`4096 -> 1`) change that selection boundary? + +## Setup and decision rule + +- Qwen3-30B-A3B BF16 on dash0 H20; community vLLM 0.20.0. +- Fixed-PD `4096 -> 256` and Fixed-PO `4096 -> 1`; 257 uniform, + prefix-disjoint requests; prefix caching off. +- Rate is frozen at 1.125 request/s/GPU, so global rates are 1.125, 2.25, and + 4.5 request/s for TP1, TP2, and TP4. +- Surface: `TP in {1,2,4}` x `MNS in {8,16,32,64}`, MBT=8192. Real + measurements use three fresh-server trials with rotated order. +- Frontier is `deadc4a`, `piecewise`, the frozen CUDA-event profile plus the + graph-compatible KERNEL_ONLY profile and measured collectives used by the + prior Trace-PD comparison. No latency calibration is fitted to this case. +- Compare exact winners, selected-config real regret, and pairwise order for + mean/p90 TTFT, E2E, and TPOT where decode exists. SLO is not scored. + +The pressure-probe measurements selected the workload but are excluded from +the real evaluation pool. A failed or incomplete Frontier cell is a coverage +failure, not a high-latency observation. + +## Expected output and decision + +The final table has one row per case/objective with Frontier winner, real +winner, regret, and pairwise agreement. If the old Fixed-PD reversal persists, +the failure is not an artifact of singleton load. If it disappears, simulator +fidelity has a load-regime boundary that must be localized. Fixed-PO isolates +whether decode-state composition is necessary for either outcome. + +The previously reviewed matrix schematic remains the figure prototype: +`../simulator-tuning-latency-matrix-v0/latency-selection-matrix-schematic.svg`. + +## Provenance + +Remote output root: +`dash0:/home/admin/cpfs/wjh/aituner/qwen30-fixed-pressure-surface-20260719-r1`. +The campaign records source/profile/model hashes, runtime versions, GPU +inventory, per-cell commands, raw request records, and an artifact checksum +manifest. + diff --git a/runs/frontier-fidelity-envelope-v1/qwen30-latency-expansion-card.md b/runs/frontier-fidelity-envelope-v1/qwen30-latency-expansion-card.md new file mode 100644 index 0000000..8ed2b8c --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/qwen30-latency-expansion-card.md @@ -0,0 +1,49 @@ +# EXP-SIMFID-Q30-LATENCY-EXPANSION: remaining Fixed/Trace × PD/P cases + +> Status: prepared for runtime-alignment preflight (2026-07-18). The already +> completed Qwen3-30B-A3B Trace-PD surface is excluded from this card. + +## Question and fixed boundary + +Does the same graph-aligned Frontier configuration that selected the correct +winner on Trace-PD also select the real vLLM winner when either the arrival/ +prefix state or the decode phase is removed? This is a selection test, not an +SLO/capacity evaluation. + +| Case | Input / output | arrival and cache | Applicable objectives | +|---|---|---|---| +| Fixed-PD | 2048 / 128 | 129 uniform requests; `t'=t/TP`; prefix cache off | mean/p90 TTFT, TPOT, E2E | +| Trace-P | exact held-out input, arrival, session and complete block-16 prefix relation / 1 | same TP-normalized 129-row trace; prefix cache on | mean/p90 TTFT, E2E; TPOT=N/A | +| Fixed-P | 2048 / 1 | 129 uniform requests; `t'=t/TP`; prefix cache off | mean/p90 TTFT, E2E; TPOT=N/A | + +All cases use Qwen3-30B-A3B BF16, community vLLM 0.20.0, H20, `TP∈{1,2,4}` +and `MNS∈{8,16,32,64}`, MBT=8192, chunked prefill, three fresh-server trials +per cell. The fixed QPS is the Trace-PD base offered rate (`129/600` req/s per +GPU), hence the global arrival rate is multiplied by TP. It makes throughput +per GPU comparable without claiming that different TP values see identical +cluster-level load. + +## Simulator and real contracts + +- Frontier uses commit `deadc4a…`, `piecewise`, the frozen CUDA-event profile + for prefill/mixed batches and KERNEL_ONLY profile for captured pure decode. +- Trace-P reuses the already verified graph buckets/KV capacities because the + server CLI is unchanged from Trace-PD. The only workload change is `OSL→1`. +- Fixed-P/PD disables vLLM prefix caching on both sides. Before freezing their + simulator surfaces, a no-request server-start preflight records the actual + graph captures and KV-block capacities for every `(TP,MNS)`; no request + latency or winner is used as calibration. +- Each real result verifies every request's input/output usage and row-vector + digest. `OSL=1` produces JSON `null` TPOT samples and is never converted to + zero. + +## Decision rule and cost + +For every applicable objective, compare the complete 12-cell simulator and +pooled three-trial real surface: winner match, selected-config real regret, +and non-tied pair direction agreement. A simulator crash or missing request +metric is a coverage failure, not a high-latency cell. + +The no-request Fixed runtime preflight is capped at 2 H20-GPUh. Each 36-run +real surface is estimated at 13 nominal / 41 worst-case H20-GPUh, plus CPU-only +Frontier replay. Launch logs record the resolved inputs, paths, and duration. diff --git a/runs/frontier-fidelity-envelope-v1/qwen30_exact_trace_client.py b/runs/frontier-fidelity-envelope-v1/qwen30_exact_trace_client.py new file mode 100644 index 0000000..b130b73 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/qwen30_exact_trace_client.py @@ -0,0 +1,261 @@ +#!/usr/bin/env python3 +"""Replay a private exact-trace anchor without emitting prompt text.""" + +from __future__ import annotations + +import argparse +import asyncio +import hashlib +import json +import math +from pathlib import Path +from typing import Any + + +TARGET_PASS_RATE = 0.95 + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--host", default="127.0.0.1") + parser.add_argument("--port", type=int, required=True) + parser.add_argument("--requests-file", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument( + "--served-model", + default=None, + help="Optional server-side model alias used only for HTTP routing.", + ) + parser.add_argument("--tpot-slo-ms", type=float, default=150.0) + parser.add_argument("--timeout-seconds", type=float, default=1800.0) + return parser.parse_args() + + +def percentile(values: list[float], fraction: float) -> float | None: + if not values: + return None + ordered = sorted(values) + return ordered[math.ceil(fraction * len(ordered)) - 1] + + +def ttft_slo_ms(input_tokens: int) -> float: + return 1000.0 + 1000.0 * input_tokens / 8000.0 + + +def row_vector_sha256(rows: list[dict[str, Any]]) -> str: + digest = hashlib.sha256() + for row in rows: + digest.update( + json.dumps( + [ + row["source_index"], + row["arrived_at"], + row["input_length"], + row["output_length"], + row["session_id"], + row["runtime_block_ids"], + ], + separators=(",", ":"), + ).encode() + ) + digest.update(b"\n") + return digest.hexdigest() + + +async def request_one( + session: aiohttp.ClientSession, + row: dict[str, Any], + *, + scheduled_at: float, + benchmark_start: float, + tpot_slo_ms: float, + served_model: str | None, +) -> dict[str, Any]: + loop = asyncio.get_running_loop() + delay = scheduled_at - loop.time() + if delay > 0: + await asyncio.sleep(delay) + admitted_at = loop.time() + record: dict[str, Any] = { + "source_index": int(row["source_index"]), + "session_id": int(row["session_id"]), + "scheduled_s": scheduled_at - benchmark_start, + "admitted_s": admitted_at - benchmark_start, + "admission_lag_ms": (admitted_at - scheduled_at) * 1000.0, + "input_tokens": int(row["input_length"]), + "requested_output_tokens": int(row["output_length"]), + "success": False, + } + body = dict(row["body"]) + if served_model is not None: + body["model"] = served_model + body.update( + { + "temperature": 0, + "stream": True, + "stream_options": {"include_usage": True}, + "return_token_ids": True, + } + ) + try: + started = loop.time() + async with session.post("/v1/completions", json=body) as response: + if response.status != 200: + detail = (await response.text())[:1000] + raise RuntimeError(f"HTTP {response.status}: {detail}") + first_token_at = None + last_token_at = None + streamed_tokens = 0 + usage = None + while True: + raw = await response.content.readline() + if not raw: + break + line = raw.decode(errors="replace").strip() + if not line.startswith("data:"): + continue + data = line[5:].strip() + if data == "[DONE]": + break + payload = json.loads(data) + if payload.get("usage"): + usage = payload["usage"] + emitted = 0 + for choice in payload.get("choices") or []: + token_ids = choice.get("token_ids") or [] + emitted += len(token_ids) if token_ids else int(bool(choice.get("text"))) + if emitted: + now = loop.time() + first_token_at = first_token_at or now + last_token_at = now + streamed_tokens += emitted + finished = loop.time() + if first_token_at is None or last_token_at is None or usage is None: + raise RuntimeError("missing streaming token or usage") + actual_input = int(usage["prompt_tokens"]) + actual_output = int(usage["completion_tokens"]) + if actual_input != int(row["input_length"]) or actual_output != int( + row["output_length"] + ): + raise RuntimeError(f"usage mismatch: {actual_input}+{actual_output}") + ttft = (first_token_at - started) * 1000.0 + tpot = ( + (last_token_at - first_token_at) * 1000.0 / (actual_output - 1) + if actual_output > 1 + else None + ) + record.update( + { + "success": True, + "actual_input_tokens": actual_input, + "actual_output_tokens": actual_output, + "streamed_token_count": streamed_tokens, + "ttft_ms": ttft, + "tpot_ms": tpot, + "e2e_ms": (finished - started) * 1000.0, + "slo_pass": ttft <= ttft_slo_ms(actual_input) + and (tpot is None or tpot <= tpot_slo_ms), + } + ) + except Exception as error: + record.update( + { + "error": f"{type(error).__name__}: {error}", + "slo_pass": False, + } + ) + return record + + +async def replay(args: argparse.Namespace, rows: list[dict[str, Any]]) -> list[dict[str, Any]]: + import aiohttp + + timeout = aiohttp.ClientTimeout(total=args.timeout_seconds) + connector = aiohttp.TCPConnector(limit=0, ttl_dns_cache=300) + benchmark_start = asyncio.get_running_loop().time() + 2.0 + async with aiohttp.ClientSession( + base_url=f"http://{args.host}:{args.port}", + timeout=timeout, + connector=connector, + ) as session: + tasks = [ + asyncio.create_task( + request_one( + session, + row, + scheduled_at=benchmark_start + float(row["arrived_at"]), + benchmark_start=benchmark_start, + tpot_slo_ms=args.tpot_slo_ms, + served_model=args.served_model, + ) + ) + for row in rows + ] + return await asyncio.gather(*tasks) + + +def main() -> None: + args = parse_args() + if args.tpot_slo_ms <= 0 or args.timeout_seconds <= 0: + raise ValueError("SLO and timeout must be positive") + rows = [json.loads(line) for line in args.requests_file.open() if line.strip()] + if not rows: + raise ValueError("requests file is empty") + arrivals = [float(row["arrived_at"]) for row in rows] + if any(right < left for left, right in zip(arrivals, arrivals[1:])): + raise ValueError("request arrival order drift") + requests = asyncio.run(replay(args, rows)) + requests.sort(key=lambda row: int(row["source_index"])) + completed = [row for row in requests if row["success"]] + passed = sum(bool(row["slo_pass"]) for row in requests) + ttfts = [float(row["ttft_ms"]) for row in completed] + tpots = [ + float(row["tpot_ms"]) + for row in completed + if row["tpot_ms"] is not None + ] + pass_rate = passed / len(requests) + payload = { + "schema": "qwen30-exact-trace-anchor-v1", + "contract": { + "requests_file": str(args.requests_file.resolve()), + "requests_file_sha256": hashlib.sha256( + args.requests_file.read_bytes() + ).hexdigest(), + "row_vector_sha256": row_vector_sha256(rows), + "requests": len(rows), + "first_arrival_s": arrivals[0], + "last_arrival_s": arrivals[-1], + "arrival": "original_trace_timestamp_and_order", + "input_output_prompt": "exact_source_values", + "served_model_alias": args.served_model, + "ttft_slo": "1000ms + 1000ms * input_tokens / 8000", + "tpot_slo_ms": args.tpot_slo_ms, + "target_pass_rate": TARGET_PASS_RATE, + "prompt_text_emitted": False, + }, + "summary": { + "completed": len(completed), + "failed": len(requests) - len(completed), + "passed": passed, + "pass_rate": pass_rate, + "feasible": pass_rate >= TARGET_PASS_RATE, + "ttft_p50_ms": percentile(ttfts, 0.50), + "ttft_p95_ms": percentile(ttfts, 0.95), + "tpot_p50_ms": percentile(tpots, 0.50), + "tpot_p95_ms": percentile(tpots, 0.95), + "admission_lag_p95_ms": percentile( + [float(row["admission_lag_ms"]) for row in requests], 0.95 + ), + }, + "requests": requests, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps(payload["summary"], sort_keys=True), flush=True) + if len(completed) != len(requests): + raise SystemExit(2) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/remote-sync-marker/README.md b/runs/frontier-fidelity-envelope-v1/remote-sync-marker/README.md new file mode 100644 index 0000000..bf2fe4b --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/remote-sync-marker/README.md @@ -0,0 +1,2 @@ +This directory is the gpu-fleet synchronization marker. Experiment code is +synchronized to dash0 through the project Git branch before dispatch. diff --git a/runs/frontier-fidelity-envelope-v1/results/a2/capacity.csv b/runs/frontier-fidelity-envelope-v1/results/a2/capacity.csv new file mode 100644 index 0000000..da354c9 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a2/capacity.csv @@ -0,0 +1,13 @@ +config,tp,mns,real,simulator +tp1_mns8,1,8,7.0,8.0 +tp1_mns16,1,16,7.0,8.0 +tp1_mns32,1,32,7.0,8.0 +tp1_mns64,1,64,7.0,8.0 +tp2_mns8,2,8,7.0,8.0 +tp2_mns16,2,16,7.0,8.0 +tp2_mns32,2,32,7.0,8.0 +tp2_mns64,2,64,7.0,8.0 +tp4_mns8,4,8,8.0,8.0 +tp4_mns16,4,16,8.0,8.0 +tp4_mns32,4,32,8.0,8.0 +tp4_mns64,4,64,8.0,8.0 diff --git a/runs/frontier-fidelity-envelope-v1/results/a2/comparison.json b/runs/frontier-fidelity-envelope-v1/results/a2/comparison.json new file mode 100644 index 0000000..4aa1522 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a2/comparison.json @@ -0,0 +1,5638 @@ +{ + "contract": { + "input_tokens": 2048, + "model": "Qwen3-30B-A3B", + "output_tokens": 1, + "prefix_caching": false, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "metrics": { + "anchor_confusion": { + "real_fail_sim_fail": 24, + "real_fail_sim_pass": 8, + "real_pass_sim_fail": 0, + "real_pass_sim_pass": 28 + }, + "anchor_grid_coverage": { + "real_only": 36, + "shared": 60, + "simulator_only": 0 + }, + "config_order": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "kendall": { + "both_ties": 34, + "concordant": 0, + "discordant": 0, + "kendall_tau_b": null, + "real_only_ties": 0, + "simulator_only_ties": 32 + }, + "pairwise_non_tied": { + "all": { + "accuracy": 0.0, + "comparable": 32, + "correct": 0 + }, + "within_tp": {} + }, + "real_best_capacity_per_gpu": 8.0, + "real_top_set": [ + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "simulator_best_capacity_per_gpu": 8.0, + "simulator_top_set": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "top1_regret_best": 0.0, + "top1_regret_worst": 0.125, + "top_set_exact_match": false, + "top_set_overlap": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns8" + ] + }, + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "real": { + "tp1_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.15452707000077, + 152.64549700077623 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r4p00.json", + "sha256": "d82e386ec3effdf57c148171311928298405da0e4f896584bf05db2485bb6bd4", + "summary": { + "admission_lag_max_ms": 0.11588609777390957, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.6234939750284, + "ttft_p50_ms": 151.05855697765946, + "ttft_p95_ms": 153.15452707000077 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r4p00.json", + "sha256": "1565be89131e3c8a9c352c6f29694ef2793734b0623a2db1cbea7e0b0e8a2bf6", + "summary": { + "admission_lag_max_ms": 0.11451996397227049, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.18561100866646, + "ttft_p50_ms": 151.18564700242132, + "ttft_p95_ms": 152.64549700077623 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.8308589477092, + 152.43642206769437 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r5p00.json", + "sha256": "e4447d94a326c541412b40157cfe283a6350680e3285b5d10fed6b8de5c0cd93", + "summary": { + "admission_lag_max_ms": 0.17109804321080446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.35138496104628, + "ttft_p50_ms": 150.91424097772688, + "ttft_p95_ms": 152.8308589477092 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r5p00.json", + "sha256": "572c17e309c6c9f23c490979e7acb885475f142bc530ecfee88916b048259ea2", + "summary": { + "admission_lag_max_ms": 0.1321269664913416, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.11930498946458, + "ttft_p50_ms": 150.50161199178547, + "ttft_p95_ms": 152.43642206769437 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.56597595382482, + 152.16859406791627 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r6p00.json", + "sha256": "68c8b39154d46f1ceda17f47e8cbd84a8c8a9f088c5e82d8f0ac25a7180dc14c", + "summary": { + "admission_lag_max_ms": 0.11641765013337135, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.50876995641738, + "ttft_p50_ms": 150.41159105021507, + "ttft_p95_ms": 153.56597595382482 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r6p00.json", + "sha256": "a970a2cdf7034026a56af3ea4da59856fd588da35d66ef4f9e15481a892edcc9", + "summary": { + "admission_lag_max_ms": 0.11551892384886742, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 168.65782206878066, + "ttft_p50_ms": 150.03317291848361, + "ttft_p95_ms": 152.16859406791627 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 300.90478900820017, + 293.0445579113439 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r7p00.json", + "sha256": "1aa74c18adbff53b36ae8a6f07b5b0865f6b68fb3e7862a70b36f67c18f0bba0", + "summary": { + "admission_lag_max_ms": 0.10505830869078636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.15459000505507, + "ttft_p50_ms": 263.44375393819064, + "ttft_p95_ms": 300.90478900820017 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r7p00.json", + "sha256": "e14bc9dba35d453862839eeed77262a1ed9b8394404ca90c60ec8cb26a1aa73b", + "summary": { + "admission_lag_max_ms": 0.10963028762489557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 295.1545650139451, + "ttft_p50_ms": 243.97043196950108, + "ttft_p95_ms": 293.0445579113439 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1323.4397460473701, + 1315.5151740647852 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r8p00.json", + "sha256": "9aeee7453ffe5098d203e02b84f5009ed347c478b736e89297bd3695dd81d477", + "summary": { + "admission_lag_max_ms": 0.1251589274033904, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1477.699184906669, + "ttft_p50_ms": 776.2576789828017, + "ttft_p95_ms": 1323.4397460473701 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r8p00.json", + "sha256": "7630f4901cf4ec4ea955738b119a1c32aba53f9ca1edb8819556fc60ae1fda09", + "summary": { + "admission_lag_max_ms": 0.11316593736410141, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1404.566578916274, + "ttft_p50_ms": 761.7035870207474, + "ttft_p95_ms": 1315.5151740647852 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4892.253074911423, + 4893.758205929771 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r16p00.json", + "sha256": "d3a2a85cf06a4f080edc9a3809f05493ac55b0aa1d890806ffbc551c8eca14f8", + "summary": { + "admission_lag_max_ms": 0.39830803871154785, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.215195022523, + "ttft_p50_ms": 2742.7029330283403, + "ttft_p95_ms": 4892.253074911423 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r16p00.json", + "sha256": "67110284c38a9a60228f00b1d2b0de9477d6dec660f1c91eb4f39e1b79e272a5", + "summary": { + "admission_lag_max_ms": 0.11306500528007746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.818040987477, + "ttft_p50_ms": 2742.512951954268, + "ttft_p95_ms": 4893.758205929771 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6650.756464921869, + 6751.845097984187 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r32p00.json", + "sha256": "affc62b27382400497bb82e9da31ed9891b60155318fc4f250dca0704ccc5e02", + "summary": { + "admission_lag_max_ms": 0.11010200250893831, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6920.621327939443, + "ttft_p50_ms": 3650.839229929261, + "ttft_p95_ms": 6650.756464921869 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r32p00.json", + "sha256": "df0c8d703b023692814be41f47da4db4ce1b7eb0c3423a742c10884b6447583f", + "summary": { + "admission_lag_max_ms": 0.1288279891014099, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6897.979240980931, + "ttft_p50_ms": 3730.34580796957, + "ttft_p95_ms": 6751.845097984187 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7732.2855240199715, + 7744.056030060165 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r64p00.json", + "sha256": "fbfad78129c843c85905c3f94f278f468a0a019c0cdf0b62140d0fea8f0aaffc", + "summary": { + "admission_lag_max_ms": 0.11090096086263657, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7836.245508980937, + "ttft_p50_ms": 4328.587644966319, + "ttft_p95_ms": 7732.2855240199715 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r64p00.json", + "sha256": "0f005c642327708e1ceb058f71f806b4d46ea666f45695605552e8f234521608", + "summary": { + "admission_lag_max_ms": 0.10862003546208143, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7847.871358972043, + "ttft_p50_ms": 4332.9262170009315, + "ttft_p95_ms": 7744.056030060165 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp1_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.44141807872802, + 152.2580359596759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r4p00.json", + "sha256": "211fa276e4a99a061126882085b9850549b453520ac3414dd2d332c89e52b431", + "summary": { + "admission_lag_max_ms": 0.09491504170000553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 169.56262104213238, + "ttft_p50_ms": 150.8563900133595, + "ttft_p95_ms": 153.44141807872802 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r4p00.json", + "sha256": "17428902a5b20e9084f20792963f0620a1af06cdc1afb76ee99fd15dd905b89e", + "summary": { + "admission_lag_max_ms": 0.12033001985400915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.99700601119548, + "ttft_p50_ms": 150.5822929320857, + "ttft_p95_ms": 152.2580359596759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.05012692604214, + 152.00478909537196 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r5p00.json", + "sha256": "b0eb19ff7ab8a26e34e91f213a983e58689c63071285e27b3719bda0a765ff32", + "summary": { + "admission_lag_max_ms": 0.11265103239566088, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.48999004345387, + "ttft_p50_ms": 150.21017810795456, + "ttft_p95_ms": 152.05012692604214 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r5p00.json", + "sha256": "3b857dff42dd16ebee95356ceacd6b58ff7bd0a9b91f3d6208d489b84985697f", + "summary": { + "admission_lag_max_ms": 0.11149293277412653, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.33813393954188, + "ttft_p50_ms": 150.45341395307332, + "ttft_p95_ms": 152.00478909537196 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.74227899499238, + 152.82704099081457 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r6p00.json", + "sha256": "7615fa664a99f02f9f6d4f1fb0f2d6cba11842772c086634b939bc08cd0f2f4d", + "summary": { + "admission_lag_max_ms": 0.1096969936043024, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.2669020127505, + "ttft_p50_ms": 150.6725640501827, + "ttft_p95_ms": 152.74227899499238 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r6p00.json", + "sha256": "345e3cd3f8cd4d7c7a5e9343f937dd9b1b5928500133830f3046b743da6df64e", + "summary": { + "admission_lag_max_ms": 0.12192933354526758, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.12304301187396, + "ttft_p50_ms": 150.68741200957447, + "ttft_p95_ms": 152.82704099081457 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 302.5522669777274, + 294.5978350471705 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r7p00.json", + "sha256": "3f54a3e3f2279a5325a173ff714d600fb1a2cdfcfba1e8e5c980f7b21b99be59", + "summary": { + "admission_lag_max_ms": 0.11122168507426977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.24335307907313, + "ttft_p50_ms": 262.4604320153594, + "ttft_p95_ms": 302.5522669777274 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r7p00.json", + "sha256": "0832169cdf1bef49c5006464c6cbcd50f4567e0ab90bec98d722f64f7b476763", + "summary": { + "admission_lag_max_ms": 0.1059951027855277, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 297.1431679325178, + "ttft_p50_ms": 249.51713997870684, + "ttft_p95_ms": 294.5978350471705 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1321.9368209829554, + 1318.7565059633926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r8p00.json", + "sha256": "a0f5900ac9d2c47dcc7d7ac08d3f3d23e1d6f529fdd0200989afa819dd3df055", + "summary": { + "admission_lag_max_ms": 0.17414195463061333, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1475.8616799954325, + "ttft_p50_ms": 775.7647309917957, + "ttft_p95_ms": 1321.9368209829554 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r8p00.json", + "sha256": "c0fc3c85095ddacc66849e6f58be0851abab6d4022ee4eb7120c8cb8ea9e5d2c", + "summary": { + "admission_lag_max_ms": 2.749277977272868, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1407.0930189918727, + "ttft_p50_ms": 763.8335369993001, + "ttft_p95_ms": 1318.7565059633926 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4893.3289990527555, + 4893.17379205022 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r16p00.json", + "sha256": "4e9ccaf5384f622b1da7c58d3de803c86099dd9deae91e363aabccc14aeabc05", + "summary": { + "admission_lag_max_ms": 0.11893699411302805, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.85822511185, + "ttft_p50_ms": 2742.8449960425496, + "ttft_p95_ms": 4893.3289990527555 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r16p00.json", + "sha256": "c207e415186aed1148e16085145bcdfd189314e7b92f74ea72fb2e4972ef4eae", + "summary": { + "admission_lag_max_ms": 0.09541702456772327, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.2590809213, + "ttft_p50_ms": 2740.232800017111, + "ttft_p95_ms": 4893.17379205022 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6653.632597066462, + 6647.155530983582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r32p00.json", + "sha256": "780f563a2739d892761905031f82ca8bd0ac4f4eb485c454390dcb2bcdfc1b81", + "summary": { + "admission_lag_max_ms": 0.105265062302351, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6922.717701992951, + "ttft_p50_ms": 3652.452490059659, + "ttft_p95_ms": 6653.632597066462 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r32p00.json", + "sha256": "2a5eb8fb37f03e759de149d24f9b23c812a26afe72ffaaffa030688904634e43", + "summary": { + "admission_lag_max_ms": 0.10375294368714094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6917.111590038985, + "ttft_p50_ms": 3649.0042679943144, + "ttft_p95_ms": 6647.155530983582 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7747.191094094887, + 7839.257381972857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r64p00.json", + "sha256": "b94a39a91842ab39d545c473fc4f872ec79eb1888943c3e03ab03c18961a9105", + "summary": { + "admission_lag_max_ms": 0.6591080455109477, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7851.370444986969, + "ttft_p50_ms": 4335.578468977474, + "ttft_p95_ms": 7747.191094094887 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r64p00.json", + "sha256": "69ab47305c4bb14f3b4ed5bc0b18a37b8a5e7c1cdbee9232e6d2b9616ed10f84", + "summary": { + "admission_lag_max_ms": 0.10753795504570007, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.125, + "passed": 8, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7885.349316056818, + "ttft_p50_ms": 4009.1016669757664, + "ttft_p95_ms": 7839.257381972857 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp1_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 156.76817402709275, + 152.9719429090619 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r4p00.json", + "sha256": "df798dd7a1cde3cfaf698286f44bbac2b0730a2c806f125268bdf7226e2bb47c", + "summary": { + "admission_lag_max_ms": 0.1147780567407608, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 162.9874319769442, + "ttft_p50_ms": 152.67050196416676, + "ttft_p95_ms": 156.76817402709275 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r4p00.json", + "sha256": "df80057224d69ce2a79f937aa45c2093d66d131765f3c7bfcc2289a9aaff3d6a", + "summary": { + "admission_lag_max_ms": 0.2131829969584942, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.917470946908, + "ttft_p50_ms": 150.96506499685347, + "ttft_p95_ms": 152.9719429090619 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.4973110537976, + 155.23677493911237 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r5p00.json", + "sha256": "e614ffaf879859407fc6420847f0140a81bda9a94fcc438fb6782c9019bf5378", + "summary": { + "admission_lag_max_ms": 0.1283780438825488, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.08639602921903, + "ttft_p50_ms": 152.1085740532726, + "ttft_p95_ms": 153.4973110537976 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r5p00.json", + "sha256": "13e825eb5ced3d0cb2b5dd68d9243fdf1cee8779b90e9aad77be04053be22be2", + "summary": { + "admission_lag_max_ms": 0.1683390000835061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.6691420264542, + "ttft_p50_ms": 152.42570801638067, + "ttft_p95_ms": 155.23677493911237 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.53963600844145, + 152.85827894695103 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r6p00.json", + "sha256": "ac530f98f57ec452538ebe2311291d0a7c83bbd5554e82f53e78a2311802e411", + "summary": { + "admission_lag_max_ms": 0.13490929268300533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.18225098680705, + "ttft_p50_ms": 150.29372205026448, + "ttft_p95_ms": 152.53963600844145 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r6p00.json", + "sha256": "2f1445ee0890b0972252d10180ef6f00cbad45ee73bd174f21b370cdf950b654", + "summary": { + "admission_lag_max_ms": 0.16514293383806944, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.88926689513028, + "ttft_p50_ms": 150.2534201135859, + "ttft_p95_ms": 152.85827894695103 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 346.96662798523903, + 295.1050230767578 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r7p00.json", + "sha256": "c4765794d442daf7eeb64a2f587792701ebe7659c7702a3bbbfc8900801fff0e", + "summary": { + "admission_lag_max_ms": 0.16206549480557442, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 354.881492909044, + "ttft_p50_ms": 310.15722593292594, + "ttft_p95_ms": 346.96662798523903 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r7p00.json", + "sha256": "30addb463839e23a7356a5cf6d4060ec1eebfa2b9564fa9500be5bcb7aecb537", + "summary": { + "admission_lag_max_ms": 0.10446517262607813, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 296.6241450048983, + "ttft_p50_ms": 246.06937903445214, + "ttft_p95_ms": 295.1050230767578 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1376.107804942876, + 1321.1636180058122 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r8p00.json", + "sha256": "89e2975f9c7319e477789ea77015541d5235efa61d90fb8bd5b0420a58f7f186", + "summary": { + "admission_lag_max_ms": 0.1416490413248539, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.890625, + "passed": 57, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1501.330335973762, + "ttft_p50_ms": 802.2130889585242, + "ttft_p95_ms": 1376.107804942876 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r8p00.json", + "sha256": "9ca6935f1320033c3ec4b72a35c5d9909ccd7b0fff2e4b2c92c6a26f52887df1", + "summary": { + "admission_lag_max_ms": 0.12422900181263685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1410.047525074333, + "ttft_p50_ms": 766.8001820566133, + "ttft_p95_ms": 1321.1636180058122 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4950.155022088438, + 4892.750842031091 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r16p00.json", + "sha256": "4fce2749435bf2e8c77999e11be0a788cd85f459a647760f3db34f1dbc0aee85", + "summary": { + "admission_lag_max_ms": 0.10520406067371368, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.171875, + "passed": 11, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 5031.207721913233, + "ttft_p50_ms": 2775.8131299633533, + "ttft_p95_ms": 4950.155022088438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r16p00.json", + "sha256": "33ded0192d8986dbcf0a7f5ab6329a440994243d4e528d6c1b278bb6b92720fd", + "summary": { + "admission_lag_max_ms": 0.14005706179887056, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.664474112913, + "ttft_p50_ms": 2743.070787983015, + "ttft_p95_ms": 4892.750842031091 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6750.775643973611, + 6754.7571890754625 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r32p00.json", + "sha256": "44030c4d3374cecebe67b886b428ba8014c87d941c00a178d6049bc57c809f46", + "summary": { + "admission_lag_max_ms": 0.7358170114457607, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6899.517056066543, + "ttft_p50_ms": 3728.6347220651805, + "ttft_p95_ms": 6750.775643973611 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r32p00.json", + "sha256": "f397bad36552c6df5bf5bcee702cfa968ddf3e20b20e9976554ddfbe838ef3b2", + "summary": { + "admission_lag_max_ms": 0.7476879982277751, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6901.606322033331, + "ttft_p50_ms": 3729.3128239689395, + "ttft_p95_ms": 6754.7571890754625 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7800.616211956367, + 7748.703246936202 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r64p00.json", + "sha256": "61eadc2af60386d6972d69fe7b083874e705864162073d54d1bde2bc0e6de7e7", + "summary": { + "admission_lag_max_ms": 0.11387700214982033, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7905.77321103774, + "ttft_p50_ms": 4366.223473916762, + "ttft_p95_ms": 7800.616211956367 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r64p00.json", + "sha256": "db02e9a814154a9259cbe3fefdf15700923458f7e1208596e861b69b6a6ddd41", + "summary": { + "admission_lag_max_ms": 0.14900392852723598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7852.748095989227, + "ttft_p50_ms": 4337.09563605953, + "ttft_p95_ms": 7748.703246936202 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp1_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.07429495826364, + 153.62840006127954 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r4p00.json", + "sha256": "23d5ab491b70c02f51c2f06dec0407edc0627819a33b9bcef1349d974a4e20fe", + "summary": { + "admission_lag_max_ms": 0.14362391084432602, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 171.46877304185182, + "ttft_p50_ms": 151.62539994344115, + "ttft_p95_ms": 154.07429495826364 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r4p00.json", + "sha256": "563aa230e0f5e420cf453dfd9b829d7df493919192d638cbe76f85eb77b6d2e7", + "summary": { + "admission_lag_max_ms": 0.11323392391204834, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.62100906390697, + "ttft_p50_ms": 151.69239102397114, + "ttft_p95_ms": 153.62840006127954 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.20477103907615, + 153.81942794192582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r5p00.json", + "sha256": "6f2a04a5c11eed5826e2a6e1ce6e6f8d98504b31d0fec30d9440df2cdb62e56d", + "summary": { + "admission_lag_max_ms": 0.10794296395033598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.37187200970948, + "ttft_p50_ms": 152.59731002151966, + "ttft_p95_ms": 154.20477103907615 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r5p00.json", + "sha256": "27a0bfa65a66e5cc77e10a5cf39a846f0fd1dd266c10926dae9db7a80c581d6e", + "summary": { + "admission_lag_max_ms": 0.1204309519380331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.40366806089878, + "ttft_p50_ms": 152.0398510619998, + "ttft_p95_ms": 153.81942794192582 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.00951700191945, + 153.5524509381503 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r6p00.json", + "sha256": "4562fae05d510cd50f2c9d7479916d056c8ac4d04851d2388b98f8410192c23a", + "summary": { + "admission_lag_max_ms": 0.10347133502364159, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.3970319693908, + "ttft_p50_ms": 150.414744974114, + "ttft_p95_ms": 153.00951700191945 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r6p00.json", + "sha256": "bd13283e2807150396b278f46ff630ae9820895e0c054a9efcfd8fc213947d47", + "summary": { + "admission_lag_max_ms": 0.13600767124444246, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 159.63539096992463, + "ttft_p50_ms": 151.83425601571798, + "ttft_p95_ms": 153.5524509381503 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 336.8800369789824, + 329.65395506471395 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r7p00.json", + "sha256": "fd5afe52238d6ddb72bdf7f7b142fddbeb3f62e160e4a1b3a639b308717964e6", + "summary": { + "admission_lag_max_ms": 0.14346837997436523, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 341.1952300230041, + "ttft_p50_ms": 294.81438896618783, + "ttft_p95_ms": 336.8800369789824 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r7p00.json", + "sha256": "b0fdc1ef617313261c490010adeb2f4ee93bbb053f1db10734ed5a63fd37fa76", + "summary": { + "admission_lag_max_ms": 0.11047185398638248, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 334.06115300022066, + "ttft_p50_ms": 281.1476809438318, + "ttft_p95_ms": 329.65395506471395 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + true + ], + "round_ttft_p95_ms": [ + 1289.2397560644895, + 1254.356418037787 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r8p00.json", + "sha256": "8877ed44b90f421afd22ec110fe6337581c67cd973b44378bcb7528fc8624de3", + "summary": { + "admission_lag_max_ms": 0.09899691212922335, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.9375, + "passed": 60, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1454.3529660440981, + "ttft_p50_ms": 813.0663179326802, + "ttft_p95_ms": 1289.2397560644895 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r8p00.json", + "sha256": "7ab24f5ab33bf166589af7e2f5c2a274ceef6a64f0029a96ec1879b42332123d", + "summary": { + "admission_lag_max_ms": 0.11694896966218948, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 0.953125, + "passed": 61, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1347.0055809011683, + "ttft_p50_ms": 744.0344559727237, + "ttft_p95_ms": 1254.356418037787 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4903.538785991259, + 4900.28300601989 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r16p00.json", + "sha256": "363d9b43164b5451bc3dd668caa0ce141d15114087cf444e7fd79d12b7563fd6", + "summary": { + "admission_lag_max_ms": 0.11420203372836113, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4982.48829995282, + "ttft_p50_ms": 2749.9175920384005, + "ttft_p95_ms": 4903.538785991259 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r16p00.json", + "sha256": "8aec70dfd6b69f4903bccc02c5d5474165ed07d9dd010c860ff9b447429da950", + "summary": { + "admission_lag_max_ms": 0.19414699636399746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4979.007771937177, + "ttft_p50_ms": 2749.115635990165, + "ttft_p95_ms": 4900.28300601989 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6760.187636013143, + 6534.733187989332 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r32p00.json", + "sha256": "7684ac0d2d6f851218b88fb37db3a5b54d877a6f1ead3bb604a6cb4d9a61cf3f", + "summary": { + "admission_lag_max_ms": 0.43534801807254553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6907.940509961918, + "ttft_p50_ms": 3735.540736000985, + "ttft_p95_ms": 6760.187636013143 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r32p00.json", + "sha256": "947974a9ae04e170b8c1aad32ddbda8c1000f3902a9b572c6338a7d0d4f23146", + "summary": { + "admission_lag_max_ms": 1.221106038428843, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6798.48151106853, + "ttft_p50_ms": 3595.78589303419, + "ttft_p95_ms": 6534.733187989332 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7754.704045015387, + 7756.428897031583 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r64p00.json", + "sha256": "abfc2a158087f0ad75b560dd420604a9f81359ffdf39f0ce8e2ea25b451e6904", + "summary": { + "admission_lag_max_ms": 0.13654399663209915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7857.916672946885, + "ttft_p50_ms": 4341.527819051407, + "ttft_p95_ms": 7754.704045015387 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r64p00.json", + "sha256": "192c0c850a52d2b64780bc6321cbfd10faddcc0ff10741ddad283b7d789b9def", + "summary": { + "admission_lag_max_ms": 0.11000398080796003, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7859.458995983005, + "ttft_p50_ms": 4341.894763987511, + "ttft_p95_ms": 7756.428897031583 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp1_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 99.1211449727416, + 101.08128399588168 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r4p00.json", + "sha256": "f149882da390114f2a2a140a3bd2ca6f15b6f7761591ea7b49b5a961c9d18df5", + "summary": { + "admission_lag_max_ms": 0.10085199028253555, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.28417506348342, + "ttft_p50_ms": 95.90165293775499, + "ttft_p95_ms": 99.1211449727416 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r4p00.json", + "sha256": "154dfaf2729001a8ef5b7430290f4ca906df1d4758e867d5b350489410f36d5c", + "summary": { + "admission_lag_max_ms": 0.11863000690937042, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 105.76263803523034, + "ttft_p50_ms": 96.27402503974736, + "ttft_p95_ms": 101.08128399588168 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.1850419882685, + 97.03529300168157 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r8p00.json", + "sha256": "9338d5cb994b0c12a6420f030ae2a9335a04986b4a81e31f4781ac838aeef2d4", + "summary": { + "admission_lag_max_ms": 0.12352492194622755, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.26434096693993, + "ttft_p50_ms": 95.07436107378453, + "ttft_p95_ms": 97.1850419882685 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r8p00.json", + "sha256": "a35b67970948976482ee835e595ee5631fa921607276d5ea7928460514f0d06b", + "summary": { + "admission_lag_max_ms": 0.11170096695423126, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.72453604172915, + "ttft_p50_ms": 94.03157397173345, + "ttft_p95_ms": 97.03529300168157 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.10817302018404, + 94.37024802900851 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r10p00.json", + "sha256": "f67e0f4f6b41861fcbcbee04b4efbf0dee46e273239ec94ef22d650ba6afa22a", + "summary": { + "admission_lag_max_ms": 0.10012194979935884, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.40214702207595, + "ttft_p50_ms": 94.17163301259279, + "ttft_p95_ms": 96.10817302018404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r10p00.json", + "sha256": "5a64b52f0298e9ecda2fddb8230cc27ab4b9c1db01575f68b5331816b324d1d7", + "summary": { + "admission_lag_max_ms": 0.10465399827808142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.07159405015409, + "ttft_p50_ms": 93.24954205658287, + "ttft_p95_ms": 94.37024802900851 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 447.0475659472868, + 454.8899739747867 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r12p00.json", + "sha256": "c6e27b9078281a54feeb1fab09616385a70de819791ce14f46b001bf540a80ab", + "summary": { + "admission_lag_max_ms": 0.12051803059875965, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 455.56937297806144, + "ttft_p50_ms": 280.09015799034387, + "ttft_p95_ms": 447.0475659472868 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r12p00.json", + "sha256": "861f9c4186c33e9ee6365d19c53e98c10615c5ee529a84e69f84dc292f53ac4f", + "summary": { + "admission_lag_max_ms": 0.11322635691612959, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 457.4526828946546, + "ttft_p50_ms": 290.1579710887745, + "ttft_p95_ms": 454.8899739747867 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1000.5082850111648, + 1004.6314439969137 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r14p00.json", + "sha256": "2dcaf1087edd1f12fbfe0618a6ab636b22484756a5729919bc0e51de153f8f4a", + "summary": { + "admission_lag_max_ms": 0.1881921198219061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1089.2393708927557, + "ttft_p50_ms": 611.2175739835948, + "ttft_p95_ms": 1000.5082850111648 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r14p00.json", + "sha256": "16937b31f72298c7a50ec3c7b9c1b9c798f305d83469b3d3c058efc4d53135da", + "summary": { + "admission_lag_max_ms": 0.10659568943083286, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1096.6680150013417, + "ttft_p50_ms": 617.3943240428343, + "ttft_p95_ms": 1004.6314439969137 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1458.4544260287657, + 1471.3564389385283 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r16p00.json", + "sha256": "5576a7fd06977f51f1c43bb73c3c474cc21a93cd288b422ec3d09db671b3afe7", + "summary": { + "admission_lag_max_ms": 0.135264010168612, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1583.2410319708288, + "ttft_p50_ms": 904.5663280412555, + "ttft_p95_ms": 1458.4544260287657 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r16p00.json", + "sha256": "2ccf16426046c412829442f72932268d5b27ce051a4312d135d32b240b48ec7c", + "summary": { + "admission_lag_max_ms": 0.11741300113499165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.78125, + "passed": 50, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1537.9131329245865, + "ttft_p50_ms": 896.1611289996654, + "ttft_p95_ms": 1471.3564389385283 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3348.027400090359, + 3348.575382027775 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r32p00.json", + "sha256": "d258760d15f1fdf2adefca72f7a3505c0b839036799dcb074450f7f526f76718", + "summary": { + "admission_lag_max_ms": 0.18610397819429636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3442.3487790627405, + "ttft_p50_ms": 1801.6119640087709, + "ttft_p95_ms": 3348.027400090359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r32p00.json", + "sha256": "edf5cccc1400639a94b5430897a04f20d2214bb1ffee7f4bfa62514150179a6a", + "summary": { + "admission_lag_max_ms": 0.1144439447671175, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3443.0311269825324, + "ttft_p50_ms": 1801.435066969134, + "ttft_p95_ms": 3348.575382027775 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4286.911493982188, + 4287.898648995906 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r64p00.json", + "sha256": "2a6303efb826637c3c6448d6f766f7de4fc4d7b63db163a48f72c20df68394f4", + "summary": { + "admission_lag_max_ms": 0.10997895151376724, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.311317006126, + "ttft_p50_ms": 2396.4726959820837, + "ttft_p95_ms": 4286.911493982188 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r64p00.json", + "sha256": "6fafe3bf206583b8689be1b2b57d6b92cbd1ba54f7fb230ca05f6c68b3ab9757", + "summary": { + "admission_lag_max_ms": 0.12669397983700037, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.538130956702, + "ttft_p50_ms": 2394.112486974336, + "ttft_p95_ms": 4287.898648995906 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp2_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.2925309324637, + 96.13083000294864 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r4p00.json", + "sha256": "812f8a108a9390fb759b46e5bd41e3ddad66bc144d47bf4024d834e42b47c47e", + "summary": { + "admission_lag_max_ms": 0.11386198457330465, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.76773096807301, + "ttft_p50_ms": 95.72782705072314, + "ttft_p95_ms": 97.2925309324637 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r4p00.json", + "sha256": "70f922eb7651d46d81d7d394b8453c5b778f921069cf48fb2ae2677afa94a842", + "summary": { + "admission_lag_max_ms": 0.30313804745674133, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 129.06646099872887, + "ttft_p50_ms": 94.53632694203407, + "ttft_p95_ms": 96.13083000294864 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 95.42673197574914, + 98.94769196398556 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r8p00.json", + "sha256": "24097d38b0ffb45f6cdbd93f0aad04dce3b7da41ff45e948eb3a794902b6547f", + "summary": { + "admission_lag_max_ms": 0.11571904178708792, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.04934400785714, + "ttft_p50_ms": 94.34181300457567, + "ttft_p95_ms": 95.42673197574914 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r8p00.json", + "sha256": "7251f076d34ba9484b08fa3fc4b898d2ba8d1199cb1eaeaaffc19930f8c889b9", + "summary": { + "admission_lag_max_ms": 0.1177559606730938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.93720292206854, + "ttft_p50_ms": 94.77824496570975, + "ttft_p95_ms": 98.94769196398556 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.28185909241438, + 95.9081610199064 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r10p00.json", + "sha256": "85796457715a25c15fc909f6def32776fbd49956eb7890df1c4abcef8c5afe78", + "summary": { + "admission_lag_max_ms": 0.14786410611122847, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.46025596279651, + "ttft_p50_ms": 93.54631893802434, + "ttft_p95_ms": 96.28185909241438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r10p00.json", + "sha256": "0bf4d4d006bf0b5bd8b989ceeeb4bacb0e3ee1b8ea66f30b3deb120a14871cf0", + "summary": { + "admission_lag_max_ms": 0.11162098962813616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.54603894241154, + "ttft_p50_ms": 93.77408598084003, + "ttft_p95_ms": 95.9081610199064 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 439.0145930228755, + 429.9719400005415 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r12p00.json", + "sha256": "68de6e36bd7c5624cab107fe5cb21bb817711691e545439a958c926a27e3fe29", + "summary": { + "admission_lag_max_ms": 0.10475562885403633, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 445.90794993564487, + "ttft_p50_ms": 270.3119309153408, + "ttft_p95_ms": 439.0145930228755 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r12p00.json", + "sha256": "d8d3a44a07b46f1673817ece2948fd97c30f41bd451d93a2fa02edf074536c63", + "summary": { + "admission_lag_max_ms": 0.4323029424995184, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 434.6071509644389, + "ttft_p50_ms": 265.2625450864434, + "ttft_p95_ms": 429.9719400005415 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1014.2896320903674, + 999.8802930349484 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r14p00.json", + "sha256": "10196c2f9a9b69c2e47f64d06bf5ee6863af4cd2b612301622f0e304949a942a", + "summary": { + "admission_lag_max_ms": 0.09005970787256956, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1105.1907669752836, + "ttft_p50_ms": 621.6445650206879, + "ttft_p95_ms": 1014.2896320903674 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r14p00.json", + "sha256": "c10cab85588a3563bed36fc4480462635d34d240938beaa0e5034f13186578a6", + "summary": { + "admission_lag_max_ms": 0.11327012907713652, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1090.6395060010254, + "ttft_p50_ms": 623.5586369875818, + "ttft_p95_ms": 999.8802930349484 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1488.3801550604403, + 1484.6198010491207 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r16p00.json", + "sha256": "ef5046c49c32178855bed6a0488ea21e8da22f786ddaabc503bea2d1025bb2f3", + "summary": { + "admission_lag_max_ms": 0.7892940193414688, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1613.616122980602, + "ttft_p50_ms": 905.2404230460525, + "ttft_p95_ms": 1488.3801550604403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r16p00.json", + "sha256": "de02c1d49414cffe7e20023abf4d502a4efc6d2e01fcb3e49e928081aaef77b6", + "summary": { + "admission_lag_max_ms": 0.10548799764364958, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1550.9842830942944, + "ttft_p50_ms": 904.9503420246765, + "ttft_p95_ms": 1484.6198010491207 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3391.811708919704, + 3362.9186518955976 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r32p00.json", + "sha256": "876c5444d17c4ce81eca907ae17d5cf7062ffa67fa266b491580e52f712614ec", + "summary": { + "admission_lag_max_ms": 0.12557301670312881, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3486.074637970887, + "ttft_p50_ms": 1840.204154024832, + "ttft_p95_ms": 3391.811708919704 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r32p00.json", + "sha256": "02b4cd93ce1f9816e822022fb608ba8d4c89265f673f6d27c8fa3d4f1523ffe0", + "summary": { + "admission_lag_max_ms": 0.11014204937964678, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3456.965696066618, + "ttft_p50_ms": 1812.343467026949, + "ttft_p95_ms": 3362.9186518955976 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4306.351027917117, + 4302.602957002819 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r64p00.json", + "sha256": "13ba282f179b2d6ce13a0aa2fd04eb29f82e579b7995e3406cb0afd69a2d5ee2", + "summary": { + "admission_lag_max_ms": 0.3526840591803193, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4359.505696920678, + "ttft_p50_ms": 2411.1130589153618, + "ttft_p95_ms": 4306.351027917117 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r64p00.json", + "sha256": "767622d62f0757ef2d8c7651ea470bfa2e62edad0ed4e75f66f84a0b722e35da", + "summary": { + "admission_lag_max_ms": 0.10626309085637331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4355.409953044727, + "ttft_p50_ms": 2409.5974090741947, + "ttft_p95_ms": 4302.602957002819 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp2_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.54812794178724, + 95.08875198662281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r4p00.json", + "sha256": "6e6a45d0c1010094c3395969815fd8cf9c35b919db202d772fb514744f981ebc", + "summary": { + "admission_lag_max_ms": 0.11306197848170996, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.27749610505998, + "ttft_p50_ms": 93.01548008807003, + "ttft_p95_ms": 96.54812794178724 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r4p00.json", + "sha256": "58aa71d694704cfea15c37cb29420dbac1dc505045c29a731e8be904846cca5f", + "summary": { + "admission_lag_max_ms": 0.11853000614792109, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 98.74943597242236, + "ttft_p50_ms": 93.57558900956064, + "ttft_p95_ms": 95.08875198662281 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.24941800534725, + 96.21043398510665 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r8p00.json", + "sha256": "311350af236cb0306aa635f05f8999b66bf5080bad7cc1f6703fffe0be31aee6", + "summary": { + "admission_lag_max_ms": 0.19771000370383263, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52211803849787, + "ttft_p50_ms": 92.4453770276159, + "ttft_p95_ms": 94.24941800534725 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r8p00.json", + "sha256": "60fe989a36a526e76f6f5c98e217b9feaf081bc680b16b9d53650bf085047c06", + "summary": { + "admission_lag_max_ms": 0.3933720290660858, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.97230398282409, + "ttft_p50_ms": 93.26327103190124, + "ttft_p95_ms": 96.21043398510665 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.98661107197404, + 92.9473100695759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r10p00.json", + "sha256": "d0a04444a0e41bf76f71ac6ea3815bfbe76f7a51f8ddf669976bc3a4896dd0c1", + "summary": { + "admission_lag_max_ms": 0.10270101483911276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 103.49742788821459, + "ttft_p50_ms": 94.55623896792531, + "ttft_p95_ms": 97.98661107197404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r10p00.json", + "sha256": "f330f26f45f6039c48150a0b6b146ca3dbb9ea8b2dff086a9efd5938ab981eb7", + "summary": { + "admission_lag_max_ms": 0.11239596642553806, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.4133670097217, + "ttft_p50_ms": 91.2688790122047, + "ttft_p95_ms": 92.9473100695759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 260.68170997314155, + 319.04888805001974 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r12p00.json", + "sha256": "fed5eacf18018402b0c2675b8c5595f92693a9bddec30f354d989098f9fe4965", + "summary": { + "admission_lag_max_ms": 0.11784560047090054, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 345.19017208367586, + "ttft_p50_ms": 211.41935500781983, + "ttft_p95_ms": 260.68170997314155 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r12p00.json", + "sha256": "f2be27cbeff3e5c1a95050418f2dc3c712a4648cada8cdb3d36382cfffc9593a", + "summary": { + "admission_lag_max_ms": 0.12930459342896938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 326.86293800361454, + "ttft_p50_ms": 229.8808820778504, + "ttft_p95_ms": 319.04888805001974 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 895.7287209341303, + 912.1733580250293 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r14p00.json", + "sha256": "c2e157808e0308cc443f3abdf5114898cef66c435688d24321f80bef2d966208", + "summary": { + "admission_lag_max_ms": 0.14902697876095772, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 962.3544370988384, + "ttft_p50_ms": 576.2168710352853, + "ttft_p95_ms": 895.7287209341303 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r14p00.json", + "sha256": "749ef7d31bf8782d42719b75ee1a2e4a443f097510af721d02cf53df191bc66e", + "summary": { + "admission_lag_max_ms": 0.42896345257759094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 986.0872740391642, + "ttft_p50_ms": 569.7114520007744, + "ttft_p95_ms": 912.1733580250293 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1364.6171110449359, + 1366.6857269126922 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r16p00.json", + "sha256": "7545ea9210348019de88ede5d2cba7527913534274deee034fa6811c173a6e72", + "summary": { + "admission_lag_max_ms": 0.12612901628017426, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1435.8308439841494, + "ttft_p50_ms": 848.9949750946835, + "ttft_p95_ms": 1364.6171110449359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r16p00.json", + "sha256": "d75e27137f6a7811126d8c957bb5a081944fc1bb7643852cdd785111773f6010", + "summary": { + "admission_lag_max_ms": 0.1258660340681672, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1436.2916270038113, + "ttft_p50_ms": 846.2129919789732, + "ttft_p95_ms": 1366.6857269126922 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3243.9196659252048, + 3214.524411014281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r32p00.json", + "sha256": "366ebddc167338ed096f373de5667740721422c13d2db1f887ed09186afafd19", + "summary": { + "admission_lag_max_ms": 0.09669095743447542, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3337.068362045102, + "ttft_p50_ms": 1751.3202880509198, + "ttft_p95_ms": 3243.9196659252048 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r32p00.json", + "sha256": "01b8f5081e36a64d5776285c8c5f423b53b1ad684c4b8f95a8caa5d997c6bf09", + "summary": { + "admission_lag_max_ms": 0.8310690755024552, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3277.297410997562, + "ttft_p50_ms": 1790.5721209244803, + "ttft_p95_ms": 3214.524411014281 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4319.725521025248, + 4367.031275993213 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r64p00.json", + "sha256": "92202709fd6bbd558ff6f1986df213848254110c4274447a7bd5879153290d46", + "summary": { + "admission_lag_max_ms": 0.10392803233116865, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4360.145599930547, + "ttft_p50_ms": 2403.5812759539112, + "ttft_p95_ms": 4319.725521025248 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r64p00.json", + "sha256": "9660d15d31bc70c5550ed29477f563313d4d81baf0e561126aebd3b7d097cfa3", + "summary": { + "admission_lag_max_ms": 0.17276196740567684, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.25, + "passed": 16, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4414.442001027055, + "ttft_p50_ms": 2260.2788450894877, + "ttft_p95_ms": 4367.031275993213 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp2_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.17482794076204, + 98.03147299680859 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r4p00.json", + "sha256": "0e315f6d9e0087ad553e874b0c16c2e6de79eb6eca22eb118dc5deca52e6efe3", + "summary": { + "admission_lag_max_ms": 0.11708191595971584, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.92840604577214, + "ttft_p50_ms": 95.13642196543515, + "ttft_p95_ms": 97.17482794076204 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r4p00.json", + "sha256": "baa394a24d2a3db40a3a7ae6dae0c5ce49ff05499a1e17eaa0bca4b479ae33bf", + "summary": { + "admission_lag_max_ms": 0.10946101974695921, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52333807107061, + "ttft_p50_ms": 93.19468098692596, + "ttft_p95_ms": 98.03147299680859 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.67980605829507, + 93.80627400241792 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r8p00.json", + "sha256": "303df27c6d4604d5af65f9056acacb7c4373fb6eb7edb6e5d42689eeaa384aaa", + "summary": { + "admission_lag_max_ms": 0.11111306957900524, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 113.2329449756071, + "ttft_p50_ms": 93.05417106952518, + "ttft_p95_ms": 97.67980605829507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r8p00.json", + "sha256": "cdbb1bca17b24b67e18ef3dd4997494d6bd2a887b3b5d1264fbf411e688c53bc", + "summary": { + "admission_lag_max_ms": 0.12463307939469814, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.12245299387723, + "ttft_p50_ms": 92.26030297577381, + "ttft_p95_ms": 93.80627400241792 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.45723297540098, + 93.57301203999668 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r10p00.json", + "sha256": "fe17ef2417319307d651bc10b01a618234d61bab797ad87a7332cab20d78ecb2", + "summary": { + "admission_lag_max_ms": 0.3267280990257859, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.28593400400132, + "ttft_p50_ms": 92.22474799025804, + "ttft_p95_ms": 94.45723297540098 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r10p00.json", + "sha256": "550f693708a34624a82d77afd4da0ab4bc01e09f42242a3a79e4532897681a51", + "summary": { + "admission_lag_max_ms": 0.11238199658691883, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.77756198309362, + "ttft_p50_ms": 92.07128698471934, + "ttft_p95_ms": 93.57301203999668 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 320.49866404850036, + 336.9630330707878 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r12p00.json", + "sha256": "edab8169453151de732923c076b8f7dca4082ee47ccd2597e65e7ba68516bc2e", + "summary": { + "admission_lag_max_ms": 0.09914662223309278, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 327.930542989634, + "ttft_p50_ms": 231.4880370395258, + "ttft_p95_ms": 320.49866404850036 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r12p00.json", + "sha256": "a6dde4363d362ca10254409e3dc734d0f0e640e63156426f3f4fd7115db49e16", + "summary": { + "admission_lag_max_ms": 0.17016706988215446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 428.6657409975305, + "ttft_p50_ms": 237.09688894450665, + "ttft_p95_ms": 336.9630330707878 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 911.7195709841326, + 910.7652489328757 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r14p00.json", + "sha256": "7eba81ee5c85c457a056f9aef28abd94d375fdaa116929cfe0a0c282a0738bd2", + "summary": { + "admission_lag_max_ms": 0.13163150288164616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 999.2953530745581, + "ttft_p50_ms": 580.023527960293, + "ttft_p95_ms": 911.7195709841326 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r14p00.json", + "sha256": "6a113d3d876ff01a9c1101e8b75efa2fc71f96683d6bb430e26d5976426b5d23", + "summary": { + "admission_lag_max_ms": 0.11634768452495337, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 988.8870939612389, + "ttft_p50_ms": 578.9327350212261, + "ttft_p95_ms": 910.7652489328757 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1389.8832038976252, + 1365.0342189939693 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r16p00.json", + "sha256": "d677251ee1a941c52469052a239c2e13be0a837285069aa8182a16e196445e48", + "summary": { + "admission_lag_max_ms": 0.11070503387600183, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1460.5093500576913, + "ttft_p50_ms": 864.3530959961936, + "ttft_p95_ms": 1389.8832038976252 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r16p00.json", + "sha256": "f5177f53aa0f87a4e7291f470978d86ad8bc46bd1a8362b79f48cd128646dc10", + "summary": { + "admission_lag_max_ms": 0.3672370221465826, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1491.050804965198, + "ttft_p50_ms": 841.7225609300658, + "ttft_p95_ms": 1365.0342189939693 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3258.010295103304, + 3258.664960041642 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r32p00.json", + "sha256": "02a6ac2b9c2b9ab97854d6fcc13b64bbb36a079e4afdae84231e4cb187fac4ef", + "summary": { + "admission_lag_max_ms": 0.1296499976888299, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.278529899195, + "ttft_p50_ms": 1755.0864539807662, + "ttft_p95_ms": 3258.010295103304 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r32p00.json", + "sha256": "9ca2f44fc7e3cb3d38ada548af858903b31bbee5cd50f1723c29255ebd4224e9", + "summary": { + "admission_lag_max_ms": 1.1343059595674276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.7101600775495, + "ttft_p50_ms": 1755.3453800501302, + "ttft_p95_ms": 3258.664960041642 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4190.255103982054, + 4189.1050330596045 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r64p00.json", + "sha256": "53908d3a254e19691839831b4337014b09591ced9189c2a9b29e0d92e80fcb56", + "summary": { + "admission_lag_max_ms": 0.11591101065278053, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4242.029857938178, + "ttft_p50_ms": 2340.5529640149325, + "ttft_p95_ms": 4190.255103982054 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r64p00.json", + "sha256": "1c8c85c873aa96c01dc0200f848474a53c1173ef72f3828e8c23aeae70e92b11", + "summary": { + "admission_lag_max_ms": 0.284243025816977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4240.770195960067, + "ttft_p50_ms": 2339.3583690049127, + "ttft_p95_ms": 4189.1050330596045 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp2_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 74.66826005838811, + 66.5705680148676 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r4p00.json", + "sha256": "fb354fc8a097caa21844d4ea5515b98b03459f400816e5e5590a6cda81cf4815", + "summary": { + "admission_lag_max_ms": 0.1035250024870038, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 93.38432096410543, + "ttft_p50_ms": 65.75745099689811, + "ttft_p95_ms": 74.66826005838811 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r4p00.json", + "sha256": "ac8e8412f79a8a32ebf085445a94aa548dd2d646fce0b9f56b4378081dcc8960", + "summary": { + "admission_lag_max_ms": 0.12108997907489538, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.89856600761414, + "ttft_p50_ms": 62.386377016082406, + "ttft_p95_ms": 66.5705680148676 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.44799195136875, + 72.29184894822538 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r8p00.json", + "sha256": "05ebbd68bbf24e9c8b25684474411c7db122b92701cc1617f9d786cf19592751", + "summary": { + "admission_lag_max_ms": 0.1825150102376938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.92095802165568, + "ttft_p50_ms": 60.598687967285514, + "ttft_p95_ms": 70.44799195136875 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r8p00.json", + "sha256": "d24f7daaf41073cc8aa030b74c06203783684c3e556770366883c37b826406f2", + "summary": { + "admission_lag_max_ms": 0.10628998279571533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 75.13270596973598, + "ttft_p50_ms": 60.161596979014575, + "ttft_p95_ms": 72.29184894822538 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.02151494007558, + 107.58775402791798 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r16p00.json", + "sha256": "a29b2128983af630e1da8f970cbda802849036bc522eb506821bf2fe1a34172f", + "summary": { + "admission_lag_max_ms": 0.10468997061252594, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.93136999942362, + "ttft_p50_ms": 59.5597909996286, + "ttft_p95_ms": 66.02151494007558 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r16p00.json", + "sha256": "275ae35651cbe2200d16c9b06a68568efda527564065255a10dee6fe93346010", + "summary": { + "admission_lag_max_ms": 0.3144120564684272, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 119.78317995090038, + "ttft_p50_ms": 59.12985489703715, + "ttft_p95_ms": 107.58775402791798 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 193.8455159543082, + 187.52167106140405 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r20p00.json", + "sha256": "4cb485f30b81c004bab379e6522d94b811c0bdc0be6d8e42c9d8ae842d9600dc", + "summary": { + "admission_lag_max_ms": 0.10407401714473963, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 197.7908629924059, + "ttft_p50_ms": 142.73560303263366, + "ttft_p95_ms": 193.8455159543082 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r20p00.json", + "sha256": "63aa9dc03a08b781c5d6e8805f2b61698c75e71673608ecd38bd1132c62e6987", + "summary": { + "admission_lag_max_ms": 0.09386602323502302, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 190.29632199089974, + "ttft_p50_ms": 136.1723990412429, + "ttft_p95_ms": 187.52167106140405 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 568.2198900030926, + 573.4539540717378 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r24p00.json", + "sha256": "5e4d5b44850090d07fe1d3abc44eddea3bfdf2b682f59b12a45a866621e38f98", + "summary": { + "admission_lag_max_ms": 0.7511250441893935, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 613.2891150191426, + "ttft_p50_ms": 375.27791305910796, + "ttft_p95_ms": 568.2198900030926 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r24p00.json", + "sha256": "239e7789aed2e9b2bcf1d127215a6b122071eb856be2db64ccc184fda153aa48", + "summary": { + "admission_lag_max_ms": 0.11180061846971512, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 614.9205249967054, + "ttft_p50_ms": 375.5133739905432, + "ttft_p95_ms": 573.4539540717378 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 860.0242879474536, + 886.1482660286129 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r28p00.json", + "sha256": "8b249b52779e5b3cd9028aaee30ee73ef34b0b0a64dcd75b9aad57266c29fbc2", + "summary": { + "admission_lag_max_ms": 0.11425174307078123, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.8854410191998, + "ttft_p50_ms": 534.7980660153553, + "ttft_p95_ms": 860.0242879474536 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r28p00.json", + "sha256": "c9b28c8f3187542a343ecf1067783569ec2abfa64631d6827a94ba6876aa7b63", + "summary": { + "admission_lag_max_ms": 0.312068616040051, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 922.4315020255744, + "ttft_p50_ms": 549.7413279954344, + "ttft_p95_ms": 886.1482660286129 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1147.1082780044526, + 1137.236590962857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r32p00.json", + "sha256": "d27ff9f940a6cd7fba339d9a176de39cf4b5bd164dcc500f2af36bcb07cdef30", + "summary": { + "admission_lag_max_ms": 0.16146607231348753, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.626160996966, + "ttft_p50_ms": 696.4708579471335, + "ttft_p95_ms": 1147.1082780044526 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r32p00.json", + "sha256": "093044e87fffdaa222e5bad3807cbf4900aa64891d5b4ba27f187c7a411252d7", + "summary": { + "admission_lag_max_ms": 0.10430300608277321, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.045956928283, + "ttft_p50_ms": 684.3201500596479, + "ttft_p95_ms": 1137.236590962857 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2064.763988018967, + 2039.6780440350994 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r64p00.json", + "sha256": "fad7bdd80d5ca603790a4e5d19dadd4f494540b106ce583e3e8c1c05d2374a58", + "summary": { + "admission_lag_max_ms": 6.394341005943716, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.53125, + "passed": 34, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2144.80376499705, + "ttft_p50_ms": 1158.5425049997866, + "ttft_p95_ms": 2064.763988018967 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r64p00.json", + "sha256": "24fdb945706dbbdd29916e0f08e0eff807334db212fded94c729da2e1b46efd3", + "summary": { + "admission_lag_max_ms": 0.10294502135366201, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.4275270039216, + "ttft_p50_ms": 1131.1213460285217, + "ttft_p95_ms": 2039.6780440350994 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.80304794665426, + 64.59337309934199 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r4p00.json", + "sha256": "2e82a30f02c2702668e59661930431cf82323746a111d42bf38b7fdc0d083b3d", + "summary": { + "admission_lag_max_ms": 0.1332589890807867, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.74050498381257, + "ttft_p50_ms": 64.10286005120724, + "ttft_p95_ms": 67.80304794665426 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r4p00.json", + "sha256": "8294196c7dc52765a6820f5e1d8439c480490daed9505263f707810f73c95783", + "summary": { + "admission_lag_max_ms": 0.1283179735764861, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.6796320155263, + "ttft_p50_ms": 60.962425894103944, + "ttft_p95_ms": 64.59337309934199 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.46496798284352, + 68.22343391831964 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r8p00.json", + "sha256": "38bbfeab47e5eb6296fcccf20f04ef25dc24db658be92300c492734bbfe333cf", + "summary": { + "admission_lag_max_ms": 0.10555307380855083, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 87.7126749837771, + "ttft_p50_ms": 62.65173899009824, + "ttft_p95_ms": 70.46496798284352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r8p00.json", + "sha256": "716f90177d6f203717c80a8e9b28697e53141f204af470493c299bc73a1ea3ee", + "summary": { + "admission_lag_max_ms": 0.24518708232790232, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 80.40577603969723, + "ttft_p50_ms": 60.29119901359081, + "ttft_p95_ms": 68.22343391831964 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 89.53388908412308, + 76.93439698778093 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r16p00.json", + "sha256": "063016efe822c31c2425e7d963eda4c054f3f7c0341df0a3cd73ab86a942e873", + "summary": { + "admission_lag_max_ms": 0.2996840048581362, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.41822991054505, + "ttft_p50_ms": 60.97789097111672, + "ttft_p95_ms": 89.53388908412308 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r16p00.json", + "sha256": "eecc6bdeff90a8bf195dfc5387c6bc2f4db20ed7c9aeb9b4ebe050efccc6051a", + "summary": { + "admission_lag_max_ms": 0.36267400719225407, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.42048599943519, + "ttft_p50_ms": 60.29245804529637, + "ttft_p95_ms": 76.93439698778093 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 206.2494510319084, + 201.13301498349756 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r20p00.json", + "sha256": "22a62729358fea7c38ca166a1a49f1235e98c7f07ecfe4948322110cda4b7178", + "summary": { + "admission_lag_max_ms": 0.11367700062692165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 255.14698098413646, + "ttft_p50_ms": 155.46583198010921, + "ttft_p95_ms": 206.2494510319084 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r20p00.json", + "sha256": "be103a759100bd2ab6dd31ee4ce41b6f0da40d232a60ebe137e3d933194a487c", + "summary": { + "admission_lag_max_ms": 0.17853605095297098, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 209.22932599205524, + "ttft_p50_ms": 155.12292610947043, + "ttft_p95_ms": 201.13301498349756 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 580.5047769099474, + 577.3971510352567 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r24p00.json", + "sha256": "890a733a97f399deb4c5efcca661d9a442ed76080cd5d8b6a2c09368aef701b7", + "summary": { + "admission_lag_max_ms": 0.1084543764591217, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 621.5670789824799, + "ttft_p50_ms": 374.6346529806033, + "ttft_p95_ms": 580.5047769099474 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r24p00.json", + "sha256": "87e8543652b1743b1b9b8c3de0fc680401b0bf9a33d61fa6132120c944b2a6eb", + "summary": { + "admission_lag_max_ms": 0.11801638174802065, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 623.2662099646404, + "ttft_p50_ms": 382.11427396163344, + "ttft_p95_ms": 577.3971510352567 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 873.0683700414374, + 889.4930009264499 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r28p00.json", + "sha256": "f877781730a5437b4cbdb76948915f1c871574b19a6df27076d21b65bb18e678", + "summary": { + "admission_lag_max_ms": 0.6807943573221564, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 943.9258739585057, + "ttft_p50_ms": 542.9852310335264, + "ttft_p95_ms": 873.0683700414374 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r28p00.json", + "sha256": "82151dc539b35302489c17976cfe89b8f4c64944298954607fad8f5c9cb9e7d8", + "summary": { + "admission_lag_max_ms": 6.750226486474276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 936.5667289821431, + "ttft_p50_ms": 546.0105559322983, + "ttft_p95_ms": 889.4930009264499 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1135.3467140579596, + 1146.743991994299 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r32p00.json", + "sha256": "082cc4c47b2eec6e49f0edd14cff5a51b880bd998d98da8b9830bbf7615483ef", + "summary": { + "admission_lag_max_ms": 0.10443700011819601, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1199.8225760180503, + "ttft_p50_ms": 680.4794340860099, + "ttft_p95_ms": 1135.3467140579596 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r32p00.json", + "sha256": "517d3cbc1104b5ea1bdb0e381c033cf978181a93449b6498d30b52cd3dd3e770", + "summary": { + "admission_lag_max_ms": 0.4415180301293731, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1210.0602959981188, + "ttft_p50_ms": 693.0603379150853, + "ttft_p95_ms": 1146.743991994299 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2084.9693010095507, + 2041.9401369290426 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r64p00.json", + "sha256": "7e11771ed598826fb3c5594c7e05d0989e01030be2d8b36fa21813aff462496d", + "summary": { + "admission_lag_max_ms": 0.10768393985927105, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.128676981665, + "ttft_p50_ms": 1113.0280290963128, + "ttft_p95_ms": 2084.9693010095507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r64p00.json", + "sha256": "cf550423832a92cba942b7e1dc14a42000e8c9507403de8cd49aa56af9db6706", + "summary": { + "admission_lag_max_ms": 0.11262902989983559, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2122.8132860269397, + "ttft_p50_ms": 1131.2310450011864, + "ttft_p95_ms": 2041.9401369290426 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.89354998525232, + 83.23234610725194 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r4p00.json", + "sha256": "266cba4088f8a9483ceb41fdc02d04a2181cb7ffb6b52372b6b0687e62f2863e", + "summary": { + "admission_lag_max_ms": 0.10568101424723864, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.09669100120664, + "ttft_p50_ms": 62.014322029426694, + "ttft_p95_ms": 66.89354998525232 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r4p00.json", + "sha256": "26f3926c40504b96a37cb7b9dae1af5040c085c1a730e20feaf1d3e9ddbf8e54", + "summary": { + "admission_lag_max_ms": 0.108166947029531, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.42452490516007, + "ttft_p50_ms": 61.261708033271134, + "ttft_p95_ms": 83.23234610725194 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.26283801253885, + 67.34374002553523 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r8p00.json", + "sha256": "e84e6ba2c8822e012aceea7cc6f376ea1b6265fe4a3bb68b7badfc36b8813545", + "summary": { + "admission_lag_max_ms": 0.15694997273385525, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54619695153087, + "ttft_p50_ms": 59.63437003083527, + "ttft_p95_ms": 67.26283801253885 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r8p00.json", + "sha256": "dae50073c9f79919181e8dcf274ca8929ba41c512f64363e40187de68839b87e", + "summary": { + "admission_lag_max_ms": 0.09962497279047966, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.00161597877741, + "ttft_p50_ms": 60.2562150452286, + "ttft_p95_ms": 67.34374002553523 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 71.37343694921583, + 64.9534990079701 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r16p00.json", + "sha256": "776c9956f65acf7c9b6c8df517a4f8c4d462289893a19e625a2a0ad7dcbf5506", + "summary": { + "admission_lag_max_ms": 0.1066719414666295, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 82.25136005785316, + "ttft_p50_ms": 58.52357891853899, + "ttft_p95_ms": 71.37343694921583 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r16p00.json", + "sha256": "90057204a5f6343ee4ee9cdd08323d87d83660d5bf8f02480882eb479d2e587e", + "summary": { + "admission_lag_max_ms": 0.10102998930960894, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 65.3582609957084, + "ttft_p50_ms": 58.98014397826046, + "ttft_p95_ms": 64.9534990079701 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 192.98715889453888, + 258.1990590551868 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r20p00.json", + "sha256": "6ce029388b206d0663acd41f4da1d67a6562389e1c84cf1826ce77c05d176d9a", + "summary": { + "admission_lag_max_ms": 0.12691703159362078, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 198.67739104665816, + "ttft_p50_ms": 142.75032398290932, + "ttft_p95_ms": 192.98715889453888 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r20p00.json", + "sha256": "0dbfb3f012574988ddebfb0af1532c735ac62d6d88863d24cd820aed50a92a59", + "summary": { + "admission_lag_max_ms": 0.3961039474233985, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 259.55891504418105, + "ttft_p50_ms": 162.6282810466364, + "ttft_p95_ms": 258.1990590551868 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 583.3089610096067, + 571.3486529421061 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r24p00.json", + "sha256": "0774ba2f285d81447c63343b32551d6fc33369b54010629381efa9a37df23a9c", + "summary": { + "admission_lag_max_ms": 0.09822729043662548, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 628.714332007803, + "ttft_p50_ms": 366.9840869260952, + "ttft_p95_ms": 583.3089610096067 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r24p00.json", + "sha256": "1986a05c366ebe73af0f89d2cd0e4efd3e577c800cf1e80997cd26ea0aaaec09", + "summary": { + "admission_lag_max_ms": 0.11414836626499891, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 617.9165309295058, + "ttft_p50_ms": 376.65714998729527, + "ttft_p95_ms": 571.3486529421061 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 869.708820944652, + 866.4831441128626 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r28p00.json", + "sha256": "ab987729fc833a59817eabaa18c320b0d0a2ecd402fd02bdb534b6c773c029d8", + "summary": { + "admission_lag_max_ms": 0.578614417463541, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 941.3277570856735, + "ttft_p50_ms": 540.4732850147411, + "ttft_p95_ms": 869.708820944652 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r28p00.json", + "sha256": "38f9c82c9ee30927cc0ba857e5e3fce2e79b8a24de03efaec77ce27258633a7e", + "summary": { + "admission_lag_max_ms": 0.12532551772892475, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 938.2033230504021, + "ttft_p50_ms": 541.139594046399, + "ttft_p95_ms": 866.4831441128626 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1132.6635719742626, + 1144.196804962121 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r32p00.json", + "sha256": "358f5b8e1b72bff5752c076f2966b966d98f6f5767ce52d5a255b7f68f061e54", + "summary": { + "admission_lag_max_ms": 0.32093306072056293, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1197.7118930080906, + "ttft_p50_ms": 678.4568330040202, + "ttft_p95_ms": 1132.6635719742626 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r32p00.json", + "sha256": "3f6dbb760582fdd7470422d8edf6b62949d158ad0d437f04015441a0a4509003", + "summary": { + "admission_lag_max_ms": 1.1304179206490517, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1193.1529619032517, + "ttft_p50_ms": 692.5535890040919, + "ttft_p95_ms": 1144.196804962121 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2042.27216495201, + 2043.943466967903 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r64p00.json", + "sha256": "a24eae7db6eb4cd105ed903917988e8fe6eef5d4cc2549d4e05e9a03dfafb107", + "summary": { + "admission_lag_max_ms": 0.134881935082376, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.23939598538, + "ttft_p50_ms": 1137.3268909519538, + "ttft_p95_ms": 2042.27216495201 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r64p00.json", + "sha256": "e85efaada0b64bb9dd686f59282171ccb256a1d8483c6f037589a68e3e1a104a", + "summary": { + "admission_lag_max_ms": 0.11892011389136314, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.3299129894003, + "ttft_p50_ms": 1134.724210947752, + "ttft_p95_ms": 2043.943466967903 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 82.33969006687403, + 69.57727600820363 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r4p00.json", + "sha256": "e69b89ad2e07e6560ebd870773532b33e4415e9ca1515e585cbe465616678569", + "summary": { + "admission_lag_max_ms": 0.14186999760568142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54833806212991, + "ttft_p50_ms": 62.749656033702195, + "ttft_p95_ms": 82.33969006687403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r4p00.json", + "sha256": "2c76dcf112a828a6720e079d29bb54a6a65d37d26abaaafdc6d5580751dbad65", + "summary": { + "admission_lag_max_ms": 0.11478399392217398, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 81.3983449479565, + "ttft_p50_ms": 62.220132909715176, + "ttft_p95_ms": 69.57727600820363 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 63.71171900536865, + 68.14328604377806 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r8p00.json", + "sha256": "0e5d687dfcdb4b6a0d1d060ceefe8a74faab94a5b241580d45b53c452f893b23", + "summary": { + "admission_lag_max_ms": 0.09818794205784798, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.63772000744939, + "ttft_p50_ms": 60.89008797425777, + "ttft_p95_ms": 63.71171900536865 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r8p00.json", + "sha256": "1bf016b603473d46fda4bc6d4daeaee5e1091ff4cda1cab7224e0c5edea512f8", + "summary": { + "admission_lag_max_ms": 0.11399004142731428, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 76.76690397784114, + "ttft_p50_ms": 60.9520060243085, + "ttft_p95_ms": 68.14328604377806 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 133.86176899075508, + 61.95738597307354 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r16p00.json", + "sha256": "197808b1b5fa8082e603a255f358826b5f8f48707223c7fd63abfe029674b817", + "summary": { + "admission_lag_max_ms": 0.10530801955610514, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 191.68931001331657, + "ttft_p50_ms": 61.92394997924566, + "ttft_p95_ms": 133.86176899075508 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r16p00.json", + "sha256": "53e809470daf9efcdabb254e85e8cc3b45d0ab8593c6c2f3aa3396e60d1656da", + "summary": { + "admission_lag_max_ms": 0.1288399798795581, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.40882098954171, + "ttft_p50_ms": 59.417470009066164, + "ttft_p95_ms": 61.95738597307354 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 201.29013701807708, + 200.0734859611839 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r20p00.json", + "sha256": "05b977824fb301517cdbe2d077b360b6053d57fee1c20acad3537321827d027a", + "summary": { + "admission_lag_max_ms": 0.3848400665447116, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.5107699232176, + "ttft_p50_ms": 150.0862929970026, + "ttft_p95_ms": 201.29013701807708 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r20p00.json", + "sha256": "baf422b7a3c95b0d3d376a31b0a30e075f69216c794ebc2ef51afafe1ec55229", + "summary": { + "admission_lag_max_ms": 0.24754798505455256, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.8192109214142, + "ttft_p50_ms": 148.99871195666492, + "ttft_p95_ms": 200.0734859611839 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 564.2021730309352, + 560.4996709153056 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r24p00.json", + "sha256": "029b2a9695322d536539c3dd7665725f933b21a8bb48fa94c0448c9f7f8bf0eb", + "summary": { + "admission_lag_max_ms": 0.14956865925341845, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 610.4218009859324, + "ttft_p50_ms": 368.3909500250593, + "ttft_p95_ms": 564.2021730309352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r24p00.json", + "sha256": "94b46583af3a7a4bbb9f7256c4c4a5ce804b3092459abe3984ea0ce67f591f3c", + "summary": { + "admission_lag_max_ms": 0.11110503692179918, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 606.6705230623484, + "ttft_p50_ms": 367.50321893487126, + "ttft_p95_ms": 560.4996709153056 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 870.7952990662307, + 894.6108610834926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r28p00.json", + "sha256": "07bfe263107bb3d2acaa356257343f90045e4295c799178bc0a0b5f1c1994284", + "summary": { + "admission_lag_max_ms": 0.6798935355618596, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 942.792383953929, + "ttft_p50_ms": 543.340316042304, + "ttft_p95_ms": 870.7952990662307 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r28p00.json", + "sha256": "861dd55076ef40a78c34824345c5279dff86262b604db120e9e38efa27b5be55", + "summary": { + "admission_lag_max_ms": 0.09912776295095682, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.2741150604561, + "ttft_p50_ms": 549.9696529004723, + "ttft_p95_ms": 894.6108610834926 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1136.603599996306, + 1134.0074760373682 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r32p00.json", + "sha256": "332b805322b571087b018ee459dd91c443dfe0b2bb539b5bdfb4de902bba6da2", + "summary": { + "admission_lag_max_ms": 0.19067595712840557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.8779760692269, + "ttft_p50_ms": 683.3873559953645, + "ttft_p95_ms": 1136.603599996306 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r32p00.json", + "sha256": "996a12a5ed962218dfbe6f30be688119030f10f6539647b6e4a9f25f032b22cb", + "summary": { + "admission_lag_max_ms": 0.11958007235080004, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1198.4891929896548, + "ttft_p50_ms": 682.7383439522237, + "ttft_p95_ms": 1134.0074760373682 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2040.8174369949847, + 2082.3236129945144 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r64p00.json", + "sha256": "c5f17f625071e905b5f1bf8045cd233b2fea0fe6ca00044935dee9c663d1526b", + "summary": { + "admission_lag_max_ms": 0.7995429914444685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2131.160691031255, + "ttft_p50_ms": 1134.8516750149429, + "ttft_p95_ms": 2040.8174369949847 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r64p00.json", + "sha256": "775db5b941347d9fd5d6df32d6ee037f8168bd47b9eae9a7bdb3dbd8c40aa990", + "summary": { + "admission_lag_max_ms": 0.17620902508497238, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2129.518053960055, + "ttft_p50_ms": 1112.3740070033818, + "ttft_p95_ms": 2082.3236129945144 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + }, + "real_campaign_resources": { + "fresh_server_anchors": 192, + "gpu_hours": 12.074444444444445, + "measured_requests": 12288, + "runs": [ + { + "duration_seconds": 908.0, + "gpu_hours": 0.25222222222222224, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive-20260716T164730291172Z", + "tp": 1 + }, + { + "duration_seconds": 540.0, + "gpu_hours": 0.15, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + { + "duration_seconds": 937.0, + "gpu_hours": 0.2602777777777778, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive-20260716T164731376588Z", + "tp": 1 + }, + { + "duration_seconds": 558.0, + "gpu_hours": 0.155, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + { + "duration_seconds": 972.0, + "gpu_hours": 0.27, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive-20260716T164732482666Z", + "tp": 1 + }, + { + "duration_seconds": 572.0, + "gpu_hours": 0.15888888888888889, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + { + "duration_seconds": 886.0, + "gpu_hours": 0.2461111111111111, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive-20260716T164729267199Z", + "tp": 1 + }, + { + "duration_seconds": 525.0, + "gpu_hours": 0.14583333333333334, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + { + "duration_seconds": 983.0, + "gpu_hours": 0.5461111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive-20260716T164734745039Z", + "tp": 2 + }, + { + "duration_seconds": 569.0, + "gpu_hours": 0.3161111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + { + "duration_seconds": 1033.0, + "gpu_hours": 0.5738888888888889, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive-20260716T170909930153Z", + "tp": 2 + }, + { + "duration_seconds": 584.0, + "gpu_hours": 0.3244444444444444, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + { + "duration_seconds": 1063.0, + "gpu_hours": 0.5905555555555555, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive-20260716T170910954319Z", + "tp": 2 + }, + { + "duration_seconds": 597.0, + "gpu_hours": 0.33166666666666667, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + { + "duration_seconds": 957.0, + "gpu_hours": 0.5316666666666666, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive-20260716T164733581881Z", + "tp": 2 + }, + { + "duration_seconds": 563.0, + "gpu_hours": 0.31277777777777777, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + { + "duration_seconds": 1008.0, + "gpu_hours": 1.12, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive-20260716T172922998830Z", + "tp": 4 + }, + { + "duration_seconds": 559.0, + "gpu_hours": 0.6211111111111111, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + { + "duration_seconds": 1024.0, + "gpu_hours": 1.1377777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive-20260716T172923955862Z", + "tp": 4 + }, + { + "duration_seconds": 556.0, + "gpu_hours": 0.6177777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + { + "duration_seconds": 959.0, + "gpu_hours": 1.0655555555555556, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive-20260716T175248149074Z", + "tp": 4 + }, + { + "duration_seconds": 567.0, + "gpu_hours": 0.63, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + { + "duration_seconds": 995.0, + "gpu_hours": 1.1055555555555556, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive-20260716T170911977980Z", + "tp": 4 + }, + { + "duration_seconds": 550.0, + "gpu_hours": 0.6111111111111112, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + ], + "successful_fleet_jobs": 24, + "warmup_requests": 4512 + }, + "schema": "qwen30-prefill-fidelity-comparison-v1", + "simulator": { + "tp1_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 636.4131651102598 + }, + { + "feasible": false, + "pass_rate": 0.453125, + "rate": 32.0, + "ttft_p95_ms": 2395.158730843253 + }, + { + "feasible": false, + "pass_rate": 0.328125, + "rate": 64.0, + "ttft_p95_ms": 3272.153523650326 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 636.4131651102598 + }, + { + "feasible": false, + "pass_rate": 0.453125, + "rate": 32.0, + "ttft_p95_ms": 2395.158730843253 + }, + { + "feasible": false, + "pass_rate": 0.328125, + "rate": 64.0, + "ttft_p95_ms": 3272.153523650326 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 636.4131651102598 + }, + { + "feasible": false, + "pass_rate": 0.453125, + "rate": 32.0, + "ttft_p95_ms": 2395.158730843253 + }, + { + "feasible": false, + "pass_rate": 0.328125, + "rate": 64.0, + "ttft_p95_ms": 3272.153523650326 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 120.88989089722979 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 636.4131651102598 + }, + { + "feasible": false, + "pass_rate": 0.453125, + "rate": 32.0, + "ttft_p95_ms": 2395.158730843253 + }, + { + "feasible": false, + "pass_rate": 0.328125, + "rate": 64.0, + "ttft_p95_ms": 3272.153523650326 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 247.38770725677827 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 32.0, + "ttft_p95_ms": 1141.943009455039 + }, + { + "feasible": false, + "pass_rate": 0.546875, + "rate": 64.0, + "ttft_p95_ms": 2030.0951199619265 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 247.38770725677827 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 32.0, + "ttft_p95_ms": 1141.943009455039 + }, + { + "feasible": false, + "pass_rate": 0.546875, + "rate": 64.0, + "ttft_p95_ms": 2030.0951199619265 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 247.38770725677827 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 32.0, + "ttft_p95_ms": 1141.943009455039 + }, + { + "feasible": false, + "pass_rate": 0.546875, + "rate": 64.0, + "ttft_p95_ms": 2030.0951199619265 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 87.63426896254956 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 247.38770725677827 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 32.0, + "ttft_p95_ms": 1141.943009455039 + }, + { + "feasible": false, + "pass_rate": 0.546875, + "rate": 64.0, + "ttft_p95_ms": 2030.0951199619265 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + } + }, + "simulator_sources": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/simulator-a2/frontier_surface_frozen.json", + "sha256": "1aa84bcabe810467c0a5c6b7f848d94bb4f14c725ab214b50e0ae8c8efe6656f" + } + ] +} diff --git a/runs/frontier-fidelity-envelope-v1/results/a2/frontier_surface_frozen.json b/runs/frontier-fidelity-envelope-v1/results/a2/frontier_surface_frozen.json new file mode 100644 index 0000000..741a63b --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a2/frontier_surface_frozen.json @@ -0,0 +1,1732 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + } + ], + "collective": { + "allreduce_csv": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv", + "allreduce_csv_sha256": "9d693fd406616b599e57bcde66c980c7fc2831b3acf37d3eb633cec80ea0070d", + "backend": "vidur" + }, + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 159.88838052749634, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.092494249343872, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.190775394439697, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 8.99299168586731, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 8.941024541854858, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.843815326690674, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.990388631820679, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.942370891571045, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.891057252883911, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.989603757858276, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 8.939709424972534, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 8.94373345375061, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 8.992903709411621, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.04556393623352, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 8.937517881393433, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.944840908050537, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.943378686904907, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.993504762649536, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.916944980621338, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.940248250961304, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 156.6941306591034, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e1c3e7924869b2b30d00d9ed7c8f5261fea58b8bae09af2222f15b909a6e29b0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032540105813934, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 8.991374015808105, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "14089aa0fe7d62dd18d486b0ab2bf02fc8bc11de772da9dd42b09ebca2e484c6", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00411222181281, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.145017147064209, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "fd82e6bd7c47563866485bb023b11fdec286c8688c09c6f00fbf75aecb846897", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 14.2457389654175, + "ttft_max_ms": 694.7908916608797, + "ttft_p50_ms": 461.27998545840177, + "ttft_p95_ms": 636.4131651102598 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.096367359161377, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "2bdd946e6b94ecda68aefa93b07566db206b6b48cf1d26b251e01ff3b6b20354", + "score": { + "feasible": false, + "pass_rate": 0.453125, + "passed": 29, + "throughput_requests_per_second": 14.66575126735821, + "ttft_max_ms": 2488.908730843253, + "ttft_p50_ms": 1333.2347933749338, + "ttft_p95_ms": 2395.158730843253 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.091314792633057, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "bf39a299a3079ec0bafe4ce77b632d793000ce77f34a65f8766c04a080e0a328", + "score": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "throughput_requests_per_second": 14.665751267358214, + "ttft_max_ms": 3410.783730843252, + "ttft_p50_ms": 1807.813828365548, + "ttft_p95_ms": 3272.153523650326 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 8.994110584259033, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e1c3e7924869b2b30d00d9ed7c8f5261fea58b8bae09af2222f15b909a6e29b0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032540105813934, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.044063806533813, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "14089aa0fe7d62dd18d486b0ab2bf02fc8bc11de772da9dd42b09ebca2e484c6", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00411222181281, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.093687534332275, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "fd82e6bd7c47563866485bb023b11fdec286c8688c09c6f00fbf75aecb846897", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 14.2457389654175, + "ttft_max_ms": 694.7908916608797, + "ttft_p50_ms": 461.27998545840177, + "ttft_p95_ms": 636.4131651102598 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.091583013534546, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "2bdd946e6b94ecda68aefa93b07566db206b6b48cf1d26b251e01ff3b6b20354", + "score": { + "feasible": false, + "pass_rate": 0.453125, + "passed": 29, + "throughput_requests_per_second": 14.66575126735821, + "ttft_max_ms": 2488.908730843253, + "ttft_p50_ms": 1333.2347933749338, + "ttft_p95_ms": 2395.158730843253 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.093712568283081, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "bf39a299a3079ec0bafe4ce77b632d793000ce77f34a65f8766c04a080e0a328", + "score": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "throughput_requests_per_second": 14.665751267358214, + "ttft_max_ms": 3410.783730843252, + "ttft_p50_ms": 1807.813828365548, + "ttft_p95_ms": 3272.153523650326 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.043614625930786, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e1c3e7924869b2b30d00d9ed7c8f5261fea58b8bae09af2222f15b909a6e29b0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032540105813934, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 8.999051809310913, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "14089aa0fe7d62dd18d486b0ab2bf02fc8bc11de772da9dd42b09ebca2e484c6", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00411222181281, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.092182636260986, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "fd82e6bd7c47563866485bb023b11fdec286c8688c09c6f00fbf75aecb846897", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 14.2457389654175, + "ttft_max_ms": 694.7908916608797, + "ttft_p50_ms": 461.27998545840177, + "ttft_p95_ms": 636.4131651102598 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.042077779769897, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "2bdd946e6b94ecda68aefa93b07566db206b6b48cf1d26b251e01ff3b6b20354", + "score": { + "feasible": false, + "pass_rate": 0.453125, + "passed": 29, + "throughput_requests_per_second": 14.66575126735821, + "ttft_max_ms": 2488.908730843253, + "ttft_p50_ms": 1333.2347933749338, + "ttft_p95_ms": 2395.158730843253 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.09327220916748, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "bf39a299a3079ec0bafe4ce77b632d793000ce77f34a65f8766c04a080e0a328", + "score": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "throughput_requests_per_second": 14.665751267358214, + "ttft_max_ms": 3410.783730843252, + "ttft_p50_ms": 1807.813828365548, + "ttft_p95_ms": 3272.153523650326 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 8.941684007644653, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e1c3e7924869b2b30d00d9ed7c8f5261fea58b8bae09af2222f15b909a6e29b0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032540105813934, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 8.992353916168213, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "14089aa0fe7d62dd18d486b0ab2bf02fc8bc11de772da9dd42b09ebca2e484c6", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00411222181281, + "ttft_max_ms": 120.88989089722979, + "ttft_p50_ms": 120.88989089722979, + "ttft_p95_ms": 120.88989089722979 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.146737098693848, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "fd82e6bd7c47563866485bb023b11fdec286c8688c09c6f00fbf75aecb846897", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 14.2457389654175, + "ttft_max_ms": 694.7908916608797, + "ttft_p50_ms": 461.27998545840177, + "ttft_p95_ms": 636.4131651102598 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.090755462646484, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "2bdd946e6b94ecda68aefa93b07566db206b6b48cf1d26b251e01ff3b6b20354", + "score": { + "feasible": false, + "pass_rate": 0.453125, + "passed": 29, + "throughput_requests_per_second": 14.66575126735821, + "ttft_max_ms": 2488.908730843253, + "ttft_p50_ms": 1333.2347933749338, + "ttft_p95_ms": 2395.158730843253 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.190985918045044, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "bf39a299a3079ec0bafe4ce77b632d793000ce77f34a65f8766c04a080e0a328", + "score": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "throughput_requests_per_second": 14.665751267358214, + "ttft_max_ms": 3410.783730843252, + "ttft_p50_ms": 1807.813828365548, + "ttft_p95_ms": 3272.153523650326 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 141.65436458587646, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "10a656450567caf0b8b628600b6e6acf8eeeb3f37c5d82713f12966b5aa9e533", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041007571782521, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.093120813369751, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "519248e66e8451d0f7bd61566a67453e862caf3b7cbfe466f83d78388f4997c4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.037541074750196, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.287490606307983, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "29cfcc459ed774299f781ade2234d11aa9c62cf5f5bb952df4b7e9c7afe7f2b4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.584885703455418, + "ttft_max_ms": 274.25967286812414, + "ttft_p50_ms": 187.36059674989036, + "ttft_p95_ms": 247.38770725677827 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.1891348361969, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "b51aecf0f91ce9d2c55d9d6b854c8a6c09c290133a65733e5cb436abd375555d", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 20.409287348343973, + "ttft_max_ms": 1204.443009455039, + "ttft_p50_ms": 693.4638889220377, + "ttft_p95_ms": 1141.943009455039 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.092704772949219, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "174484e0f73c517ada2adce7dea0c4737f89b6252eab700d80c0c4055493418c", + "score": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "throughput_requests_per_second": 20.862900609687326, + "ttft_max_ms": 2114.521306587048, + "ttft_p50_ms": 1129.8633894955506, + "ttft_p95_ms": 2030.0951199619265 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 8.94224739074707, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "10a656450567caf0b8b628600b6e6acf8eeeb3f37c5d82713f12966b5aa9e533", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041007571782521, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 8.993168115615845, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "519248e66e8451d0f7bd61566a67453e862caf3b7cbfe466f83d78388f4997c4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.037541074750196, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.14280891418457, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "29cfcc459ed774299f781ade2234d11aa9c62cf5f5bb952df4b7e9c7afe7f2b4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.584885703455418, + "ttft_max_ms": 274.25967286812414, + "ttft_p50_ms": 187.36059674989036, + "ttft_p95_ms": 247.38770725677827 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.090214967727661, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "b51aecf0f91ce9d2c55d9d6b854c8a6c09c290133a65733e5cb436abd375555d", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 20.409287348343973, + "ttft_max_ms": 1204.443009455039, + "ttft_p50_ms": 693.4638889220377, + "ttft_p95_ms": 1141.943009455039 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.094237804412842, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "174484e0f73c517ada2adce7dea0c4737f89b6252eab700d80c0c4055493418c", + "score": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "throughput_requests_per_second": 20.862900609687326, + "ttft_max_ms": 2114.521306587048, + "ttft_p50_ms": 1129.8633894955506, + "ttft_p95_ms": 2030.0951199619265 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.140523672103882, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "10a656450567caf0b8b628600b6e6acf8eeeb3f37c5d82713f12966b5aa9e533", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041007571782521, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.091987609863281, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "519248e66e8451d0f7bd61566a67453e862caf3b7cbfe466f83d78388f4997c4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.037541074750196, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.141303777694702, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "29cfcc459ed774299f781ade2234d11aa9c62cf5f5bb952df4b7e9c7afe7f2b4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.584885703455418, + "ttft_max_ms": 274.25967286812414, + "ttft_p50_ms": 187.36059674989036, + "ttft_p95_ms": 247.38770725677827 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.240424156188965, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "b51aecf0f91ce9d2c55d9d6b854c8a6c09c290133a65733e5cb436abd375555d", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 20.409287348343973, + "ttft_max_ms": 1204.443009455039, + "ttft_p50_ms": 693.4638889220377, + "ttft_p95_ms": 1141.943009455039 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.140634536743164, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "174484e0f73c517ada2adce7dea0c4737f89b6252eab700d80c0c4055493418c", + "score": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "throughput_requests_per_second": 20.862900609687326, + "ttft_max_ms": 2114.521306587048, + "ttft_p50_ms": 1129.8633894955506, + "ttft_p95_ms": 2030.0951199619265 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.143561601638794, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "10a656450567caf0b8b628600b6e6acf8eeeb3f37c5d82713f12966b5aa9e533", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041007571782521, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.091549396514893, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "519248e66e8451d0f7bd61566a67453e862caf3b7cbfe466f83d78388f4997c4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.037541074750196, + "ttft_max_ms": 87.63426896254956, + "ttft_p50_ms": 87.63426896254956, + "ttft_p95_ms": 87.63426896254956 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.191416263580322, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "29cfcc459ed774299f781ade2234d11aa9c62cf5f5bb952df4b7e9c7afe7f2b4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.584885703455418, + "ttft_max_ms": 274.25967286812414, + "ttft_p50_ms": 187.36059674989036, + "ttft_p95_ms": 247.38770725677827 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.043817520141602, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "b51aecf0f91ce9d2c55d9d6b854c8a6c09c290133a65733e5cb436abd375555d", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 20.409287348343973, + "ttft_max_ms": 1204.443009455039, + "ttft_p50_ms": 693.4638889220377, + "ttft_p95_ms": 1141.943009455039 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.046180486679077, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "174484e0f73c517ada2adce7dea0c4737f89b6252eab700d80c0c4055493418c", + "score": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "throughput_requests_per_second": 20.862900609687326, + "ttft_max_ms": 2114.521306587048, + "ttft_p50_ms": 1129.8633894955506, + "ttft_p95_ms": 2030.0951199619265 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 4.0, + 8.0, + 16.0, + 32.0, + 64.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/cc_backend/backends/vidur_cc_backend.py\n M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n?? tests/unit/test_vidur_cc_large_payload.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "frozen_before_real" +} diff --git a/runs/frontier-fidelity-envelope-v1/results/a2/qwen30-prefill-ranking.png b/runs/frontier-fidelity-envelope-v1/results/a2/qwen30-prefill-ranking.png new file mode 100644 index 0000000..ed8a221 Binary files /dev/null and b/runs/frontier-fidelity-envelope-v1/results/a2/qwen30-prefill-ranking.png differ diff --git a/runs/frontier-fidelity-envelope-v1/results/a3/capacity.csv b/runs/frontier-fidelity-envelope-v1/results/a3/capacity.csv new file mode 100644 index 0000000..da354c9 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a3/capacity.csv @@ -0,0 +1,13 @@ +config,tp,mns,real,simulator +tp1_mns8,1,8,7.0,8.0 +tp1_mns16,1,16,7.0,8.0 +tp1_mns32,1,32,7.0,8.0 +tp1_mns64,1,64,7.0,8.0 +tp2_mns8,2,8,7.0,8.0 +tp2_mns16,2,16,7.0,8.0 +tp2_mns32,2,32,7.0,8.0 +tp2_mns64,2,64,7.0,8.0 +tp4_mns8,4,8,8.0,8.0 +tp4_mns16,4,16,8.0,8.0 +tp4_mns32,4,32,8.0,8.0 +tp4_mns64,4,64,8.0,8.0 diff --git a/runs/frontier-fidelity-envelope-v1/results/a3/comparison.json b/runs/frontier-fidelity-envelope-v1/results/a3/comparison.json new file mode 100644 index 0000000..b011206 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a3/comparison.json @@ -0,0 +1,5638 @@ +{ + "contract": { + "input_tokens": 2048, + "model": "Qwen3-30B-A3B", + "output_tokens": 1, + "prefix_caching": false, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "metrics": { + "anchor_confusion": { + "real_fail_sim_fail": 24, + "real_fail_sim_pass": 8, + "real_pass_sim_fail": 0, + "real_pass_sim_pass": 28 + }, + "anchor_grid_coverage": { + "real_only": 36, + "shared": 60, + "simulator_only": 0 + }, + "config_order": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "kendall": { + "both_ties": 34, + "concordant": 0, + "discordant": 0, + "kendall_tau_b": null, + "real_only_ties": 0, + "simulator_only_ties": 32 + }, + "pairwise_non_tied": { + "all": { + "accuracy": 0.0, + "comparable": 32, + "correct": 0 + }, + "within_tp": {} + }, + "real_best_capacity_per_gpu": 8.0, + "real_top_set": [ + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "simulator_best_capacity_per_gpu": 8.0, + "simulator_top_set": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "top1_regret_best": 0.0, + "top1_regret_worst": 0.125, + "top_set_exact_match": false, + "top_set_overlap": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64", + "tp4_mns8" + ] + }, + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "real": { + "tp1_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.15452707000077, + 152.64549700077623 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r4p00.json", + "sha256": "d82e386ec3effdf57c148171311928298405da0e4f896584bf05db2485bb6bd4", + "summary": { + "admission_lag_max_ms": 0.11588609777390957, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.6234939750284, + "ttft_p50_ms": 151.05855697765946, + "ttft_p95_ms": 153.15452707000077 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r4p00.json", + "sha256": "1565be89131e3c8a9c352c6f29694ef2793734b0623a2db1cbea7e0b0e8a2bf6", + "summary": { + "admission_lag_max_ms": 0.11451996397227049, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.18561100866646, + "ttft_p50_ms": 151.18564700242132, + "ttft_p95_ms": 152.64549700077623 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.8308589477092, + 152.43642206769437 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r5p00.json", + "sha256": "e4447d94a326c541412b40157cfe283a6350680e3285b5d10fed6b8de5c0cd93", + "summary": { + "admission_lag_max_ms": 0.17109804321080446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.35138496104628, + "ttft_p50_ms": 150.91424097772688, + "ttft_p95_ms": 152.8308589477092 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r5p00.json", + "sha256": "572c17e309c6c9f23c490979e7acb885475f142bc530ecfee88916b048259ea2", + "summary": { + "admission_lag_max_ms": 0.1321269664913416, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.11930498946458, + "ttft_p50_ms": 150.50161199178547, + "ttft_p95_ms": 152.43642206769437 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.56597595382482, + 152.16859406791627 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r6p00.json", + "sha256": "68c8b39154d46f1ceda17f47e8cbd84a8c8a9f088c5e82d8f0ac25a7180dc14c", + "summary": { + "admission_lag_max_ms": 0.11641765013337135, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.50876995641738, + "ttft_p50_ms": 150.41159105021507, + "ttft_p95_ms": 153.56597595382482 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r6p00.json", + "sha256": "a970a2cdf7034026a56af3ea4da59856fd588da35d66ef4f9e15481a892edcc9", + "summary": { + "admission_lag_max_ms": 0.11551892384886742, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 168.65782206878066, + "ttft_p50_ms": 150.03317291848361, + "ttft_p95_ms": 152.16859406791627 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 300.90478900820017, + 293.0445579113439 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r7p00.json", + "sha256": "1aa74c18adbff53b36ae8a6f07b5b0865f6b68fb3e7862a70b36f67c18f0bba0", + "summary": { + "admission_lag_max_ms": 0.10505830869078636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.15459000505507, + "ttft_p50_ms": 263.44375393819064, + "ttft_p95_ms": 300.90478900820017 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r7p00.json", + "sha256": "e14bc9dba35d453862839eeed77262a1ed9b8394404ca90c60ec8cb26a1aa73b", + "summary": { + "admission_lag_max_ms": 0.10963028762489557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 295.1545650139451, + "ttft_p50_ms": 243.97043196950108, + "ttft_p95_ms": 293.0445579113439 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1323.4397460473701, + 1315.5151740647852 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r8p00.json", + "sha256": "9aeee7453ffe5098d203e02b84f5009ed347c478b736e89297bd3695dd81d477", + "summary": { + "admission_lag_max_ms": 0.1251589274033904, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1477.699184906669, + "ttft_p50_ms": 776.2576789828017, + "ttft_p95_ms": 1323.4397460473701 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r8p00.json", + "sha256": "7630f4901cf4ec4ea955738b119a1c32aba53f9ca1edb8819556fc60ae1fda09", + "summary": { + "admission_lag_max_ms": 0.11316593736410141, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1404.566578916274, + "ttft_p50_ms": 761.7035870207474, + "ttft_p95_ms": 1315.5151740647852 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4892.253074911423, + 4893.758205929771 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r16p00.json", + "sha256": "d3a2a85cf06a4f080edc9a3809f05493ac55b0aa1d890806ffbc551c8eca14f8", + "summary": { + "admission_lag_max_ms": 0.39830803871154785, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.215195022523, + "ttft_p50_ms": 2742.7029330283403, + "ttft_p95_ms": 4892.253074911423 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r16p00.json", + "sha256": "67110284c38a9a60228f00b1d2b0de9477d6dec660f1c91eb4f39e1b79e272a5", + "summary": { + "admission_lag_max_ms": 0.11306500528007746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.818040987477, + "ttft_p50_ms": 2742.512951954268, + "ttft_p95_ms": 4893.758205929771 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6650.756464921869, + 6751.845097984187 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r32p00.json", + "sha256": "affc62b27382400497bb82e9da31ed9891b60155318fc4f250dca0704ccc5e02", + "summary": { + "admission_lag_max_ms": 0.11010200250893831, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6920.621327939443, + "ttft_p50_ms": 3650.839229929261, + "ttft_p95_ms": 6650.756464921869 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r32p00.json", + "sha256": "df0c8d703b023692814be41f47da4db4ce1b7eb0c3423a742c10884b6447583f", + "summary": { + "admission_lag_max_ms": 0.1288279891014099, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6897.979240980931, + "ttft_p50_ms": 3730.34580796957, + "ttft_p95_ms": 6751.845097984187 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7732.2855240199715, + 7744.056030060165 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r64p00.json", + "sha256": "fbfad78129c843c85905c3f94f278f468a0a019c0cdf0b62140d0fea8f0aaffc", + "summary": { + "admission_lag_max_ms": 0.11090096086263657, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7836.245508980937, + "ttft_p50_ms": 4328.587644966319, + "ttft_p95_ms": 7732.2855240199715 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r64p00.json", + "sha256": "0f005c642327708e1ceb058f71f806b4d46ea666f45695605552e8f234521608", + "summary": { + "admission_lag_max_ms": 0.10862003546208143, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7847.871358972043, + "ttft_p50_ms": 4332.9262170009315, + "ttft_p95_ms": 7744.056030060165 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp1_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.44141807872802, + 152.2580359596759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r4p00.json", + "sha256": "211fa276e4a99a061126882085b9850549b453520ac3414dd2d332c89e52b431", + "summary": { + "admission_lag_max_ms": 0.09491504170000553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 169.56262104213238, + "ttft_p50_ms": 150.8563900133595, + "ttft_p95_ms": 153.44141807872802 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r4p00.json", + "sha256": "17428902a5b20e9084f20792963f0620a1af06cdc1afb76ee99fd15dd905b89e", + "summary": { + "admission_lag_max_ms": 0.12033001985400915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.99700601119548, + "ttft_p50_ms": 150.5822929320857, + "ttft_p95_ms": 152.2580359596759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.05012692604214, + 152.00478909537196 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r5p00.json", + "sha256": "b0eb19ff7ab8a26e34e91f213a983e58689c63071285e27b3719bda0a765ff32", + "summary": { + "admission_lag_max_ms": 0.11265103239566088, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.48999004345387, + "ttft_p50_ms": 150.21017810795456, + "ttft_p95_ms": 152.05012692604214 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r5p00.json", + "sha256": "3b857dff42dd16ebee95356ceacd6b58ff7bd0a9b91f3d6208d489b84985697f", + "summary": { + "admission_lag_max_ms": 0.11149293277412653, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.33813393954188, + "ttft_p50_ms": 150.45341395307332, + "ttft_p95_ms": 152.00478909537196 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.74227899499238, + 152.82704099081457 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r6p00.json", + "sha256": "7615fa664a99f02f9f6d4f1fb0f2d6cba11842772c086634b939bc08cd0f2f4d", + "summary": { + "admission_lag_max_ms": 0.1096969936043024, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.2669020127505, + "ttft_p50_ms": 150.6725640501827, + "ttft_p95_ms": 152.74227899499238 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r6p00.json", + "sha256": "345e3cd3f8cd4d7c7a5e9343f937dd9b1b5928500133830f3046b743da6df64e", + "summary": { + "admission_lag_max_ms": 0.12192933354526758, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.12304301187396, + "ttft_p50_ms": 150.68741200957447, + "ttft_p95_ms": 152.82704099081457 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 302.5522669777274, + 294.5978350471705 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r7p00.json", + "sha256": "3f54a3e3f2279a5325a173ff714d600fb1a2cdfcfba1e8e5c980f7b21b99be59", + "summary": { + "admission_lag_max_ms": 0.11122168507426977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.24335307907313, + "ttft_p50_ms": 262.4604320153594, + "ttft_p95_ms": 302.5522669777274 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r7p00.json", + "sha256": "0832169cdf1bef49c5006464c6cbcd50f4567e0ab90bec98d722f64f7b476763", + "summary": { + "admission_lag_max_ms": 0.1059951027855277, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 297.1431679325178, + "ttft_p50_ms": 249.51713997870684, + "ttft_p95_ms": 294.5978350471705 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1321.9368209829554, + 1318.7565059633926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r8p00.json", + "sha256": "a0f5900ac9d2c47dcc7d7ac08d3f3d23e1d6f529fdd0200989afa819dd3df055", + "summary": { + "admission_lag_max_ms": 0.17414195463061333, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1475.8616799954325, + "ttft_p50_ms": 775.7647309917957, + "ttft_p95_ms": 1321.9368209829554 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r8p00.json", + "sha256": "c0fc3c85095ddacc66849e6f58be0851abab6d4022ee4eb7120c8cb8ea9e5d2c", + "summary": { + "admission_lag_max_ms": 2.749277977272868, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1407.0930189918727, + "ttft_p50_ms": 763.8335369993001, + "ttft_p95_ms": 1318.7565059633926 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4893.3289990527555, + 4893.17379205022 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r16p00.json", + "sha256": "4e9ccaf5384f622b1da7c58d3de803c86099dd9deae91e363aabccc14aeabc05", + "summary": { + "admission_lag_max_ms": 0.11893699411302805, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.85822511185, + "ttft_p50_ms": 2742.8449960425496, + "ttft_p95_ms": 4893.3289990527555 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r16p00.json", + "sha256": "c207e415186aed1148e16085145bcdfd189314e7b92f74ea72fb2e4972ef4eae", + "summary": { + "admission_lag_max_ms": 0.09541702456772327, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.2590809213, + "ttft_p50_ms": 2740.232800017111, + "ttft_p95_ms": 4893.17379205022 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6653.632597066462, + 6647.155530983582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r32p00.json", + "sha256": "780f563a2739d892761905031f82ca8bd0ac4f4eb485c454390dcb2bcdfc1b81", + "summary": { + "admission_lag_max_ms": 0.105265062302351, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6922.717701992951, + "ttft_p50_ms": 3652.452490059659, + "ttft_p95_ms": 6653.632597066462 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r32p00.json", + "sha256": "2a5eb8fb37f03e759de149d24f9b23c812a26afe72ffaaffa030688904634e43", + "summary": { + "admission_lag_max_ms": 0.10375294368714094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6917.111590038985, + "ttft_p50_ms": 3649.0042679943144, + "ttft_p95_ms": 6647.155530983582 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7747.191094094887, + 7839.257381972857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r64p00.json", + "sha256": "b94a39a91842ab39d545c473fc4f872ec79eb1888943c3e03ab03c18961a9105", + "summary": { + "admission_lag_max_ms": 0.6591080455109477, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7851.370444986969, + "ttft_p50_ms": 4335.578468977474, + "ttft_p95_ms": 7747.191094094887 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r64p00.json", + "sha256": "69ab47305c4bb14f3b4ed5bc0b18a37b8a5e7c1cdbee9232e6d2b9616ed10f84", + "summary": { + "admission_lag_max_ms": 0.10753795504570007, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.125, + "passed": 8, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7885.349316056818, + "ttft_p50_ms": 4009.1016669757664, + "ttft_p95_ms": 7839.257381972857 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp1_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 156.76817402709275, + 152.9719429090619 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r4p00.json", + "sha256": "df798dd7a1cde3cfaf698286f44bbac2b0730a2c806f125268bdf7226e2bb47c", + "summary": { + "admission_lag_max_ms": 0.1147780567407608, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 162.9874319769442, + "ttft_p50_ms": 152.67050196416676, + "ttft_p95_ms": 156.76817402709275 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r4p00.json", + "sha256": "df80057224d69ce2a79f937aa45c2093d66d131765f3c7bfcc2289a9aaff3d6a", + "summary": { + "admission_lag_max_ms": 0.2131829969584942, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.917470946908, + "ttft_p50_ms": 150.96506499685347, + "ttft_p95_ms": 152.9719429090619 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.4973110537976, + 155.23677493911237 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r5p00.json", + "sha256": "e614ffaf879859407fc6420847f0140a81bda9a94fcc438fb6782c9019bf5378", + "summary": { + "admission_lag_max_ms": 0.1283780438825488, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.08639602921903, + "ttft_p50_ms": 152.1085740532726, + "ttft_p95_ms": 153.4973110537976 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r5p00.json", + "sha256": "13e825eb5ced3d0cb2b5dd68d9243fdf1cee8779b90e9aad77be04053be22be2", + "summary": { + "admission_lag_max_ms": 0.1683390000835061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.6691420264542, + "ttft_p50_ms": 152.42570801638067, + "ttft_p95_ms": 155.23677493911237 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.53963600844145, + 152.85827894695103 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r6p00.json", + "sha256": "ac530f98f57ec452538ebe2311291d0a7c83bbd5554e82f53e78a2311802e411", + "summary": { + "admission_lag_max_ms": 0.13490929268300533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.18225098680705, + "ttft_p50_ms": 150.29372205026448, + "ttft_p95_ms": 152.53963600844145 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r6p00.json", + "sha256": "2f1445ee0890b0972252d10180ef6f00cbad45ee73bd174f21b370cdf950b654", + "summary": { + "admission_lag_max_ms": 0.16514293383806944, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.88926689513028, + "ttft_p50_ms": 150.2534201135859, + "ttft_p95_ms": 152.85827894695103 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 346.96662798523903, + 295.1050230767578 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r7p00.json", + "sha256": "c4765794d442daf7eeb64a2f587792701ebe7659c7702a3bbbfc8900801fff0e", + "summary": { + "admission_lag_max_ms": 0.16206549480557442, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 354.881492909044, + "ttft_p50_ms": 310.15722593292594, + "ttft_p95_ms": 346.96662798523903 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r7p00.json", + "sha256": "30addb463839e23a7356a5cf6d4060ec1eebfa2b9564fa9500be5bcb7aecb537", + "summary": { + "admission_lag_max_ms": 0.10446517262607813, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 296.6241450048983, + "ttft_p50_ms": 246.06937903445214, + "ttft_p95_ms": 295.1050230767578 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1376.107804942876, + 1321.1636180058122 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r8p00.json", + "sha256": "89e2975f9c7319e477789ea77015541d5235efa61d90fb8bd5b0420a58f7f186", + "summary": { + "admission_lag_max_ms": 0.1416490413248539, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.890625, + "passed": 57, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1501.330335973762, + "ttft_p50_ms": 802.2130889585242, + "ttft_p95_ms": 1376.107804942876 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r8p00.json", + "sha256": "9ca6935f1320033c3ec4b72a35c5d9909ccd7b0fff2e4b2c92c6a26f52887df1", + "summary": { + "admission_lag_max_ms": 0.12422900181263685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1410.047525074333, + "ttft_p50_ms": 766.8001820566133, + "ttft_p95_ms": 1321.1636180058122 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4950.155022088438, + 4892.750842031091 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r16p00.json", + "sha256": "4fce2749435bf2e8c77999e11be0a788cd85f459a647760f3db34f1dbc0aee85", + "summary": { + "admission_lag_max_ms": 0.10520406067371368, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.171875, + "passed": 11, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 5031.207721913233, + "ttft_p50_ms": 2775.8131299633533, + "ttft_p95_ms": 4950.155022088438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r16p00.json", + "sha256": "33ded0192d8986dbcf0a7f5ab6329a440994243d4e528d6c1b278bb6b92720fd", + "summary": { + "admission_lag_max_ms": 0.14005706179887056, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.664474112913, + "ttft_p50_ms": 2743.070787983015, + "ttft_p95_ms": 4892.750842031091 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6750.775643973611, + 6754.7571890754625 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r32p00.json", + "sha256": "44030c4d3374cecebe67b886b428ba8014c87d941c00a178d6049bc57c809f46", + "summary": { + "admission_lag_max_ms": 0.7358170114457607, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6899.517056066543, + "ttft_p50_ms": 3728.6347220651805, + "ttft_p95_ms": 6750.775643973611 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r32p00.json", + "sha256": "f397bad36552c6df5bf5bcee702cfa968ddf3e20b20e9976554ddfbe838ef3b2", + "summary": { + "admission_lag_max_ms": 0.7476879982277751, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6901.606322033331, + "ttft_p50_ms": 3729.3128239689395, + "ttft_p95_ms": 6754.7571890754625 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7800.616211956367, + 7748.703246936202 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r64p00.json", + "sha256": "61eadc2af60386d6972d69fe7b083874e705864162073d54d1bde2bc0e6de7e7", + "summary": { + "admission_lag_max_ms": 0.11387700214982033, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7905.77321103774, + "ttft_p50_ms": 4366.223473916762, + "ttft_p95_ms": 7800.616211956367 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r64p00.json", + "sha256": "db02e9a814154a9259cbe3fefdf15700923458f7e1208596e861b69b6a6ddd41", + "summary": { + "admission_lag_max_ms": 0.14900392852723598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7852.748095989227, + "ttft_p50_ms": 4337.09563605953, + "ttft_p95_ms": 7748.703246936202 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp1_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.07429495826364, + 153.62840006127954 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r4p00.json", + "sha256": "23d5ab491b70c02f51c2f06dec0407edc0627819a33b9bcef1349d974a4e20fe", + "summary": { + "admission_lag_max_ms": 0.14362391084432602, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 171.46877304185182, + "ttft_p50_ms": 151.62539994344115, + "ttft_p95_ms": 154.07429495826364 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r4p00.json", + "sha256": "563aa230e0f5e420cf453dfd9b829d7df493919192d638cbe76f85eb77b6d2e7", + "summary": { + "admission_lag_max_ms": 0.11323392391204834, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.62100906390697, + "ttft_p50_ms": 151.69239102397114, + "ttft_p95_ms": 153.62840006127954 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.20477103907615, + 153.81942794192582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r5p00.json", + "sha256": "6f2a04a5c11eed5826e2a6e1ce6e6f8d98504b31d0fec30d9440df2cdb62e56d", + "summary": { + "admission_lag_max_ms": 0.10794296395033598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.37187200970948, + "ttft_p50_ms": 152.59731002151966, + "ttft_p95_ms": 154.20477103907615 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r5p00.json", + "sha256": "27a0bfa65a66e5cc77e10a5cf39a846f0fd1dd266c10926dae9db7a80c581d6e", + "summary": { + "admission_lag_max_ms": 0.1204309519380331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.40366806089878, + "ttft_p50_ms": 152.0398510619998, + "ttft_p95_ms": 153.81942794192582 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.00951700191945, + 153.5524509381503 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r6p00.json", + "sha256": "4562fae05d510cd50f2c9d7479916d056c8ac4d04851d2388b98f8410192c23a", + "summary": { + "admission_lag_max_ms": 0.10347133502364159, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.3970319693908, + "ttft_p50_ms": 150.414744974114, + "ttft_p95_ms": 153.00951700191945 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r6p00.json", + "sha256": "bd13283e2807150396b278f46ff630ae9820895e0c054a9efcfd8fc213947d47", + "summary": { + "admission_lag_max_ms": 0.13600767124444246, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 159.63539096992463, + "ttft_p50_ms": 151.83425601571798, + "ttft_p95_ms": 153.5524509381503 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 336.8800369789824, + 329.65395506471395 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r7p00.json", + "sha256": "fd5afe52238d6ddb72bdf7f7b142fddbeb3f62e160e4a1b3a639b308717964e6", + "summary": { + "admission_lag_max_ms": 0.14346837997436523, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 341.1952300230041, + "ttft_p50_ms": 294.81438896618783, + "ttft_p95_ms": 336.8800369789824 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r7p00.json", + "sha256": "b0fdc1ef617313261c490010adeb2f4ee93bbb053f1db10734ed5a63fd37fa76", + "summary": { + "admission_lag_max_ms": 0.11047185398638248, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 334.06115300022066, + "ttft_p50_ms": 281.1476809438318, + "ttft_p95_ms": 329.65395506471395 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + true + ], + "round_ttft_p95_ms": [ + 1289.2397560644895, + 1254.356418037787 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r8p00.json", + "sha256": "8877ed44b90f421afd22ec110fe6337581c67cd973b44378bcb7528fc8624de3", + "summary": { + "admission_lag_max_ms": 0.09899691212922335, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.9375, + "passed": 60, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1454.3529660440981, + "ttft_p50_ms": 813.0663179326802, + "ttft_p95_ms": 1289.2397560644895 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r8p00.json", + "sha256": "7ab24f5ab33bf166589af7e2f5c2a274ceef6a64f0029a96ec1879b42332123d", + "summary": { + "admission_lag_max_ms": 0.11694896966218948, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 0.953125, + "passed": 61, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1347.0055809011683, + "ttft_p50_ms": 744.0344559727237, + "ttft_p95_ms": 1254.356418037787 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4903.538785991259, + 4900.28300601989 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r16p00.json", + "sha256": "363d9b43164b5451bc3dd668caa0ce141d15114087cf444e7fd79d12b7563fd6", + "summary": { + "admission_lag_max_ms": 0.11420203372836113, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4982.48829995282, + "ttft_p50_ms": 2749.9175920384005, + "ttft_p95_ms": 4903.538785991259 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r16p00.json", + "sha256": "8aec70dfd6b69f4903bccc02c5d5474165ed07d9dd010c860ff9b447429da950", + "summary": { + "admission_lag_max_ms": 0.19414699636399746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4979.007771937177, + "ttft_p50_ms": 2749.115635990165, + "ttft_p95_ms": 4900.28300601989 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6760.187636013143, + 6534.733187989332 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r32p00.json", + "sha256": "7684ac0d2d6f851218b88fb37db3a5b54d877a6f1ead3bb604a6cb4d9a61cf3f", + "summary": { + "admission_lag_max_ms": 0.43534801807254553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6907.940509961918, + "ttft_p50_ms": 3735.540736000985, + "ttft_p95_ms": 6760.187636013143 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r32p00.json", + "sha256": "947974a9ae04e170b8c1aad32ddbda8c1000f3902a9b572c6338a7d0d4f23146", + "summary": { + "admission_lag_max_ms": 1.221106038428843, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6798.48151106853, + "ttft_p50_ms": 3595.78589303419, + "ttft_p95_ms": 6534.733187989332 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7754.704045015387, + 7756.428897031583 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r64p00.json", + "sha256": "abfc2a158087f0ad75b560dd420604a9f81359ffdf39f0ce8e2ea25b451e6904", + "summary": { + "admission_lag_max_ms": 0.13654399663209915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7857.916672946885, + "ttft_p50_ms": 4341.527819051407, + "ttft_p95_ms": 7754.704045015387 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r64p00.json", + "sha256": "192c0c850a52d2b64780bc6321cbfd10faddcc0ff10741ddad283b7d789b9def", + "summary": { + "admission_lag_max_ms": 0.11000398080796003, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7859.458995983005, + "ttft_p50_ms": 4341.894763987511, + "ttft_p95_ms": 7756.428897031583 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp1_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 99.1211449727416, + 101.08128399588168 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r4p00.json", + "sha256": "f149882da390114f2a2a140a3bd2ca6f15b6f7761591ea7b49b5a961c9d18df5", + "summary": { + "admission_lag_max_ms": 0.10085199028253555, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.28417506348342, + "ttft_p50_ms": 95.90165293775499, + "ttft_p95_ms": 99.1211449727416 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r4p00.json", + "sha256": "154dfaf2729001a8ef5b7430290f4ca906df1d4758e867d5b350489410f36d5c", + "summary": { + "admission_lag_max_ms": 0.11863000690937042, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 105.76263803523034, + "ttft_p50_ms": 96.27402503974736, + "ttft_p95_ms": 101.08128399588168 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.1850419882685, + 97.03529300168157 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r8p00.json", + "sha256": "9338d5cb994b0c12a6420f030ae2a9335a04986b4a81e31f4781ac838aeef2d4", + "summary": { + "admission_lag_max_ms": 0.12352492194622755, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.26434096693993, + "ttft_p50_ms": 95.07436107378453, + "ttft_p95_ms": 97.1850419882685 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r8p00.json", + "sha256": "a35b67970948976482ee835e595ee5631fa921607276d5ea7928460514f0d06b", + "summary": { + "admission_lag_max_ms": 0.11170096695423126, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.72453604172915, + "ttft_p50_ms": 94.03157397173345, + "ttft_p95_ms": 97.03529300168157 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.10817302018404, + 94.37024802900851 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r10p00.json", + "sha256": "f67e0f4f6b41861fcbcbee04b4efbf0dee46e273239ec94ef22d650ba6afa22a", + "summary": { + "admission_lag_max_ms": 0.10012194979935884, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.40214702207595, + "ttft_p50_ms": 94.17163301259279, + "ttft_p95_ms": 96.10817302018404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r10p00.json", + "sha256": "5a64b52f0298e9ecda2fddb8230cc27ab4b9c1db01575f68b5331816b324d1d7", + "summary": { + "admission_lag_max_ms": 0.10465399827808142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.07159405015409, + "ttft_p50_ms": 93.24954205658287, + "ttft_p95_ms": 94.37024802900851 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 447.0475659472868, + 454.8899739747867 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r12p00.json", + "sha256": "c6e27b9078281a54feeb1fab09616385a70de819791ce14f46b001bf540a80ab", + "summary": { + "admission_lag_max_ms": 0.12051803059875965, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 455.56937297806144, + "ttft_p50_ms": 280.09015799034387, + "ttft_p95_ms": 447.0475659472868 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r12p00.json", + "sha256": "861f9c4186c33e9ee6365d19c53e98c10615c5ee529a84e69f84dc292f53ac4f", + "summary": { + "admission_lag_max_ms": 0.11322635691612959, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 457.4526828946546, + "ttft_p50_ms": 290.1579710887745, + "ttft_p95_ms": 454.8899739747867 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1000.5082850111648, + 1004.6314439969137 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r14p00.json", + "sha256": "2dcaf1087edd1f12fbfe0618a6ab636b22484756a5729919bc0e51de153f8f4a", + "summary": { + "admission_lag_max_ms": 0.1881921198219061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1089.2393708927557, + "ttft_p50_ms": 611.2175739835948, + "ttft_p95_ms": 1000.5082850111648 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r14p00.json", + "sha256": "16937b31f72298c7a50ec3c7b9c1b9c798f305d83469b3d3c058efc4d53135da", + "summary": { + "admission_lag_max_ms": 0.10659568943083286, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1096.6680150013417, + "ttft_p50_ms": 617.3943240428343, + "ttft_p95_ms": 1004.6314439969137 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1458.4544260287657, + 1471.3564389385283 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r16p00.json", + "sha256": "5576a7fd06977f51f1c43bb73c3c474cc21a93cd288b422ec3d09db671b3afe7", + "summary": { + "admission_lag_max_ms": 0.135264010168612, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1583.2410319708288, + "ttft_p50_ms": 904.5663280412555, + "ttft_p95_ms": 1458.4544260287657 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r16p00.json", + "sha256": "2ccf16426046c412829442f72932268d5b27ce051a4312d135d32b240b48ec7c", + "summary": { + "admission_lag_max_ms": 0.11741300113499165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.78125, + "passed": 50, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1537.9131329245865, + "ttft_p50_ms": 896.1611289996654, + "ttft_p95_ms": 1471.3564389385283 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3348.027400090359, + 3348.575382027775 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r32p00.json", + "sha256": "d258760d15f1fdf2adefca72f7a3505c0b839036799dcb074450f7f526f76718", + "summary": { + "admission_lag_max_ms": 0.18610397819429636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3442.3487790627405, + "ttft_p50_ms": 1801.6119640087709, + "ttft_p95_ms": 3348.027400090359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r32p00.json", + "sha256": "edf5cccc1400639a94b5430897a04f20d2214bb1ffee7f4bfa62514150179a6a", + "summary": { + "admission_lag_max_ms": 0.1144439447671175, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3443.0311269825324, + "ttft_p50_ms": 1801.435066969134, + "ttft_p95_ms": 3348.575382027775 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4286.911493982188, + 4287.898648995906 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r64p00.json", + "sha256": "2a6303efb826637c3c6448d6f766f7de4fc4d7b63db163a48f72c20df68394f4", + "summary": { + "admission_lag_max_ms": 0.10997895151376724, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.311317006126, + "ttft_p50_ms": 2396.4726959820837, + "ttft_p95_ms": 4286.911493982188 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r64p00.json", + "sha256": "6fafe3bf206583b8689be1b2b57d6b92cbd1ba54f7fb230ca05f6c68b3ab9757", + "summary": { + "admission_lag_max_ms": 0.12669397983700037, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.538130956702, + "ttft_p50_ms": 2394.112486974336, + "ttft_p95_ms": 4287.898648995906 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp2_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.2925309324637, + 96.13083000294864 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r4p00.json", + "sha256": "812f8a108a9390fb759b46e5bd41e3ddad66bc144d47bf4024d834e42b47c47e", + "summary": { + "admission_lag_max_ms": 0.11386198457330465, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.76773096807301, + "ttft_p50_ms": 95.72782705072314, + "ttft_p95_ms": 97.2925309324637 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r4p00.json", + "sha256": "70f922eb7651d46d81d7d394b8453c5b778f921069cf48fb2ae2677afa94a842", + "summary": { + "admission_lag_max_ms": 0.30313804745674133, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 129.06646099872887, + "ttft_p50_ms": 94.53632694203407, + "ttft_p95_ms": 96.13083000294864 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 95.42673197574914, + 98.94769196398556 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r8p00.json", + "sha256": "24097d38b0ffb45f6cdbd93f0aad04dce3b7da41ff45e948eb3a794902b6547f", + "summary": { + "admission_lag_max_ms": 0.11571904178708792, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.04934400785714, + "ttft_p50_ms": 94.34181300457567, + "ttft_p95_ms": 95.42673197574914 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r8p00.json", + "sha256": "7251f076d34ba9484b08fa3fc4b898d2ba8d1199cb1eaeaaffc19930f8c889b9", + "summary": { + "admission_lag_max_ms": 0.1177559606730938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.93720292206854, + "ttft_p50_ms": 94.77824496570975, + "ttft_p95_ms": 98.94769196398556 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.28185909241438, + 95.9081610199064 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r10p00.json", + "sha256": "85796457715a25c15fc909f6def32776fbd49956eb7890df1c4abcef8c5afe78", + "summary": { + "admission_lag_max_ms": 0.14786410611122847, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.46025596279651, + "ttft_p50_ms": 93.54631893802434, + "ttft_p95_ms": 96.28185909241438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r10p00.json", + "sha256": "0bf4d4d006bf0b5bd8b989ceeeb4bacb0e3ee1b8ea66f30b3deb120a14871cf0", + "summary": { + "admission_lag_max_ms": 0.11162098962813616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.54603894241154, + "ttft_p50_ms": 93.77408598084003, + "ttft_p95_ms": 95.9081610199064 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 439.0145930228755, + 429.9719400005415 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r12p00.json", + "sha256": "68de6e36bd7c5624cab107fe5cb21bb817711691e545439a958c926a27e3fe29", + "summary": { + "admission_lag_max_ms": 0.10475562885403633, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 445.90794993564487, + "ttft_p50_ms": 270.3119309153408, + "ttft_p95_ms": 439.0145930228755 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r12p00.json", + "sha256": "d8d3a44a07b46f1673817ece2948fd97c30f41bd451d93a2fa02edf074536c63", + "summary": { + "admission_lag_max_ms": 0.4323029424995184, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 434.6071509644389, + "ttft_p50_ms": 265.2625450864434, + "ttft_p95_ms": 429.9719400005415 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1014.2896320903674, + 999.8802930349484 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r14p00.json", + "sha256": "10196c2f9a9b69c2e47f64d06bf5ee6863af4cd2b612301622f0e304949a942a", + "summary": { + "admission_lag_max_ms": 0.09005970787256956, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1105.1907669752836, + "ttft_p50_ms": 621.6445650206879, + "ttft_p95_ms": 1014.2896320903674 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r14p00.json", + "sha256": "c10cab85588a3563bed36fc4480462635d34d240938beaa0e5034f13186578a6", + "summary": { + "admission_lag_max_ms": 0.11327012907713652, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1090.6395060010254, + "ttft_p50_ms": 623.5586369875818, + "ttft_p95_ms": 999.8802930349484 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1488.3801550604403, + 1484.6198010491207 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r16p00.json", + "sha256": "ef5046c49c32178855bed6a0488ea21e8da22f786ddaabc503bea2d1025bb2f3", + "summary": { + "admission_lag_max_ms": 0.7892940193414688, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1613.616122980602, + "ttft_p50_ms": 905.2404230460525, + "ttft_p95_ms": 1488.3801550604403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r16p00.json", + "sha256": "de02c1d49414cffe7e20023abf4d502a4efc6d2e01fcb3e49e928081aaef77b6", + "summary": { + "admission_lag_max_ms": 0.10548799764364958, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1550.9842830942944, + "ttft_p50_ms": 904.9503420246765, + "ttft_p95_ms": 1484.6198010491207 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3391.811708919704, + 3362.9186518955976 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r32p00.json", + "sha256": "876c5444d17c4ce81eca907ae17d5cf7062ffa67fa266b491580e52f712614ec", + "summary": { + "admission_lag_max_ms": 0.12557301670312881, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3486.074637970887, + "ttft_p50_ms": 1840.204154024832, + "ttft_p95_ms": 3391.811708919704 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r32p00.json", + "sha256": "02b4cd93ce1f9816e822022fb608ba8d4c89265f673f6d27c8fa3d4f1523ffe0", + "summary": { + "admission_lag_max_ms": 0.11014204937964678, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3456.965696066618, + "ttft_p50_ms": 1812.343467026949, + "ttft_p95_ms": 3362.9186518955976 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4306.351027917117, + 4302.602957002819 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r64p00.json", + "sha256": "13ba282f179b2d6ce13a0aa2fd04eb29f82e579b7995e3406cb0afd69a2d5ee2", + "summary": { + "admission_lag_max_ms": 0.3526840591803193, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4359.505696920678, + "ttft_p50_ms": 2411.1130589153618, + "ttft_p95_ms": 4306.351027917117 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r64p00.json", + "sha256": "767622d62f0757ef2d8c7651ea470bfa2e62edad0ed4e75f66f84a0b722e35da", + "summary": { + "admission_lag_max_ms": 0.10626309085637331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4355.409953044727, + "ttft_p50_ms": 2409.5974090741947, + "ttft_p95_ms": 4302.602957002819 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp2_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.54812794178724, + 95.08875198662281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r4p00.json", + "sha256": "6e6a45d0c1010094c3395969815fd8cf9c35b919db202d772fb514744f981ebc", + "summary": { + "admission_lag_max_ms": 0.11306197848170996, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.27749610505998, + "ttft_p50_ms": 93.01548008807003, + "ttft_p95_ms": 96.54812794178724 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r4p00.json", + "sha256": "58aa71d694704cfea15c37cb29420dbac1dc505045c29a731e8be904846cca5f", + "summary": { + "admission_lag_max_ms": 0.11853000614792109, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 98.74943597242236, + "ttft_p50_ms": 93.57558900956064, + "ttft_p95_ms": 95.08875198662281 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.24941800534725, + 96.21043398510665 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r8p00.json", + "sha256": "311350af236cb0306aa635f05f8999b66bf5080bad7cc1f6703fffe0be31aee6", + "summary": { + "admission_lag_max_ms": 0.19771000370383263, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52211803849787, + "ttft_p50_ms": 92.4453770276159, + "ttft_p95_ms": 94.24941800534725 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r8p00.json", + "sha256": "60fe989a36a526e76f6f5c98e217b9feaf081bc680b16b9d53650bf085047c06", + "summary": { + "admission_lag_max_ms": 0.3933720290660858, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.97230398282409, + "ttft_p50_ms": 93.26327103190124, + "ttft_p95_ms": 96.21043398510665 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.98661107197404, + 92.9473100695759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r10p00.json", + "sha256": "d0a04444a0e41bf76f71ac6ea3815bfbe76f7a51f8ddf669976bc3a4896dd0c1", + "summary": { + "admission_lag_max_ms": 0.10270101483911276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 103.49742788821459, + "ttft_p50_ms": 94.55623896792531, + "ttft_p95_ms": 97.98661107197404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r10p00.json", + "sha256": "f330f26f45f6039c48150a0b6b146ca3dbb9ea8b2dff086a9efd5938ab981eb7", + "summary": { + "admission_lag_max_ms": 0.11239596642553806, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.4133670097217, + "ttft_p50_ms": 91.2688790122047, + "ttft_p95_ms": 92.9473100695759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 260.68170997314155, + 319.04888805001974 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r12p00.json", + "sha256": "fed5eacf18018402b0c2675b8c5595f92693a9bddec30f354d989098f9fe4965", + "summary": { + "admission_lag_max_ms": 0.11784560047090054, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 345.19017208367586, + "ttft_p50_ms": 211.41935500781983, + "ttft_p95_ms": 260.68170997314155 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r12p00.json", + "sha256": "f2be27cbeff3e5c1a95050418f2dc3c712a4648cada8cdb3d36382cfffc9593a", + "summary": { + "admission_lag_max_ms": 0.12930459342896938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 326.86293800361454, + "ttft_p50_ms": 229.8808820778504, + "ttft_p95_ms": 319.04888805001974 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 895.7287209341303, + 912.1733580250293 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r14p00.json", + "sha256": "c2e157808e0308cc443f3abdf5114898cef66c435688d24321f80bef2d966208", + "summary": { + "admission_lag_max_ms": 0.14902697876095772, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 962.3544370988384, + "ttft_p50_ms": 576.2168710352853, + "ttft_p95_ms": 895.7287209341303 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r14p00.json", + "sha256": "749ef7d31bf8782d42719b75ee1a2e4a443f097510af721d02cf53df191bc66e", + "summary": { + "admission_lag_max_ms": 0.42896345257759094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 986.0872740391642, + "ttft_p50_ms": 569.7114520007744, + "ttft_p95_ms": 912.1733580250293 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1364.6171110449359, + 1366.6857269126922 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r16p00.json", + "sha256": "7545ea9210348019de88ede5d2cba7527913534274deee034fa6811c173a6e72", + "summary": { + "admission_lag_max_ms": 0.12612901628017426, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1435.8308439841494, + "ttft_p50_ms": 848.9949750946835, + "ttft_p95_ms": 1364.6171110449359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r16p00.json", + "sha256": "d75e27137f6a7811126d8c957bb5a081944fc1bb7643852cdd785111773f6010", + "summary": { + "admission_lag_max_ms": 0.1258660340681672, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1436.2916270038113, + "ttft_p50_ms": 846.2129919789732, + "ttft_p95_ms": 1366.6857269126922 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3243.9196659252048, + 3214.524411014281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r32p00.json", + "sha256": "366ebddc167338ed096f373de5667740721422c13d2db1f887ed09186afafd19", + "summary": { + "admission_lag_max_ms": 0.09669095743447542, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3337.068362045102, + "ttft_p50_ms": 1751.3202880509198, + "ttft_p95_ms": 3243.9196659252048 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r32p00.json", + "sha256": "01b8f5081e36a64d5776285c8c5f423b53b1ad684c4b8f95a8caa5d997c6bf09", + "summary": { + "admission_lag_max_ms": 0.8310690755024552, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3277.297410997562, + "ttft_p50_ms": 1790.5721209244803, + "ttft_p95_ms": 3214.524411014281 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4319.725521025248, + 4367.031275993213 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r64p00.json", + "sha256": "92202709fd6bbd558ff6f1986df213848254110c4274447a7bd5879153290d46", + "summary": { + "admission_lag_max_ms": 0.10392803233116865, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4360.145599930547, + "ttft_p50_ms": 2403.5812759539112, + "ttft_p95_ms": 4319.725521025248 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r64p00.json", + "sha256": "9660d15d31bc70c5550ed29477f563313d4d81baf0e561126aebd3b7d097cfa3", + "summary": { + "admission_lag_max_ms": 0.17276196740567684, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.25, + "passed": 16, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4414.442001027055, + "ttft_p50_ms": 2260.2788450894877, + "ttft_p95_ms": 4367.031275993213 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp2_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.17482794076204, + 98.03147299680859 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r4p00.json", + "sha256": "0e315f6d9e0087ad553e874b0c16c2e6de79eb6eca22eb118dc5deca52e6efe3", + "summary": { + "admission_lag_max_ms": 0.11708191595971584, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.92840604577214, + "ttft_p50_ms": 95.13642196543515, + "ttft_p95_ms": 97.17482794076204 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r4p00.json", + "sha256": "baa394a24d2a3db40a3a7ae6dae0c5ce49ff05499a1e17eaa0bca4b479ae33bf", + "summary": { + "admission_lag_max_ms": 0.10946101974695921, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52333807107061, + "ttft_p50_ms": 93.19468098692596, + "ttft_p95_ms": 98.03147299680859 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.67980605829507, + 93.80627400241792 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r8p00.json", + "sha256": "303df27c6d4604d5af65f9056acacb7c4373fb6eb7edb6e5d42689eeaa384aaa", + "summary": { + "admission_lag_max_ms": 0.11111306957900524, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 113.2329449756071, + "ttft_p50_ms": 93.05417106952518, + "ttft_p95_ms": 97.67980605829507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r8p00.json", + "sha256": "cdbb1bca17b24b67e18ef3dd4997494d6bd2a887b3b5d1264fbf411e688c53bc", + "summary": { + "admission_lag_max_ms": 0.12463307939469814, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.12245299387723, + "ttft_p50_ms": 92.26030297577381, + "ttft_p95_ms": 93.80627400241792 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.45723297540098, + 93.57301203999668 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r10p00.json", + "sha256": "fe17ef2417319307d651bc10b01a618234d61bab797ad87a7332cab20d78ecb2", + "summary": { + "admission_lag_max_ms": 0.3267280990257859, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.28593400400132, + "ttft_p50_ms": 92.22474799025804, + "ttft_p95_ms": 94.45723297540098 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r10p00.json", + "sha256": "550f693708a34624a82d77afd4da0ab4bc01e09f42242a3a79e4532897681a51", + "summary": { + "admission_lag_max_ms": 0.11238199658691883, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.77756198309362, + "ttft_p50_ms": 92.07128698471934, + "ttft_p95_ms": 93.57301203999668 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 320.49866404850036, + 336.9630330707878 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r12p00.json", + "sha256": "edab8169453151de732923c076b8f7dca4082ee47ccd2597e65e7ba68516bc2e", + "summary": { + "admission_lag_max_ms": 0.09914662223309278, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 327.930542989634, + "ttft_p50_ms": 231.4880370395258, + "ttft_p95_ms": 320.49866404850036 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r12p00.json", + "sha256": "a6dde4363d362ca10254409e3dc734d0f0e640e63156426f3f4fd7115db49e16", + "summary": { + "admission_lag_max_ms": 0.17016706988215446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 428.6657409975305, + "ttft_p50_ms": 237.09688894450665, + "ttft_p95_ms": 336.9630330707878 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 911.7195709841326, + 910.7652489328757 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r14p00.json", + "sha256": "7eba81ee5c85c457a056f9aef28abd94d375fdaa116929cfe0a0c282a0738bd2", + "summary": { + "admission_lag_max_ms": 0.13163150288164616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 999.2953530745581, + "ttft_p50_ms": 580.023527960293, + "ttft_p95_ms": 911.7195709841326 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r14p00.json", + "sha256": "6a113d3d876ff01a9c1101e8b75efa2fc71f96683d6bb430e26d5976426b5d23", + "summary": { + "admission_lag_max_ms": 0.11634768452495337, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 988.8870939612389, + "ttft_p50_ms": 578.9327350212261, + "ttft_p95_ms": 910.7652489328757 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1389.8832038976252, + 1365.0342189939693 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r16p00.json", + "sha256": "d677251ee1a941c52469052a239c2e13be0a837285069aa8182a16e196445e48", + "summary": { + "admission_lag_max_ms": 0.11070503387600183, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1460.5093500576913, + "ttft_p50_ms": 864.3530959961936, + "ttft_p95_ms": 1389.8832038976252 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r16p00.json", + "sha256": "f5177f53aa0f87a4e7291f470978d86ad8bc46bd1a8362b79f48cd128646dc10", + "summary": { + "admission_lag_max_ms": 0.3672370221465826, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1491.050804965198, + "ttft_p50_ms": 841.7225609300658, + "ttft_p95_ms": 1365.0342189939693 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3258.010295103304, + 3258.664960041642 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r32p00.json", + "sha256": "02a6ac2b9c2b9ab97854d6fcc13b64bbb36a079e4afdae84231e4cb187fac4ef", + "summary": { + "admission_lag_max_ms": 0.1296499976888299, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.278529899195, + "ttft_p50_ms": 1755.0864539807662, + "ttft_p95_ms": 3258.010295103304 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r32p00.json", + "sha256": "9ca2f44fc7e3cb3d38ada548af858903b31bbee5cd50f1723c29255ebd4224e9", + "summary": { + "admission_lag_max_ms": 1.1343059595674276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.7101600775495, + "ttft_p50_ms": 1755.3453800501302, + "ttft_p95_ms": 3258.664960041642 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4190.255103982054, + 4189.1050330596045 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r64p00.json", + "sha256": "53908d3a254e19691839831b4337014b09591ced9189c2a9b29e0d92e80fcb56", + "summary": { + "admission_lag_max_ms": 0.11591101065278053, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4242.029857938178, + "ttft_p50_ms": 2340.5529640149325, + "ttft_p95_ms": 4190.255103982054 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r64p00.json", + "sha256": "1c8c85c873aa96c01dc0200f848474a53c1173ef72f3828e8c23aeae70e92b11", + "summary": { + "admission_lag_max_ms": 0.284243025816977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4240.770195960067, + "ttft_p50_ms": 2339.3583690049127, + "ttft_p95_ms": 4189.1050330596045 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp2_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 74.66826005838811, + 66.5705680148676 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r4p00.json", + "sha256": "fb354fc8a097caa21844d4ea5515b98b03459f400816e5e5590a6cda81cf4815", + "summary": { + "admission_lag_max_ms": 0.1035250024870038, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 93.38432096410543, + "ttft_p50_ms": 65.75745099689811, + "ttft_p95_ms": 74.66826005838811 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r4p00.json", + "sha256": "ac8e8412f79a8a32ebf085445a94aa548dd2d646fce0b9f56b4378081dcc8960", + "summary": { + "admission_lag_max_ms": 0.12108997907489538, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.89856600761414, + "ttft_p50_ms": 62.386377016082406, + "ttft_p95_ms": 66.5705680148676 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.44799195136875, + 72.29184894822538 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r8p00.json", + "sha256": "05ebbd68bbf24e9c8b25684474411c7db122b92701cc1617f9d786cf19592751", + "summary": { + "admission_lag_max_ms": 0.1825150102376938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.92095802165568, + "ttft_p50_ms": 60.598687967285514, + "ttft_p95_ms": 70.44799195136875 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r8p00.json", + "sha256": "d24f7daaf41073cc8aa030b74c06203783684c3e556770366883c37b826406f2", + "summary": { + "admission_lag_max_ms": 0.10628998279571533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 75.13270596973598, + "ttft_p50_ms": 60.161596979014575, + "ttft_p95_ms": 72.29184894822538 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.02151494007558, + 107.58775402791798 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r16p00.json", + "sha256": "a29b2128983af630e1da8f970cbda802849036bc522eb506821bf2fe1a34172f", + "summary": { + "admission_lag_max_ms": 0.10468997061252594, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.93136999942362, + "ttft_p50_ms": 59.5597909996286, + "ttft_p95_ms": 66.02151494007558 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r16p00.json", + "sha256": "275ae35651cbe2200d16c9b06a68568efda527564065255a10dee6fe93346010", + "summary": { + "admission_lag_max_ms": 0.3144120564684272, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 119.78317995090038, + "ttft_p50_ms": 59.12985489703715, + "ttft_p95_ms": 107.58775402791798 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 193.8455159543082, + 187.52167106140405 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r20p00.json", + "sha256": "4cb485f30b81c004bab379e6522d94b811c0bdc0be6d8e42c9d8ae842d9600dc", + "summary": { + "admission_lag_max_ms": 0.10407401714473963, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 197.7908629924059, + "ttft_p50_ms": 142.73560303263366, + "ttft_p95_ms": 193.8455159543082 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r20p00.json", + "sha256": "63aa9dc03a08b781c5d6e8805f2b61698c75e71673608ecd38bd1132c62e6987", + "summary": { + "admission_lag_max_ms": 0.09386602323502302, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 190.29632199089974, + "ttft_p50_ms": 136.1723990412429, + "ttft_p95_ms": 187.52167106140405 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 568.2198900030926, + 573.4539540717378 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r24p00.json", + "sha256": "5e4d5b44850090d07fe1d3abc44eddea3bfdf2b682f59b12a45a866621e38f98", + "summary": { + "admission_lag_max_ms": 0.7511250441893935, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 613.2891150191426, + "ttft_p50_ms": 375.27791305910796, + "ttft_p95_ms": 568.2198900030926 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r24p00.json", + "sha256": "239e7789aed2e9b2bcf1d127215a6b122071eb856be2db64ccc184fda153aa48", + "summary": { + "admission_lag_max_ms": 0.11180061846971512, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 614.9205249967054, + "ttft_p50_ms": 375.5133739905432, + "ttft_p95_ms": 573.4539540717378 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 860.0242879474536, + 886.1482660286129 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r28p00.json", + "sha256": "8b249b52779e5b3cd9028aaee30ee73ef34b0b0a64dcd75b9aad57266c29fbc2", + "summary": { + "admission_lag_max_ms": 0.11425174307078123, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.8854410191998, + "ttft_p50_ms": 534.7980660153553, + "ttft_p95_ms": 860.0242879474536 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r28p00.json", + "sha256": "c9b28c8f3187542a343ecf1067783569ec2abfa64631d6827a94ba6876aa7b63", + "summary": { + "admission_lag_max_ms": 0.312068616040051, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 922.4315020255744, + "ttft_p50_ms": 549.7413279954344, + "ttft_p95_ms": 886.1482660286129 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1147.1082780044526, + 1137.236590962857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r32p00.json", + "sha256": "d27ff9f940a6cd7fba339d9a176de39cf4b5bd164dcc500f2af36bcb07cdef30", + "summary": { + "admission_lag_max_ms": 0.16146607231348753, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.626160996966, + "ttft_p50_ms": 696.4708579471335, + "ttft_p95_ms": 1147.1082780044526 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r32p00.json", + "sha256": "093044e87fffdaa222e5bad3807cbf4900aa64891d5b4ba27f187c7a411252d7", + "summary": { + "admission_lag_max_ms": 0.10430300608277321, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.045956928283, + "ttft_p50_ms": 684.3201500596479, + "ttft_p95_ms": 1137.236590962857 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2064.763988018967, + 2039.6780440350994 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r64p00.json", + "sha256": "fad7bdd80d5ca603790a4e5d19dadd4f494540b106ce583e3e8c1c05d2374a58", + "summary": { + "admission_lag_max_ms": 6.394341005943716, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.53125, + "passed": 34, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2144.80376499705, + "ttft_p50_ms": 1158.5425049997866, + "ttft_p95_ms": 2064.763988018967 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r64p00.json", + "sha256": "24fdb945706dbbdd29916e0f08e0eff807334db212fded94c729da2e1b46efd3", + "summary": { + "admission_lag_max_ms": 0.10294502135366201, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.4275270039216, + "ttft_p50_ms": 1131.1213460285217, + "ttft_p95_ms": 2039.6780440350994 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.80304794665426, + 64.59337309934199 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r4p00.json", + "sha256": "2e82a30f02c2702668e59661930431cf82323746a111d42bf38b7fdc0d083b3d", + "summary": { + "admission_lag_max_ms": 0.1332589890807867, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.74050498381257, + "ttft_p50_ms": 64.10286005120724, + "ttft_p95_ms": 67.80304794665426 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r4p00.json", + "sha256": "8294196c7dc52765a6820f5e1d8439c480490daed9505263f707810f73c95783", + "summary": { + "admission_lag_max_ms": 0.1283179735764861, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.6796320155263, + "ttft_p50_ms": 60.962425894103944, + "ttft_p95_ms": 64.59337309934199 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.46496798284352, + 68.22343391831964 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r8p00.json", + "sha256": "38bbfeab47e5eb6296fcccf20f04ef25dc24db658be92300c492734bbfe333cf", + "summary": { + "admission_lag_max_ms": 0.10555307380855083, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 87.7126749837771, + "ttft_p50_ms": 62.65173899009824, + "ttft_p95_ms": 70.46496798284352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r8p00.json", + "sha256": "716f90177d6f203717c80a8e9b28697e53141f204af470493c299bc73a1ea3ee", + "summary": { + "admission_lag_max_ms": 0.24518708232790232, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 80.40577603969723, + "ttft_p50_ms": 60.29119901359081, + "ttft_p95_ms": 68.22343391831964 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 89.53388908412308, + 76.93439698778093 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r16p00.json", + "sha256": "063016efe822c31c2425e7d963eda4c054f3f7c0341df0a3cd73ab86a942e873", + "summary": { + "admission_lag_max_ms": 0.2996840048581362, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.41822991054505, + "ttft_p50_ms": 60.97789097111672, + "ttft_p95_ms": 89.53388908412308 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r16p00.json", + "sha256": "eecc6bdeff90a8bf195dfc5387c6bc2f4db20ed7c9aeb9b4ebe050efccc6051a", + "summary": { + "admission_lag_max_ms": 0.36267400719225407, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.42048599943519, + "ttft_p50_ms": 60.29245804529637, + "ttft_p95_ms": 76.93439698778093 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 206.2494510319084, + 201.13301498349756 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r20p00.json", + "sha256": "22a62729358fea7c38ca166a1a49f1235e98c7f07ecfe4948322110cda4b7178", + "summary": { + "admission_lag_max_ms": 0.11367700062692165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 255.14698098413646, + "ttft_p50_ms": 155.46583198010921, + "ttft_p95_ms": 206.2494510319084 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r20p00.json", + "sha256": "be103a759100bd2ab6dd31ee4ce41b6f0da40d232a60ebe137e3d933194a487c", + "summary": { + "admission_lag_max_ms": 0.17853605095297098, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 209.22932599205524, + "ttft_p50_ms": 155.12292610947043, + "ttft_p95_ms": 201.13301498349756 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 580.5047769099474, + 577.3971510352567 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r24p00.json", + "sha256": "890a733a97f399deb4c5efcca661d9a442ed76080cd5d8b6a2c09368aef701b7", + "summary": { + "admission_lag_max_ms": 0.1084543764591217, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 621.5670789824799, + "ttft_p50_ms": 374.6346529806033, + "ttft_p95_ms": 580.5047769099474 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r24p00.json", + "sha256": "87e8543652b1743b1b9b8c3de0fc680401b0bf9a33d61fa6132120c944b2a6eb", + "summary": { + "admission_lag_max_ms": 0.11801638174802065, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 623.2662099646404, + "ttft_p50_ms": 382.11427396163344, + "ttft_p95_ms": 577.3971510352567 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 873.0683700414374, + 889.4930009264499 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r28p00.json", + "sha256": "f877781730a5437b4cbdb76948915f1c871574b19a6df27076d21b65bb18e678", + "summary": { + "admission_lag_max_ms": 0.6807943573221564, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 943.9258739585057, + "ttft_p50_ms": 542.9852310335264, + "ttft_p95_ms": 873.0683700414374 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r28p00.json", + "sha256": "82151dc539b35302489c17976cfe89b8f4c64944298954607fad8f5c9cb9e7d8", + "summary": { + "admission_lag_max_ms": 6.750226486474276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 936.5667289821431, + "ttft_p50_ms": 546.0105559322983, + "ttft_p95_ms": 889.4930009264499 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1135.3467140579596, + 1146.743991994299 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r32p00.json", + "sha256": "082cc4c47b2eec6e49f0edd14cff5a51b880bd998d98da8b9830bbf7615483ef", + "summary": { + "admission_lag_max_ms": 0.10443700011819601, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1199.8225760180503, + "ttft_p50_ms": 680.4794340860099, + "ttft_p95_ms": 1135.3467140579596 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r32p00.json", + "sha256": "517d3cbc1104b5ea1bdb0e381c033cf978181a93449b6498d30b52cd3dd3e770", + "summary": { + "admission_lag_max_ms": 0.4415180301293731, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1210.0602959981188, + "ttft_p50_ms": 693.0603379150853, + "ttft_p95_ms": 1146.743991994299 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2084.9693010095507, + 2041.9401369290426 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r64p00.json", + "sha256": "7e11771ed598826fb3c5594c7e05d0989e01030be2d8b36fa21813aff462496d", + "summary": { + "admission_lag_max_ms": 0.10768393985927105, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.128676981665, + "ttft_p50_ms": 1113.0280290963128, + "ttft_p95_ms": 2084.9693010095507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r64p00.json", + "sha256": "cf550423832a92cba942b7e1dc14a42000e8c9507403de8cd49aa56af9db6706", + "summary": { + "admission_lag_max_ms": 0.11262902989983559, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2122.8132860269397, + "ttft_p50_ms": 1131.2310450011864, + "ttft_p95_ms": 2041.9401369290426 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.89354998525232, + 83.23234610725194 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r4p00.json", + "sha256": "266cba4088f8a9483ceb41fdc02d04a2181cb7ffb6b52372b6b0687e62f2863e", + "summary": { + "admission_lag_max_ms": 0.10568101424723864, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.09669100120664, + "ttft_p50_ms": 62.014322029426694, + "ttft_p95_ms": 66.89354998525232 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r4p00.json", + "sha256": "26f3926c40504b96a37cb7b9dae1af5040c085c1a730e20feaf1d3e9ddbf8e54", + "summary": { + "admission_lag_max_ms": 0.108166947029531, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.42452490516007, + "ttft_p50_ms": 61.261708033271134, + "ttft_p95_ms": 83.23234610725194 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.26283801253885, + 67.34374002553523 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r8p00.json", + "sha256": "e84e6ba2c8822e012aceea7cc6f376ea1b6265fe4a3bb68b7badfc36b8813545", + "summary": { + "admission_lag_max_ms": 0.15694997273385525, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54619695153087, + "ttft_p50_ms": 59.63437003083527, + "ttft_p95_ms": 67.26283801253885 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r8p00.json", + "sha256": "dae50073c9f79919181e8dcf274ca8929ba41c512f64363e40187de68839b87e", + "summary": { + "admission_lag_max_ms": 0.09962497279047966, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.00161597877741, + "ttft_p50_ms": 60.2562150452286, + "ttft_p95_ms": 67.34374002553523 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 71.37343694921583, + 64.9534990079701 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r16p00.json", + "sha256": "776c9956f65acf7c9b6c8df517a4f8c4d462289893a19e625a2a0ad7dcbf5506", + "summary": { + "admission_lag_max_ms": 0.1066719414666295, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 82.25136005785316, + "ttft_p50_ms": 58.52357891853899, + "ttft_p95_ms": 71.37343694921583 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r16p00.json", + "sha256": "90057204a5f6343ee4ee9cdd08323d87d83660d5bf8f02480882eb479d2e587e", + "summary": { + "admission_lag_max_ms": 0.10102998930960894, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 65.3582609957084, + "ttft_p50_ms": 58.98014397826046, + "ttft_p95_ms": 64.9534990079701 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 192.98715889453888, + 258.1990590551868 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r20p00.json", + "sha256": "6ce029388b206d0663acd41f4da1d67a6562389e1c84cf1826ce77c05d176d9a", + "summary": { + "admission_lag_max_ms": 0.12691703159362078, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 198.67739104665816, + "ttft_p50_ms": 142.75032398290932, + "ttft_p95_ms": 192.98715889453888 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r20p00.json", + "sha256": "0dbfb3f012574988ddebfb0af1532c735ac62d6d88863d24cd820aed50a92a59", + "summary": { + "admission_lag_max_ms": 0.3961039474233985, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 259.55891504418105, + "ttft_p50_ms": 162.6282810466364, + "ttft_p95_ms": 258.1990590551868 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 583.3089610096067, + 571.3486529421061 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r24p00.json", + "sha256": "0774ba2f285d81447c63343b32551d6fc33369b54010629381efa9a37df23a9c", + "summary": { + "admission_lag_max_ms": 0.09822729043662548, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 628.714332007803, + "ttft_p50_ms": 366.9840869260952, + "ttft_p95_ms": 583.3089610096067 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r24p00.json", + "sha256": "1986a05c366ebe73af0f89d2cd0e4efd3e577c800cf1e80997cd26ea0aaaec09", + "summary": { + "admission_lag_max_ms": 0.11414836626499891, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 617.9165309295058, + "ttft_p50_ms": 376.65714998729527, + "ttft_p95_ms": 571.3486529421061 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 869.708820944652, + 866.4831441128626 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r28p00.json", + "sha256": "ab987729fc833a59817eabaa18c320b0d0a2ecd402fd02bdb534b6c773c029d8", + "summary": { + "admission_lag_max_ms": 0.578614417463541, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 941.3277570856735, + "ttft_p50_ms": 540.4732850147411, + "ttft_p95_ms": 869.708820944652 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r28p00.json", + "sha256": "38f9c82c9ee30927cc0ba857e5e3fce2e79b8a24de03efaec77ce27258633a7e", + "summary": { + "admission_lag_max_ms": 0.12532551772892475, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 938.2033230504021, + "ttft_p50_ms": 541.139594046399, + "ttft_p95_ms": 866.4831441128626 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1132.6635719742626, + 1144.196804962121 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r32p00.json", + "sha256": "358f5b8e1b72bff5752c076f2966b966d98f6f5767ce52d5a255b7f68f061e54", + "summary": { + "admission_lag_max_ms": 0.32093306072056293, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1197.7118930080906, + "ttft_p50_ms": 678.4568330040202, + "ttft_p95_ms": 1132.6635719742626 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r32p00.json", + "sha256": "3f6dbb760582fdd7470422d8edf6b62949d158ad0d437f04015441a0a4509003", + "summary": { + "admission_lag_max_ms": 1.1304179206490517, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1193.1529619032517, + "ttft_p50_ms": 692.5535890040919, + "ttft_p95_ms": 1144.196804962121 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2042.27216495201, + 2043.943466967903 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r64p00.json", + "sha256": "a24eae7db6eb4cd105ed903917988e8fe6eef5d4cc2549d4e05e9a03dfafb107", + "summary": { + "admission_lag_max_ms": 0.134881935082376, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.23939598538, + "ttft_p50_ms": 1137.3268909519538, + "ttft_p95_ms": 2042.27216495201 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r64p00.json", + "sha256": "e85efaada0b64bb9dd686f59282171ccb256a1d8483c6f037589a68e3e1a104a", + "summary": { + "admission_lag_max_ms": 0.11892011389136314, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.3299129894003, + "ttft_p50_ms": 1134.724210947752, + "ttft_p95_ms": 2043.943466967903 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 82.33969006687403, + 69.57727600820363 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r4p00.json", + "sha256": "e69b89ad2e07e6560ebd870773532b33e4415e9ca1515e585cbe465616678569", + "summary": { + "admission_lag_max_ms": 0.14186999760568142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54833806212991, + "ttft_p50_ms": 62.749656033702195, + "ttft_p95_ms": 82.33969006687403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r4p00.json", + "sha256": "2c76dcf112a828a6720e079d29bb54a6a65d37d26abaaafdc6d5580751dbad65", + "summary": { + "admission_lag_max_ms": 0.11478399392217398, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 81.3983449479565, + "ttft_p50_ms": 62.220132909715176, + "ttft_p95_ms": 69.57727600820363 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 63.71171900536865, + 68.14328604377806 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r8p00.json", + "sha256": "0e5d687dfcdb4b6a0d1d060ceefe8a74faab94a5b241580d45b53c452f893b23", + "summary": { + "admission_lag_max_ms": 0.09818794205784798, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.63772000744939, + "ttft_p50_ms": 60.89008797425777, + "ttft_p95_ms": 63.71171900536865 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r8p00.json", + "sha256": "1bf016b603473d46fda4bc6d4daeaee5e1091ff4cda1cab7224e0c5edea512f8", + "summary": { + "admission_lag_max_ms": 0.11399004142731428, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 76.76690397784114, + "ttft_p50_ms": 60.9520060243085, + "ttft_p95_ms": 68.14328604377806 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 133.86176899075508, + 61.95738597307354 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r16p00.json", + "sha256": "197808b1b5fa8082e603a255f358826b5f8f48707223c7fd63abfe029674b817", + "summary": { + "admission_lag_max_ms": 0.10530801955610514, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 191.68931001331657, + "ttft_p50_ms": 61.92394997924566, + "ttft_p95_ms": 133.86176899075508 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r16p00.json", + "sha256": "53e809470daf9efcdabb254e85e8cc3b45d0ab8593c6c2f3aa3396e60d1656da", + "summary": { + "admission_lag_max_ms": 0.1288399798795581, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.40882098954171, + "ttft_p50_ms": 59.417470009066164, + "ttft_p95_ms": 61.95738597307354 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 201.29013701807708, + 200.0734859611839 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r20p00.json", + "sha256": "05b977824fb301517cdbe2d077b360b6053d57fee1c20acad3537321827d027a", + "summary": { + "admission_lag_max_ms": 0.3848400665447116, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.5107699232176, + "ttft_p50_ms": 150.0862929970026, + "ttft_p95_ms": 201.29013701807708 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r20p00.json", + "sha256": "baf422b7a3c95b0d3d376a31b0a30e075f69216c794ebc2ef51afafe1ec55229", + "summary": { + "admission_lag_max_ms": 0.24754798505455256, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.8192109214142, + "ttft_p50_ms": 148.99871195666492, + "ttft_p95_ms": 200.0734859611839 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 564.2021730309352, + 560.4996709153056 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r24p00.json", + "sha256": "029b2a9695322d536539c3dd7665725f933b21a8bb48fa94c0448c9f7f8bf0eb", + "summary": { + "admission_lag_max_ms": 0.14956865925341845, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 610.4218009859324, + "ttft_p50_ms": 368.3909500250593, + "ttft_p95_ms": 564.2021730309352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r24p00.json", + "sha256": "94b46583af3a7a4bbb9f7256c4c4a5ce804b3092459abe3984ea0ce67f591f3c", + "summary": { + "admission_lag_max_ms": 0.11110503692179918, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 606.6705230623484, + "ttft_p50_ms": 367.50321893487126, + "ttft_p95_ms": 560.4996709153056 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 870.7952990662307, + 894.6108610834926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r28p00.json", + "sha256": "07bfe263107bb3d2acaa356257343f90045e4295c799178bc0a0b5f1c1994284", + "summary": { + "admission_lag_max_ms": 0.6798935355618596, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 942.792383953929, + "ttft_p50_ms": 543.340316042304, + "ttft_p95_ms": 870.7952990662307 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r28p00.json", + "sha256": "861dd55076ef40a78c34824345c5279dff86262b604db120e9e38efa27b5be55", + "summary": { + "admission_lag_max_ms": 0.09912776295095682, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.2741150604561, + "ttft_p50_ms": 549.9696529004723, + "ttft_p95_ms": 894.6108610834926 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1136.603599996306, + 1134.0074760373682 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r32p00.json", + "sha256": "332b805322b571087b018ee459dd91c443dfe0b2bb539b5bdfb4de902bba6da2", + "summary": { + "admission_lag_max_ms": 0.19067595712840557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.8779760692269, + "ttft_p50_ms": 683.3873559953645, + "ttft_p95_ms": 1136.603599996306 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r32p00.json", + "sha256": "996a12a5ed962218dfbe6f30be688119030f10f6539647b6e4a9f25f032b22cb", + "summary": { + "admission_lag_max_ms": 0.11958007235080004, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1198.4891929896548, + "ttft_p50_ms": 682.7383439522237, + "ttft_p95_ms": 1134.0074760373682 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2040.8174369949847, + 2082.3236129945144 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r64p00.json", + "sha256": "c5f17f625071e905b5f1bf8045cd233b2fea0fe6ca00044935dee9c663d1526b", + "summary": { + "admission_lag_max_ms": 0.7995429914444685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2131.160691031255, + "ttft_p50_ms": 1134.8516750149429, + "ttft_p95_ms": 2040.8174369949847 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r64p00.json", + "sha256": "775db5b941347d9fd5d6df32d6ee037f8168bd47b9eae9a7bdb3dbd8c40aa990", + "summary": { + "admission_lag_max_ms": 0.17620902508497238, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2129.518053960055, + "ttft_p50_ms": 1112.3740070033818, + "ttft_p95_ms": 2082.3236129945144 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + }, + "real_campaign_resources": { + "fresh_server_anchors": 192, + "gpu_hours": 12.074444444444445, + "measured_requests": 12288, + "runs": [ + { + "duration_seconds": 908.0, + "gpu_hours": 0.25222222222222224, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive-20260716T164730291172Z", + "tp": 1 + }, + { + "duration_seconds": 540.0, + "gpu_hours": 0.15, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + { + "duration_seconds": 937.0, + "gpu_hours": 0.2602777777777778, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive-20260716T164731376588Z", + "tp": 1 + }, + { + "duration_seconds": 558.0, + "gpu_hours": 0.155, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + { + "duration_seconds": 972.0, + "gpu_hours": 0.27, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive-20260716T164732482666Z", + "tp": 1 + }, + { + "duration_seconds": 572.0, + "gpu_hours": 0.15888888888888889, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + { + "duration_seconds": 886.0, + "gpu_hours": 0.2461111111111111, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive-20260716T164729267199Z", + "tp": 1 + }, + { + "duration_seconds": 525.0, + "gpu_hours": 0.14583333333333334, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + { + "duration_seconds": 983.0, + "gpu_hours": 0.5461111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive-20260716T164734745039Z", + "tp": 2 + }, + { + "duration_seconds": 569.0, + "gpu_hours": 0.3161111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + { + "duration_seconds": 1033.0, + "gpu_hours": 0.5738888888888889, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive-20260716T170909930153Z", + "tp": 2 + }, + { + "duration_seconds": 584.0, + "gpu_hours": 0.3244444444444444, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + { + "duration_seconds": 1063.0, + "gpu_hours": 0.5905555555555555, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive-20260716T170910954319Z", + "tp": 2 + }, + { + "duration_seconds": 597.0, + "gpu_hours": 0.33166666666666667, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + { + "duration_seconds": 957.0, + "gpu_hours": 0.5316666666666666, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive-20260716T164733581881Z", + "tp": 2 + }, + { + "duration_seconds": 563.0, + "gpu_hours": 0.31277777777777777, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + { + "duration_seconds": 1008.0, + "gpu_hours": 1.12, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive-20260716T172922998830Z", + "tp": 4 + }, + { + "duration_seconds": 559.0, + "gpu_hours": 0.6211111111111111, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + { + "duration_seconds": 1024.0, + "gpu_hours": 1.1377777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive-20260716T172923955862Z", + "tp": 4 + }, + { + "duration_seconds": 556.0, + "gpu_hours": 0.6177777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + { + "duration_seconds": 959.0, + "gpu_hours": 1.0655555555555556, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive-20260716T175248149074Z", + "tp": 4 + }, + { + "duration_seconds": 567.0, + "gpu_hours": 0.63, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + { + "duration_seconds": 995.0, + "gpu_hours": 1.1055555555555556, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive-20260716T170911977980Z", + "tp": 4 + }, + { + "duration_seconds": 550.0, + "gpu_hours": 0.6111111111111112, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + ], + "successful_fleet_jobs": 24, + "warmup_requests": 4512 + }, + "schema": "qwen30-prefill-fidelity-comparison-v1", + "simulator": { + "tp1_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 166.6461504479848 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 575.6631856791392 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 16.0, + "ttft_p95_ms": 3529.1762077816857 + }, + { + "feasible": false, + "pass_rate": 0.1875, + "rate": 32.0, + "ttft_p95_ms": 5290.0301465092725 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 6180.6551465092725 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 166.6461504479848 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 575.6631856791392 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 16.0, + "ttft_p95_ms": 3529.1762077816857 + }, + { + "feasible": false, + "pass_rate": 0.1875, + "rate": 32.0, + "ttft_p95_ms": 5290.0301465092725 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 6180.6551465092725 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 166.6461504479848 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 575.6631856791392 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 16.0, + "ttft_p95_ms": 3529.1762077816857 + }, + { + "feasible": false, + "pass_rate": 0.1875, + "rate": 32.0, + "ttft_p95_ms": 5290.0301465092725 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 6180.6551465092725 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 166.6461504479848 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 575.6631856791392 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 16.0, + "ttft_p95_ms": 3529.1762077816857 + }, + { + "feasible": false, + "pass_rate": 0.1875, + "rate": 32.0, + "ttft_p95_ms": 5290.0301465092725 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 6180.6551465092725 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 118.55152318709017 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 118.55152318708929 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 786.9875986853465 + }, + { + "feasible": false, + "pass_rate": 0.421875, + "rate": 32.0, + "ttft_p95_ms": 2566.908354521784 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 64.0, + "ttft_p95_ms": 3439.824414996849 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 118.55152318709017 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 118.55152318708929 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 786.9875986853465 + }, + { + "feasible": false, + "pass_rate": 0.421875, + "rate": 32.0, + "ttft_p95_ms": 2566.908354521784 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 64.0, + "ttft_p95_ms": 3439.824414996849 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 118.55152318709017 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 118.55152318708929 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 786.9875986853465 + }, + { + "feasible": false, + "pass_rate": 0.421875, + "rate": 32.0, + "ttft_p95_ms": 2566.908354521784 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 64.0, + "ttft_p95_ms": 3439.824414996849 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 118.55152318709017 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 118.55152318708929 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 786.9875986853465 + }, + { + "feasible": false, + "pass_rate": 0.421875, + "rate": 32.0, + "ttft_p95_ms": 2566.908354521784 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 64.0, + "ttft_p95_ms": 3439.824414996849 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 86.51189290936045 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 86.51189290936001 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 252.16908885537626 + }, + { + "feasible": true, + "pass_rate": 0.984375, + "rate": 32.0, + "ttft_p95_ms": 1219.2127732631457 + }, + { + "feasible": false, + "pass_rate": 0.515625, + "rate": 64.0, + "ttft_p95_ms": 2106.0636248827373 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 86.51189290936045 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 86.51189290936001 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 252.16908885537626 + }, + { + "feasible": true, + "pass_rate": 0.984375, + "rate": 32.0, + "ttft_p95_ms": 1219.2127732631457 + }, + { + "feasible": false, + "pass_rate": 0.515625, + "rate": 64.0, + "ttft_p95_ms": 2106.0636248827373 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 86.51189290936045 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 86.51189290936001 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 252.16908885537626 + }, + { + "feasible": true, + "pass_rate": 0.984375, + "rate": 32.0, + "ttft_p95_ms": 1219.2127732631457 + }, + { + "feasible": false, + "pass_rate": 0.515625, + "rate": 64.0, + "ttft_p95_ms": 2106.0636248827373 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 86.51189290936045 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 86.51189290936001 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 252.16908885537626 + }, + { + "feasible": true, + "pass_rate": 0.984375, + "rate": 32.0, + "ttft_p95_ms": 1219.2127732631457 + }, + { + "feasible": false, + "pass_rate": 0.515625, + "rate": 64.0, + "ttft_p95_ms": 2106.0636248827373 + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + } + }, + "simulator_sources": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/simulator-a3/frontier_surface_frozen.json", + "sha256": "fe7392fbcd20f6a96562773b3644df9dac23c8562462356acc86585f7f1d40f0" + } + ] +} diff --git a/runs/frontier-fidelity-envelope-v1/results/a3/frontier_surface_frozen.json b/runs/frontier-fidelity-envelope-v1/results/a3/frontier_surface_frozen.json new file mode 100644 index 0000000..34009eb --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/a3/frontier_surface_frozen.json @@ -0,0 +1,1736 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + } + ], + "collective": { + "allreduce_csv": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv", + "allreduce_csv_sha256": "9d693fd406616b599e57bcde66c980c7fc2831b3acf37d3eb633cec80ea0070d", + "backend": "vidur" + }, + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 157.35246539115906, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "705075942dea05ebacb4c83c2587a865e40e1c7af243d2894fa7a05409f8fe59", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.020947591286289, + "ttft_max_ms": 166.6461504479848, + "ttft_p50_ms": 166.64615044798398, + "ttft_p95_ms": 166.6461504479848 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.141019344329834, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "5dbdc7f1672445ef892ed104c0ea1bfb02baba3f87e8a3e707c523707861b747", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.75274759930899, + "ttft_max_ms": 586.937565208359, + "ttft_p50_ms": 436.98066306434936, + "ttft_p95_ms": 575.6631856791392 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.098863363265991, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "58835b434e8f85b91a8d14681bf4a6a0df5f38c69ead31bbc6e02c1c50f8a9c1", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 8.523874834682788, + "ttft_max_ms": 3654.1762077816857, + "ttft_p50_ms": 1994.5136762864174, + "ttft_p95_ms": 3529.1762077816857 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.089977025985718, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a83336424fd7682e60e1a02a62d5c1edbcded34f14919b6011c8bc76a3f24de1", + "score": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 5483.402803563629, + "ttft_p50_ms": 2911.6176150140045, + "ttft_p95_ms": 5290.0301465092725 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.088061094284058, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "000767c0b12d8ec4c9e83f978b3552c8c6b99a2b42414100fc1a7b3e8e5c721e", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 6436.527803563629, + "ttft_p50_ms": 3380.3676150140045, + "ttft_p95_ms": 6180.6551465092725 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 8.947673320770264, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "705075942dea05ebacb4c83c2587a865e40e1c7af243d2894fa7a05409f8fe59", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.020947591286289, + "ttft_max_ms": 166.6461504479848, + "ttft_p50_ms": 166.64615044798398, + "ttft_p95_ms": 166.6461504479848 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.042973518371582, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "5dbdc7f1672445ef892ed104c0ea1bfb02baba3f87e8a3e707c523707861b747", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.75274759930899, + "ttft_max_ms": 586.937565208359, + "ttft_p50_ms": 436.98066306434936, + "ttft_p95_ms": 575.6631856791392 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.092506170272827, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "58835b434e8f85b91a8d14681bf4a6a0df5f38c69ead31bbc6e02c1c50f8a9c1", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 8.523874834682788, + "ttft_max_ms": 3654.1762077816857, + "ttft_p50_ms": 1994.5136762864174, + "ttft_p95_ms": 3529.1762077816857 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.044708013534546, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a83336424fd7682e60e1a02a62d5c1edbcded34f14919b6011c8bc76a3f24de1", + "score": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 5483.402803563629, + "ttft_p50_ms": 2911.6176150140045, + "ttft_p95_ms": 5290.0301465092725 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.29175853729248, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "000767c0b12d8ec4c9e83f978b3552c8c6b99a2b42414100fc1a7b3e8e5c721e", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 6436.527803563629, + "ttft_p50_ms": 3380.3676150140045, + "ttft_p95_ms": 6180.6551465092725 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 8.991463661193848, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "705075942dea05ebacb4c83c2587a865e40e1c7af243d2894fa7a05409f8fe59", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.020947591286289, + "ttft_max_ms": 166.6461504479848, + "ttft_p50_ms": 166.64615044798398, + "ttft_p95_ms": 166.6461504479848 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.138724327087402, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "5dbdc7f1672445ef892ed104c0ea1bfb02baba3f87e8a3e707c523707861b747", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.75274759930899, + "ttft_max_ms": 586.937565208359, + "ttft_p50_ms": 436.98066306434936, + "ttft_p95_ms": 575.6631856791392 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.188970565795898, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "58835b434e8f85b91a8d14681bf4a6a0df5f38c69ead31bbc6e02c1c50f8a9c1", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 8.523874834682788, + "ttft_max_ms": 3654.1762077816857, + "ttft_p50_ms": 1994.5136762864174, + "ttft_p95_ms": 3529.1762077816857 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.189127683639526, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a83336424fd7682e60e1a02a62d5c1edbcded34f14919b6011c8bc76a3f24de1", + "score": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 5483.402803563629, + "ttft_p50_ms": 2911.6176150140045, + "ttft_p95_ms": 5290.0301465092725 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.189879894256592, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "000767c0b12d8ec4c9e83f978b3552c8c6b99a2b42414100fc1a7b3e8e5c721e", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 6436.527803563629, + "ttft_p50_ms": 3380.3676150140045, + "ttft_p95_ms": 6180.6551465092725 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.999529123306274, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "705075942dea05ebacb4c83c2587a865e40e1c7af243d2894fa7a05409f8fe59", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.020947591286289, + "ttft_max_ms": 166.6461504479848, + "ttft_p50_ms": 166.64615044798398, + "ttft_p95_ms": 166.6461504479848 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.192278385162354, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "5dbdc7f1672445ef892ed104c0ea1bfb02baba3f87e8a3e707c523707861b747", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.75274759930899, + "ttft_max_ms": 586.937565208359, + "ttft_p50_ms": 436.98066306434936, + "ttft_p95_ms": 575.6631856791392 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.040328979492188, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "58835b434e8f85b91a8d14681bf4a6a0df5f38c69ead31bbc6e02c1c50f8a9c1", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 8.523874834682788, + "ttft_max_ms": 3654.1762077816857, + "ttft_p50_ms": 1994.5136762864174, + "ttft_p95_ms": 3529.1762077816857 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 8.990798473358154, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a83336424fd7682e60e1a02a62d5c1edbcded34f14919b6011c8bc76a3f24de1", + "score": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 5483.402803563629, + "ttft_p50_ms": 2911.6176150140045, + "ttft_p95_ms": 5290.0301465092725 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.192291736602783, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "000767c0b12d8ec4c9e83f978b3552c8c6b99a2b42414100fc1a7b3e8e5c721e", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 8.660758725922316, + "ttft_max_ms": 6436.527803563629, + "ttft_p50_ms": 3380.3676150140045, + "ttft_p95_ms": 6180.6551465092725 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.093063116073608, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "195c2750784a03930b1ae369648ed6361fb8d1ea506f0828bc7c9818a4f55961", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.033134335322499, + "ttft_max_ms": 118.55152318709017, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318709017 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 160.23531913757324, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "733c6a6bca1efc56f9a160dd89a9156446581d5a1c4c072b90f0a7659acc637e", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.006453678862723, + "ttft_max_ms": 118.55152318708929, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318708929 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.289193630218506, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "37ba1e79ce560f7449049a3ef35429ee1601b35f249800f5e2e0b4d96121f0a4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.760059242174107, + "ttft_max_ms": 849.4875986853465, + "ttft_p50_ms": 539.7284419614418, + "ttft_p95_ms": 786.9875986853465 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.239543676376343, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "1fb27df916c2d9002f5584c9a3ac3ca5fceed2f5fece4f40e69748dbad7aee3b", + "score": { + "feasible": false, + "pass_rate": 0.421875, + "passed": 27, + "throughput_requests_per_second": 14.110410219984, + "ttft_max_ms": 2660.658354521784, + "ttft_p50_ms": 1414.3128122232463, + "ttft_p95_ms": 2566.908354521784 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.141466856002808, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "ccefacf7e9b214037083892f6fa84c5e0fd288f041ebfe21880b9709c8b25727", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 14.110410219983999, + "ttft_max_ms": 3582.533354521785, + "ttft_p50_ms": 1896.1470654856275, + "ttft_p95_ms": 3439.824414996849 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.042891025543213, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "195c2750784a03930b1ae369648ed6361fb8d1ea506f0828bc7c9818a4f55961", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.033134335322499, + "ttft_max_ms": 118.55152318709017, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318709017 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.136998653411865, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "733c6a6bca1efc56f9a160dd89a9156446581d5a1c4c072b90f0a7659acc637e", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.006453678862723, + "ttft_max_ms": 118.55152318708929, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318708929 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.13957166671753, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "37ba1e79ce560f7449049a3ef35429ee1601b35f249800f5e2e0b4d96121f0a4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.760059242174107, + "ttft_max_ms": 849.4875986853465, + "ttft_p50_ms": 539.7284419614418, + "ttft_p95_ms": 786.9875986853465 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.098120927810669, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "1fb27df916c2d9002f5584c9a3ac3ca5fceed2f5fece4f40e69748dbad7aee3b", + "score": { + "feasible": false, + "pass_rate": 0.421875, + "passed": 27, + "throughput_requests_per_second": 14.110410219984, + "ttft_max_ms": 2660.658354521784, + "ttft_p50_ms": 1414.3128122232463, + "ttft_p95_ms": 2566.908354521784 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.140004634857178, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "ccefacf7e9b214037083892f6fa84c5e0fd288f041ebfe21880b9709c8b25727", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 14.110410219983999, + "ttft_max_ms": 3582.533354521785, + "ttft_p50_ms": 1896.1470654856275, + "ttft_p95_ms": 3439.824414996849 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.090512752532959, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "195c2750784a03930b1ae369648ed6361fb8d1ea506f0828bc7c9818a4f55961", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.033134335322499, + "ttft_max_ms": 118.55152318709017, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318709017 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.102084636688232, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "733c6a6bca1efc56f9a160dd89a9156446581d5a1c4c072b90f0a7659acc637e", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.006453678862723, + "ttft_max_ms": 118.55152318708929, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318708929 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.09265947341919, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "37ba1e79ce560f7449049a3ef35429ee1601b35f249800f5e2e0b4d96121f0a4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.760059242174107, + "ttft_max_ms": 849.4875986853465, + "ttft_p50_ms": 539.7284419614418, + "ttft_p95_ms": 786.9875986853465 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.14569616317749, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "1fb27df916c2d9002f5584c9a3ac3ca5fceed2f5fece4f40e69748dbad7aee3b", + "score": { + "feasible": false, + "pass_rate": 0.421875, + "passed": 27, + "throughput_requests_per_second": 14.110410219984, + "ttft_max_ms": 2660.658354521784, + "ttft_p50_ms": 1414.3128122232463, + "ttft_p95_ms": 2566.908354521784 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.191733598709106, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "ccefacf7e9b214037083892f6fa84c5e0fd288f041ebfe21880b9709c8b25727", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 14.110410219983999, + "ttft_max_ms": 3582.533354521785, + "ttft_p50_ms": 1896.1470654856275, + "ttft_p95_ms": 3439.824414996849 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.146432876586914, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "195c2750784a03930b1ae369648ed6361fb8d1ea506f0828bc7c9818a4f55961", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.033134335322499, + "ttft_max_ms": 118.55152318709017, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318709017 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.068153619766235, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "733c6a6bca1efc56f9a160dd89a9156446581d5a1c4c072b90f0a7659acc637e", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.006453678862723, + "ttft_max_ms": 118.55152318708929, + "ttft_p50_ms": 118.55152318708929, + "ttft_p95_ms": 118.55152318708929 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.139062404632568, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "37ba1e79ce560f7449049a3ef35429ee1601b35f249800f5e2e0b4d96121f0a4", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.760059242174107, + "ttft_max_ms": 849.4875986853465, + "ttft_p50_ms": 539.7284419614418, + "ttft_p95_ms": 786.9875986853465 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.143300771713257, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "1fb27df916c2d9002f5584c9a3ac3ca5fceed2f5fece4f40e69748dbad7aee3b", + "score": { + "feasible": false, + "pass_rate": 0.421875, + "passed": 27, + "throughput_requests_per_second": 14.110410219984, + "ttft_max_ms": 2660.658354521784, + "ttft_p50_ms": 1414.3128122232463, + "ttft_p95_ms": 2566.908354521784 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.092010021209717, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "ccefacf7e9b214037083892f6fa84c5e0fd288f041ebfe21880b9709c8b25727", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 14.110410219983999, + "ttft_max_ms": 3582.533354521785, + "ttft_p50_ms": 1896.1470654856275, + "ttft_p95_ms": 3439.824414996849 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 161.351567029953, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "d1d9086461911338e6c87a8303aed090f25e2bbbfd184341e2df531f64c23355", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041293968822475, + "ttft_max_ms": 86.51189290936045, + "ttft_p50_ms": 86.51189290936001, + "ttft_p95_ms": 86.51189290936045 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.191051244735718, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "ee23707d601c6b4559c7f019b09020acdb589844eeffb79f6b46f2365805ac99", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.038674169035575, + "ttft_max_ms": 86.51189290936001, + "ttft_p50_ms": 86.51189290935957, + "ttft_p95_ms": 86.51189290936001 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.050537109375, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "63cce9144c69236a07bb94cdef18c7d09dddeaa1c0ba549a36ff29742c780c2a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.535832140198398, + "ttft_max_ms": 276.9276050512883, + "ttft_p50_ms": 189.66908885537626, + "ttft_p95_ms": 252.16908885537626 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.04306149482727, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "568476977590c981958cc0acc4e07547c079a2ae00520b59f8ed7761acb67bfb", + "score": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63, + "throughput_requests_per_second": 19.925437356037587, + "ttft_max_ms": 1281.7127732631457, + "ttft_p50_ms": 729.6185162106785, + "ttft_p95_ms": 1219.2127732631457 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.14296841621399, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "f789fb967a981edf5c112d0e1a5d0a034456e2cf51f49d6e2e8597afe9df0f2f", + "score": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33, + "throughput_requests_per_second": 20.34996647854275, + "ttft_max_ms": 2191.8433254998167, + "ttft_p50_ms": 1169.8561499618286, + "ttft_p95_ms": 2106.0636248827373 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.14107608795166, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "d1d9086461911338e6c87a8303aed090f25e2bbbfd184341e2df531f64c23355", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041293968822475, + "ttft_max_ms": 86.51189290936045, + "ttft_p50_ms": 86.51189290936001, + "ttft_p95_ms": 86.51189290936045 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.139281272888184, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "ee23707d601c6b4559c7f019b09020acdb589844eeffb79f6b46f2365805ac99", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.038674169035575, + "ttft_max_ms": 86.51189290936001, + "ttft_p50_ms": 86.51189290935957, + "ttft_p95_ms": 86.51189290936001 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.240875244140625, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "63cce9144c69236a07bb94cdef18c7d09dddeaa1c0ba549a36ff29742c780c2a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.535832140198398, + "ttft_max_ms": 276.9276050512883, + "ttft_p50_ms": 189.66908885537626, + "ttft_p95_ms": 252.16908885537626 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.090924739837646, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "568476977590c981958cc0acc4e07547c079a2ae00520b59f8ed7761acb67bfb", + "score": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63, + "throughput_requests_per_second": 19.925437356037587, + "ttft_max_ms": 1281.7127732631457, + "ttft_p50_ms": 729.6185162106785, + "ttft_p95_ms": 1219.2127732631457 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.041213274002075, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "f789fb967a981edf5c112d0e1a5d0a034456e2cf51f49d6e2e8597afe9df0f2f", + "score": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33, + "throughput_requests_per_second": 20.34996647854275, + "ttft_max_ms": 2191.8433254998167, + "ttft_p50_ms": 1169.8561499618286, + "ttft_p95_ms": 2106.0636248827373 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.092233896255493, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "d1d9086461911338e6c87a8303aed090f25e2bbbfd184341e2df531f64c23355", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041293968822475, + "ttft_max_ms": 86.51189290936045, + "ttft_p50_ms": 86.51189290936001, + "ttft_p95_ms": 86.51189290936045 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.093050956726074, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "ee23707d601c6b4559c7f019b09020acdb589844eeffb79f6b46f2365805ac99", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.038674169035575, + "ttft_max_ms": 86.51189290936001, + "ttft_p50_ms": 86.51189290935957, + "ttft_p95_ms": 86.51189290936001 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.242159605026245, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "63cce9144c69236a07bb94cdef18c7d09dddeaa1c0ba549a36ff29742c780c2a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.535832140198398, + "ttft_max_ms": 276.9276050512883, + "ttft_p50_ms": 189.66908885537626, + "ttft_p95_ms": 252.16908885537626 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.19024658203125, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "568476977590c981958cc0acc4e07547c079a2ae00520b59f8ed7761acb67bfb", + "score": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63, + "throughput_requests_per_second": 19.925437356037587, + "ttft_max_ms": 1281.7127732631457, + "ttft_p50_ms": 729.6185162106785, + "ttft_p95_ms": 1219.2127732631457 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.18896198272705, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "f789fb967a981edf5c112d0e1a5d0a034456e2cf51f49d6e2e8597afe9df0f2f", + "score": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33, + "throughput_requests_per_second": 20.34996647854275, + "ttft_max_ms": 2191.8433254998167, + "ttft_p50_ms": 1169.8561499618286, + "ttft_p95_ms": 2106.0636248827373 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.088153839111328, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "d1d9086461911338e6c87a8303aed090f25e2bbbfd184341e2df531f64c23355", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.041293968822475, + "ttft_max_ms": 86.51189290936045, + "ttft_p50_ms": 86.51189290936001, + "ttft_p95_ms": 86.51189290936045 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.048948764801025, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "ee23707d601c6b4559c7f019b09020acdb589844eeffb79f6b46f2365805ac99", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.038674169035575, + "ttft_max_ms": 86.51189290936001, + "ttft_p50_ms": 86.51189290935957, + "ttft_p95_ms": 86.51189290936001 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.04839539527893, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "63cce9144c69236a07bb94cdef18c7d09dddeaa1c0ba549a36ff29742c780c2a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.535832140198398, + "ttft_max_ms": 276.9276050512883, + "ttft_p50_ms": 189.66908885537626, + "ttft_p95_ms": 252.16908885537626 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.139516353607178, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "568476977590c981958cc0acc4e07547c079a2ae00520b59f8ed7761acb67bfb", + "score": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63, + "throughput_requests_per_second": 19.925437356037587, + "ttft_max_ms": 1281.7127732631457, + "ttft_p50_ms": 729.6185162106785, + "ttft_p95_ms": 1219.2127732631457 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.145836114883423, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "f789fb967a981edf5c112d0e1a5d0a034456e2cf51f49d6e2e8597afe9df0f2f", + "score": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33, + "throughput_requests_per_second": 20.34996647854275, + "ttft_max_ms": 2191.8433254998167, + "ttft_p50_ms": 1169.8561499618286, + "ttft_p95_ms": 2106.0636248827373 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 4.0, + 8.0, + 16.0, + 32.0, + 64.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/cc_backend/backends/vidur_cc_backend.py\n M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n?? tests/unit/test_vidur_cc_large_payload.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp1-20260717-v1-20260717T014550833983Z/artifacts/artifacts/attention-composition-tp1-v1/raw/flashattn-composition-tp1.json": "ce461be92a5059e2d168b2cbf21b55d9a55ee71c3b0586a3e63cd8288938168d", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp2-20260717-v1-20260717T014551831545Z/artifacts/artifacts/attention-composition-tp2-v1/raw/flashattn-composition-tp2.json": "997d67fb3a28b488c3aa4945ae7c6d4ecf8a08726ee3cb4652d16c221ff395ad", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp4-20260717-v1-20260717T014552897206Z/artifacts/artifacts/attention-composition-tp4-v1/raw/flashattn-composition-tp4.json": "046cffd4c11d8a36513167ec25d3e1ca016324715679ea688a58d4a71dfe62b2", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "03180eec28087a301c9d8c185874c50bdb28b31e5e9af35fcce5a6a5a9e62147", + "attention_true_mixed_fused.csv": "cccf76e6cb6a9359f188b07d4dc2942442568a6e8cf3ff3d77ab3d35a052cecb", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving-pure-prefill-batch-composition", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_pure_prefill_batch_composition": "Direct FA3 measurements for 2/4 requests at query length 2048 and 2/4/8/16 requests at query length 512 for each TP; included only when batch-composition attention inputs are supplied", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 150, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v3" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final", + "sha256": { + "attention": "03180eec28087a301c9d8c185874c50bdb28b31e5e9af35fcce5a6a5a9e62147", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "70aecb6d70f9e509e9816b1a0929163b9505760c2ac30bbd4cd8e51badfb7635", + "moe": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "frozen_before_real" +} diff --git a/runs/frontier-fidelity-envelope-v1/results/a3/qwen30-prefill-ranking.png b/runs/frontier-fidelity-envelope-v1/results/a3/qwen30-prefill-ranking.png new file mode 100644 index 0000000..ed8a221 Binary files /dev/null and b/runs/frontier-fidelity-envelope-v1/results/a3/qwen30-prefill-ranking.png differ diff --git a/runs/frontier-fidelity-envelope-v1/results/f1-pilot/README.md b/runs/frontier-fidelity-envelope-v1/results/f1-pilot/README.md new file mode 100644 index 0000000..ef9e445 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/f1-pilot/README.md @@ -0,0 +1,10 @@ +# F1 64-request pilot（非 decision-bearing） + +该 surface 使用固定 `ISL=512/OSL=1`、uniform QPS、prefix off 和每个 anchor +64 requests。它完成了 runner correctness smoke，但不能用于 capacity/ranking claim：在 +512 req/s 时，64 个请求只覆盖 0.123 秒，TP4 的 64/64 SLO pass 只说明这个有限 burst +能在 TTFT SLO 内排空，并不说明系统能持续承载 512 req/s。 + +Decision-bearing F1 改为每个 anchor 至少 64 requests 且至少 10 秒 steady arrivals； +物化 contract 位于 `../../workloads/f1-isl512-osl1-steady10s/manifest.json`。原始 pilot +manifest 保留用于审计，不与 steady surface 合并。 diff --git a/runs/frontier-fidelity-envelope-v1/results/f1-pilot/frontier_trace_surface_frozen.json b/runs/frontier-fidelity-envelope-v1/results/f1-pilot/frontier_trace_surface_frozen.json new file mode 100644 index 0000000..bb21c11 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/results/f1-pilot/frontier_trace_surface_frozen.json @@ -0,0 +1,3912 @@ +{ + "collective": { + "allreduce_csv": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv", + "allreduce_csv_sha256": "9d693fd406616b599e57bcde66c980c7fc2831b3acf37d3eb633cec80ea0070d", + "backend": "vidur" + }, + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.23979640007019, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "ad907fe8c9476f468f30802108ab33b012d857f0b2082a55159b4645e201980a", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 216.29366351634627, + "ttft_p95_ms": 283.4691372063174 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.139918327331543, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "63476fae7272db8d2636c94e535f4f14bafa5e8b7b0e6d2cb9b1e840dd686115", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 259.0172330518654, + "ttft_p95_ms": 347.0536850751076 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.040833234786987, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "5b277eafe865f738e74074ea50ad124c7acc79e034f09c78ec4c588db7bcac95", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 749.1180253669007, + "ttft_p95_ms": 1235.420300361212 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 8.89000153541565, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "07bd77c8416c5ff0e9b83861669f833c126ac2fb3b0a9a6ae2830fc8e2e9041b", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 952.2430253669007, + "ttft_p95_ms": 1704.170300361212 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.043700695037842, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "f939698c1a72ad9c6da784797fe3044c0eaeec87f311bcf2a9b35fc699aa9877", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 1053.8055253669008, + "ttft_p95_ms": 1938.545300361212 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 9.08841609954834, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 512.0, + "request_count": 64, + "request_metrics_sha256": "3734b160b72632c9d6eff25ab66e2db66a33685af385f22ff779454825604871", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 1104.5867753669008, + "ttft_p95_ms": 2055.732800361212 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.090706586837769, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "ad907fe8c9476f468f30802108ab33b012d857f0b2082a55159b4645e201980a", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 216.29366351634627, + "ttft_p95_ms": 283.4691372063174 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.0899977684021, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "63476fae7272db8d2636c94e535f4f14bafa5e8b7b0e6d2cb9b1e840dd686115", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 259.0172330518654, + "ttft_p95_ms": 347.0536850751076 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.190520524978638, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "63a37dec7df065e7e8c8ef5ada849898c99795af0873bbe32c4120596580e435", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 733.829341199415, + "ttft_p95_ms": 1045.2671022385387 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.088061571121216, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "043cb4e37ec062487ecf8bf2910d3660fe3d5846be13214ac2a8beb668e19192", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 870.0662724508372, + "ttft_p95_ms": 1477.3167945290845 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.192060708999634, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "e7f32c2152c42a95203201dba11f2c561eeaf3cfc9c6e9e68d89df2294475510", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 940.3787724508372, + "ttft_p95_ms": 1680.4417945290845 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 9.089672565460205, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 512.0, + "request_count": 64, + "request_metrics_sha256": "d879f833e5dc08d711f48fb6c6ccf48f7891fe7694470b107747ed0a1b95dac7", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 975.5350224508372, + "ttft_p95_ms": 1782.0042945290845 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.191780090332031, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "ad907fe8c9476f468f30802108ab33b012d857f0b2082a55159b4645e201980a", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 216.29366351634627, + "ttft_p95_ms": 283.4691372063174 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.189058303833008, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "63476fae7272db8d2636c94e535f4f14bafa5e8b7b0e6d2cb9b1e840dd686115", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 259.0172330518654, + "ttft_p95_ms": 347.0536850751076 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.190616607666016, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "63a37dec7df065e7e8c8ef5ada849898c99795af0873bbe32c4120596580e435", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 733.829341199415, + "ttft_p95_ms": 1045.2671022385387 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.140897035598755, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "043cb4e37ec062487ecf8bf2910d3660fe3d5846be13214ac2a8beb668e19192", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 870.0662724508372, + "ttft_p95_ms": 1477.3167945290845 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.195079803466797, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "e7f32c2152c42a95203201dba11f2c561eeaf3cfc9c6e9e68d89df2294475510", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 940.3787724508372, + "ttft_p95_ms": 1680.4417945290845 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.291805744171143, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 512.0, + "request_count": 64, + "request_metrics_sha256": "d879f833e5dc08d711f48fb6c6ccf48f7891fe7694470b107747ed0a1b95dac7", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 975.5350224508372, + "ttft_p95_ms": 1782.0042945290845 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.336015462875366, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "ad907fe8c9476f468f30802108ab33b012d857f0b2082a55159b4645e201980a", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 216.29366351634627, + "ttft_p95_ms": 283.4691372063174 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.189040660858154, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "63476fae7272db8d2636c94e535f4f14bafa5e8b7b0e6d2cb9b1e840dd686115", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 259.0172330518654, + "ttft_p95_ms": 347.0536850751076 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.28804874420166, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "63a37dec7df065e7e8c8ef5ada849898c99795af0873bbe32c4120596580e435", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 0.984375, + "passed": 63 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 733.829341199415, + "ttft_p95_ms": 1045.2671022385387 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.286862850189209, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "043cb4e37ec062487ecf8bf2910d3660fe3d5846be13214ac2a8beb668e19192", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 870.0662724508372, + "ttft_p95_ms": 1477.3167945290845 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.135990619659424, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "e7f32c2152c42a95203201dba11f2c561eeaf3cfc9c6e9e68d89df2294475510", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 940.3787724508372, + "ttft_p95_ms": 1680.4417945290845 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.236075401306152, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 512.0, + "request_count": 64, + "request_metrics_sha256": "d879f833e5dc08d711f48fb6c6ccf48f7891fe7694470b107747ed0a1b95dac7", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.515625, + "passed": 33 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 975.5350224508372, + "ttft_p95_ms": 1782.0042945290845 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.188935279846191, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "ef8e3157f73a0421abb1d6c8ef036db2c3858b52a81c47c52bf2b24803674ab6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 145.3083076454922, + "ttft_p95_ms": 191.94509020206718 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.238474130630493, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "b3e01bb71b1a148c2586b6f5a9cfe05bb9b31e656f4e7993ebd27a4d14d6c994", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 160.71212013197345, + "ttft_p95_ms": 215.40746023330092 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.044495582580566, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "be6729dc2729694448aa8a59685d2a1cfd6863cdc40baa8d8aff7edc86633c63", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 342.3859670790913, + "ttft_p95_ms": 516.3686608843987 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.19176721572876, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "4a23d18f2b8a29fd0e341ab68c7499b93864e91aebf164a39a819450d728e7fc", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 568.5763491205233, + "ttft_p95_ms": 976.9340429258309 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.190044641494751, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "f1d8c6820f95946bcf3a4977585b99514ae35d870d2ff09bb9f7c7baef96464e", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.796875, + "passed": 51 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 670.1388491205233, + "ttft_p95_ms": 1211.309042925831 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.139874696731567, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "307d2d128d6032dd613bc67d5f78e0d3d9c9f05e7f1a72c5751fd1cec0b589b3", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 720.9200991205233, + "ttft_p95_ms": 1328.496542925831 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.338894128799438, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "ef8e3157f73a0421abb1d6c8ef036db2c3858b52a81c47c52bf2b24803674ab6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 145.3083076454922, + "ttft_p95_ms": 191.94509020206718 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.092449188232422, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "b3e01bb71b1a148c2586b6f5a9cfe05bb9b31e656f4e7993ebd27a4d14d6c994", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 160.71212013197345, + "ttft_p95_ms": 215.40746023330092 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.191263914108276, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "82720ff92616f5f6544e03f52ce3635328d72bd28cbfb26e18a0ad8f43e21abf", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 308.93499355358654, + "ttft_p95_ms": 484.23333144654856 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.09642219543457, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "e8dae4b0df880217f1c36c91435bf1afe920ccec1d776428611a7a49f3861d79", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 504.6855225446456, + "ttft_p95_ms": 731.2781755076838 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.189780473709106, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "172a6cae007a956ca9640e313a3c82b2ffbb8b19c05bb1db808d1f390e9c8cc3", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 565.5028311011398, + "ttft_p95_ms": 970.7870068870639 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.236796617507935, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "18b637ba04030fc15b1b39a20e3c651c829f4bd06bc908ac441ebb640653ef93", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 600.6590811011398, + "ttft_p95_ms": 1072.3495068870639 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.48096251487732, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "ef8e3157f73a0421abb1d6c8ef036db2c3858b52a81c47c52bf2b24803674ab6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 145.3083076454922, + "ttft_p95_ms": 191.94509020206718 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.192594051361084, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "b3e01bb71b1a148c2586b6f5a9cfe05bb9b31e656f4e7993ebd27a4d14d6c994", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 160.71212013197345, + "ttft_p95_ms": 215.40746023330092 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.440584182739258, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "82720ff92616f5f6544e03f52ce3635328d72bd28cbfb26e18a0ad8f43e21abf", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 308.93499355358654, + "ttft_p95_ms": 484.23333144654856 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.340449810028076, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "e8dae4b0df880217f1c36c91435bf1afe920ccec1d776428611a7a49f3861d79", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 504.6855225446456, + "ttft_p95_ms": 731.2781755076838 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.138880968093872, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "172a6cae007a956ca9640e313a3c82b2ffbb8b19c05bb1db808d1f390e9c8cc3", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 565.5028311011398, + "ttft_p95_ms": 970.7870068870639 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.097595930099487, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "18b637ba04030fc15b1b39a20e3c651c829f4bd06bc908ac441ebb640653ef93", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 600.6590811011398, + "ttft_p95_ms": 1072.3495068870639 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 8.998417615890503, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "ef8e3157f73a0421abb1d6c8ef036db2c3858b52a81c47c52bf2b24803674ab6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 145.3083076454922, + "ttft_p95_ms": 191.94509020206718 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.042184114456177, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "b3e01bb71b1a148c2586b6f5a9cfe05bb9b31e656f4e7993ebd27a4d14d6c994", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 160.71212013197345, + "ttft_p95_ms": 215.40746023330092 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.095743656158447, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "82720ff92616f5f6544e03f52ce3635328d72bd28cbfb26e18a0ad8f43e21abf", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 308.93499355358654, + "ttft_p95_ms": 484.23333144654856 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.151461124420166, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "e8dae4b0df880217f1c36c91435bf1afe920ccec1d776428611a7a49f3861d79", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 504.6855225446456, + "ttft_p95_ms": 731.2781755076838 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.000789642333984, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "172a6cae007a956ca9640e313a3c82b2ffbb8b19c05bb1db808d1f390e9c8cc3", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 565.5028311011398, + "ttft_p95_ms": 970.7870068870639 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 8.99738073348999, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 256.0, + "request_count": 64, + "request_metrics_sha256": "18b637ba04030fc15b1b39a20e3c651c829f4bd06bc908ac441ebb640653ef93", + "score": { + "slos": { + "tpot_100ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_150ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_180ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + }, + "tpot_50ms": { + "feasible": false, + "pass_rate": 0.875, + "passed": 56 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 600.6590811011398, + "ttft_p95_ms": 1072.3495068870639 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.145608186721802, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "cf7604290c74aa06e82791dccac8f57adbb43b861ec03601cf7353bcdb5cf85d", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 113.95995004232296, + "ttft_p95_ms": 148.6208545233651 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.04130220413208, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "667039850bf100a1214dff99ba4a6f15ff47f9d3986736dfa2460979e2e644e6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 123.08122913633257, + "ttft_p95_ms": 155.48122649863294 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.094721555709839, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "8fc12fe9c0c8ee1d2cae00e41bf57f36e90ecda35fed2e6c1b4fcaa52afb096b", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 132.5292253513205, + "ttft_p95_ms": 179.06372047949105 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.047569274902344, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a657986eb6c2f0500d902176c3344861ffca678da2498bca87bbaf0989097229", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 369.5552485735299, + "ttft_p95_ms": 599.7693411300143 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.150342464447021, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "7f8328f0bba2873fb721543cb20ef17fd1c43610e8adff9d26f4033157f86d92", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 471.1177485735299, + "ttft_p95_ms": 834.1443411300143 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.043598413467407, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "056ab2df7050c5080c7e03ba1bb9875d82fbcad2e8b1dc69b34a5c184fd89cb1", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 521.89899857353, + "ttft_p95_ms": 951.3318411300143 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.041609525680542, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "cf7604290c74aa06e82791dccac8f57adbb43b861ec03601cf7353bcdb5cf85d", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 113.95995004232296, + "ttft_p95_ms": 148.6208545233651 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.040033102035522, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "667039850bf100a1214dff99ba4a6f15ff47f9d3986736dfa2460979e2e644e6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 123.08122913633257, + "ttft_p95_ms": 155.48122649863294 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.14271068572998, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "8fc12fe9c0c8ee1d2cae00e41bf57f36e90ecda35fed2e6c1b4fcaa52afb096b", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 132.5292253513205, + "ttft_p95_ms": 179.06372047949105 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.04430103302002, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a1ad1f58cce44b8bc7b72b2997413758c4d40e971ce51fc68f9530bb2279d281", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 302.6860632179096, + "ttft_p95_ms": 411.8879773090427 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 8.998274326324463, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "6271e503f9158379f1b1043acab957d96c452adb195e45c63b311667677dfd75", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 381.3555701081787, + "ttft_p95_ms": 623.3699841993117 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.042869806289673, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "1bddc1f7a64764acb948087958ba9c20ed08b818545ccde67a3cec4cb48a9c35", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 416.5118201081787, + "ttft_p95_ms": 724.9324841993117 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.03968071937561, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "cf7604290c74aa06e82791dccac8f57adbb43b861ec03601cf7353bcdb5cf85d", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 113.95995004232296, + "ttft_p95_ms": 148.6208545233651 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.040271520614624, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "667039850bf100a1214dff99ba4a6f15ff47f9d3986736dfa2460979e2e644e6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 123.08122913633257, + "ttft_p95_ms": 155.48122649863294 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.089779615402222, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "8fc12fe9c0c8ee1d2cae00e41bf57f36e90ecda35fed2e6c1b4fcaa52afb096b", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 132.5292253513205, + "ttft_p95_ms": 179.06372047949105 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.046946287155151, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a1ad1f58cce44b8bc7b72b2997413758c4d40e971ce51fc68f9530bb2279d281", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 302.6860632179096, + "ttft_p95_ms": 411.8879773090427 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.149730682373047, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "6271e503f9158379f1b1043acab957d96c452adb195e45c63b311667677dfd75", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 381.3555701081787, + "ttft_p95_ms": 623.3699841993117 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.099130392074585, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "1bddc1f7a64764acb948087958ba9c20ed08b818545ccde67a3cec4cb48a9c35", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 416.5118201081787, + "ttft_p95_ms": 724.9324841993117 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.091226577758789, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "cf7604290c74aa06e82791dccac8f57adbb43b861ec03601cf7353bcdb5cf85d", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 113.95995004232296, + "ttft_p95_ms": 148.6208545233651 + }, + "status": "completed", + "trace_label": "r16", + "trace_sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.049186944961548, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "667039850bf100a1214dff99ba4a6f15ff47f9d3986736dfa2460979e2e644e6", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 123.08122913633257, + "ttft_p95_ms": 155.48122649863294 + }, + "status": "completed", + "trace_label": "r32", + "trace_sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.14399766921997, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "8fc12fe9c0c8ee1d2cae00e41bf57f36e90ecda35fed2e6c1b4fcaa52afb096b", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 132.5292253513205, + "ttft_p95_ms": 179.06372047949105 + }, + "status": "completed", + "trace_label": "r64", + "trace_sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.000115871429443, + "offered_request_rate": 128.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a1ad1f58cce44b8bc7b72b2997413758c4d40e971ce51fc68f9530bb2279d281", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 302.6860632179096, + "ttft_p95_ms": 411.8879773090427 + }, + "status": "completed", + "trace_label": "r128", + "trace_sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.093112468719482, + "offered_request_rate": 256.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "6271e503f9158379f1b1043acab957d96c452adb195e45c63b311667677dfd75", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 381.3555701081787, + "ttft_p95_ms": 623.3699841993117 + }, + "status": "completed", + "trace_label": "r256", + "trace_sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.14099645614624, + "offered_request_rate": 512.0, + "offered_request_rate_per_gpu": 128.0, + "request_count": 64, + "request_metrics_sha256": "1bddc1f7a64764acb948087958ba9c20ed08b818545ccde67a3cec4cb48a9c35", + "score": { + "slos": { + "tpot_100ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_150ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_180ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + }, + "tpot_50ms": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64 + } + }, + "tpot_p50_ms": null, + "tpot_p95_ms": null, + "ttft_p50_ms": 416.5118201081787, + "ttft_p95_ms": 724.9324841993117 + }, + "status": "completed", + "trace_label": "r512", + "trace_sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] + } + ], + "contract": { + "arrival": "original_trace_timestamp_and_order", + "input_output": "exact_source_values", + "prefix_caching": false, + "primary_tpot_slo_ms": 150.0, + "rate_contract": "uniform-spacing", + "target_pass_rate": 0.95, + "tpot_slos_ms": [ + 50.0, + 100.0, + 150.0, + 180.0 + ], + "ttft_slo": "1000ms + 1000ms * input_tokens / 8000", + "window_seconds": null + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/cc_backend/backends/vidur_cc_backend.py\n M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n?? tests/unit/test_vidur_cc_large_payload.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp1-20260717-v1-20260717T014550833983Z/artifacts/artifacts/attention-composition-tp1-v1/raw/flashattn-composition-tp1.json": "ce461be92a5059e2d168b2cbf21b55d9a55ee71c3b0586a3e63cd8288938168d", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp2-20260717-v1-20260717T014551831545Z/artifacts/artifacts/attention-composition-tp2-v1/raw/flashattn-composition-tp2.json": "997d67fb3a28b488c3aa4945ae7c6d4ecf8a08726ee3cb4652d16c221ff395ad", + "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/fleet-artifacts/qwen30-attention-composition-tp4-20260717-v1-20260717T014552897206Z/artifacts/artifacts/attention-composition-tp4-v1/raw/flashattn-composition-tp4.json": "046cffd4c11d8a36513167ec25d3e1ca016324715679ea688a58d4a71dfe62b2", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "03180eec28087a301c9d8c185874c50bdb28b31e5e9af35fcce5a6a5a9e62147", + "attention_true_mixed_fused.csv": "cccf76e6cb6a9359f188b07d4dc2942442568a6e8cf3ff3d77ab3d35a052cecb", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving-pure-prefill-batch-composition", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_pure_prefill_batch_composition": "Direct FA3 measurements for 2/4 requests at query length 2048 and 2/4/8/16 requests at query length 512 for each TP; included only when batch-composition attention inputs are supplied", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 150, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v3" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/profiles/profile-v3-batch-final", + "sha256": { + "attention": "03180eec28087a301c9d8c185874c50bdb28b31e5e9af35fcce5a6a5a9e62147", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "70aecb6d70f9e509e9816b1a0929163b9505760c2ac30bbd4cd8e51badfb7635", + "moe": "dd698113975e5ed77f51dfd8a69aebf764ed651c3f6f34b71914d25d6e001415" + } + }, + "rankings": { + "tpot_100ms": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 128.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 32.0, + "upper_censored": false + } + ], + "tpot_150ms": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 128.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 32.0, + "upper_censored": false + } + ], + "tpot_180ms": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 128.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 32.0, + "upper_censored": false + } + ], + "tpot_50ms": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 256.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": false + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 512.0, + "maximum_tested_feasible_request_rate_per_gpu": 128.0, + "upper_censored": true + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 64.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 128.0, + "maximum_tested_feasible_request_rate_per_gpu": 64.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 32.0, + "maximum_tested_feasible_request_rate_per_gpu": 32.0, + "upper_censored": false + } + ] + }, + "schema": "frontier-qwen30-exact-trace-surface-v1", + "status": "frozen_before_real", + "traces": [ + { + "first_arrival_s": 0.0, + "label": "r16", + "last_arrival_s": 3.9375, + "offered_request_rate": 16.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r16.csv", + "requests": 64, + "sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "first_arrival_s": 0.0, + "label": "r32", + "last_arrival_s": 1.96875, + "offered_request_rate": 32.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r32.csv", + "requests": 64, + "sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "first_arrival_s": 0.0, + "label": "r64", + "last_arrival_s": 0.984375, + "offered_request_rate": 64.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r64.csv", + "requests": 64, + "sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "first_arrival_s": 0.0, + "label": "r128", + "last_arrival_s": 0.4921875, + "offered_request_rate": 128.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r128.csv", + "requests": 64, + "sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "first_arrival_s": 0.0, + "label": "r256", + "last_arrival_s": 0.24609375, + "offered_request_rate": 256.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r256.csv", + "requests": 64, + "sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "first_arrival_s": 0.0, + "label": "r512", + "last_arrival_s": 0.123046875, + "offered_request_rate": 512.0, + "path": "/home/gahow/phd/aituner/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r512.csv", + "requests": 64, + "sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ] +} diff --git a/runs/frontier-fidelity-envelope-v1/run_flashattn_composition.sh b/runs/frontier-fidelity-envelope-v1/run_flashattn_composition.sh new file mode 100644 index 0000000..03afe43 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_flashattn_composition.sh @@ -0,0 +1,65 @@ +#!/usr/bin/env bash + +set -euo pipefail + +TP="${TP:?TP must be set to 1, 2, or 4}" +case "${TP}" in + 1|2|4) ;; + *) echo "ERROR: invalid TP=${TP}" >&2; exit 1 ;; +esac + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +CAMPAIGN_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" +PROFILE_SCRIPT="${CAMPAIGN_ROOT}/../frontier-qwen30-vllm020-profile-v1/profile_vllm020_flashattn.py" +LOG_DIR="${OUTPUT_ROOT}/logs" +PROVENANCE_DIR="${OUTPUT_ROOT}/provenance" +BATCH_SPECS=(2q512 4q512 8q512 16q512 2q2k 4q2k) + +mkdir -p "${LOG_DIR}" "${PROVENANCE_DIR}" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${LOG_DIR}/composition.log") 2>&1 + +if [[ -z "${CUDA_VISIBLE_DEVICES:-}" ]]; then + echo "ERROR: CUDA_VISIBLE_DEVICES must contain the fleet-allocated GPU" >&2 + exit 1 +fi +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=FlashAttention3 tp=${TP} batch_specs=${BATCH_SPECS[*]} profile_script=${PROFILE_SCRIPT} output=${OUTPUT_ROOT} expected_wall=3-8m hard_wall=900s hard_gpu_cap=0.25_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test -x "${VENV_ROOT}/bin/python" +test -f "${VLLM_SOURCE}/benchmarks/attention_benchmarks/runner.py" +test -f "${MODEL}/config.json" +test -f "${PROFILE_SCRIPT}" + +git -C "${CAMPAIGN_ROOT}/../.." rev-parse HEAD > "${PROVENANCE_DIR}/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${PROVENANCE_DIR}/vllm-source.commit" +sha256sum "${PROFILE_SCRIPT}" "${BASH_SOURCE[0]}" > "${PROVENANCE_DIR}/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" > "${PROVENANCE_DIR}/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total --format=csv,noheader > "${PROVENANCE_DIR}/gpus.csv" +printf '%s\n' "${BATCH_SPECS[@]}" > "${PROVENANCE_DIR}/batch-specs.txt" + +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" "${PROFILE_SCRIPT}" \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/flashattn-composition-tp${TP}.json" \ + --tp "${TP}" \ + --batch-specs "${BATCH_SPECS[@]}" \ + --warmup-iters 5 \ + --repeats 10 \ + --profile-kv-update + +test -s "${OUTPUT_ROOT}/raw/flashattn-composition-tp${TP}.json" +sha256sum "${OUTPUT_ROOT}/raw/flashattn-composition-tp${TP}.json" "${PROVENANCE_DIR}"/* > "${OUTPUT_ROOT}/artifacts.sha256" +nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FLASHATTN_COMPOSITION_COMPLETE tp=${TP} cases=${#BATCH_SPECS[@]}" diff --git a/runs/frontier-fidelity-envelope-v1/run_flashattn_long_context.sh b/runs/frontier-fidelity-envelope-v1/run_flashattn_long_context.sh new file mode 100755 index 0000000..5230e16 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_flashattn_long_context.sh @@ -0,0 +1,65 @@ +#!/usr/bin/env bash + +set -euo pipefail + +TP="${TP:?TP must be set to 1, 2, or 4}" +case "${TP}" in + 1|2|4) ;; + *) echo "ERROR: invalid TP=${TP}" >&2; exit 1 ;; +esac + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +CAMPAIGN_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" +PROFILE_SCRIPT="${CAMPAIGN_ROOT}/../frontier-qwen30-vllm020-profile-v1/profile_vllm020_flashattn.py" +BATCH_SPECS=( + q1s16k 8q1s16k 16q1s16k 32q1s16k 64q1s16k + q1s32k 8q1s32k 16q1s32k 32q1s32k 64q1s32k + q1s40k 8q1s40k 16q1s40k 32q1s40k 64q1s40k + q512_8q1s16k q1k_16q1s16k q2k_32q1s16k + q4k_16q1s32k q2k_32q1s32k q1k_64q1s32k +) + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/long-context.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=FlashAttention3 tp=${TP} batch_specs=${BATCH_SPECS[*]} profile_script=${PROFILE_SCRIPT} output=${OUTPUT_ROOT} expected_wall=1-5m hard_wall=900s hard_gpu_cap=0.25_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test -x "${VENV_ROOT}/bin/python" +test -f "${VLLM_SOURCE}/benchmarks/attention_benchmarks/runner.py" +test -f "${MODEL}/config.json" +test -f "${PROFILE_SCRIPT}" + +git -C "${CAMPAIGN_ROOT}/../.." rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum "${PROFILE_SCRIPT}" "${BASH_SOURCE[0]}" > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" +printf '%s\n' "${BATCH_SPECS[@]}" > "${OUTPUT_ROOT}/provenance/batch-specs.txt" + +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" "${PROFILE_SCRIPT}" \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/flashattn-long-context-tp${TP}.json" \ + --tp "${TP}" \ + --batch-specs "${BATCH_SPECS[@]}" \ + --warmup-iters 5 \ + --repeats 10 \ + --profile-kv-update + +test -s "${OUTPUT_ROOT}/raw/flashattn-long-context-tp${TP}.json" +sha256sum "${OUTPUT_ROOT}/raw/flashattn-long-context-tp${TP}.json" "${OUTPUT_ROOT}/provenance"/* > "${OUTPUT_ROOT}/artifacts.sha256" +nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FLASHATTN_LONG_CONTEXT_COMPLETE tp=${TP} cases=${#BATCH_SPECS[@]}" diff --git a/runs/frontier-fidelity-envelope-v1/run_frontier_qwen235_v020_surface.py b/runs/frontier-fidelity-envelope-v1/run_frontier_qwen235_v020_surface.py new file mode 100644 index 0000000..2f25871 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_frontier_qwen235_v020_surface.py @@ -0,0 +1,240 @@ +#!/usr/bin/env python3 +"""Run Frontier on the four-cell Qwen235 vLLM 0.20 latency surface.""" + +from __future__ import annotations + +import argparse +import importlib.util +import json +import os +import subprocess +import sys +import time +from dataclasses import asdict, dataclass +from pathlib import Path + + +HERE = Path(__file__).resolve().parent +Q30_RUNNER = HERE / "run_frontier_qwen30_exact_trace_surface.py" + + +def load_q30(): + spec = importlib.util.spec_from_file_location("q30_exact_surface", Q30_RUNNER) + if spec is None or spec.loader is None: + raise ImportError(Q30_RUNNER) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +Q30 = load_q30() +BASE = Q30.BASE +MODEL = "Qwen3-235B-A22B" + + +@dataclass(frozen=True) +class Config: + tp: int + mns: int + moe_tp: int + moe_ep: int + + @property + def name(self) -> str: + return f"tp{self.tp}_ep{self.moe_ep}_mns{self.mns}" + + +GRID = tuple( + Config(tp, mns, 4 if tp == 4 else 1, 1 if tp == 4 else 8) + for tp in (4, 8) + for mns in (64, 128) +) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--frontier-source", type=Path, required=True) + parser.add_argument("--replayserve-root", type=Path, required=True) + parser.add_argument("--profile-root", type=Path, required=True) + parser.add_argument("--python-deps", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument("--runtime-contract", type=Path, required=True) + parser.add_argument("--trace-tp", action="append", required=True, help="TP=PATH") + parser.add_argument("--config", action="append") + parser.add_argument("--prefix-caching", action=argparse.BooleanOptionalAction, default=True) + parser.add_argument("--allreduce-csv", type=Path, required=True) + parser.add_argument("--timeout-seconds", type=float, default=7200) + parser.add_argument("--resume", action="store_true") + return parser.parse_args() + + +def profile_paths(root: Path) -> dict[str, Path]: + paths = { + "linear": root / "linear_op.csv", + "attention": root / "attention.csv", + "moe": root / "moe.csv", + "linear_kernel": root / "linear_op_kernel_only.csv", + "attention_kernel": root / "attention_kernel_only.csv", + "moe_kernel": root / "moe_kernel_only.csv", + "manifest": root / "manifest.json", + } + missing = [str(path) for path in paths.values() if not path.is_file()] + if missing: + raise FileNotFoundError(missing) + manifest = json.loads(paths["manifest"].read_text()) + for name, path in paths.items(): + if name == "manifest": + continue + expected = manifest.get("outputs", {}).get(path.name) + if expected != BASE.sha256(path): + raise ValueError(f"profile hash mismatch: {path}") + return paths + + +def knobs(config: Config, paths: dict[str, Path], contract: dict, cache: Path, prefix: bool): + resolved = contract[config.name] + return { + "simulation_mode": "online", + "sys_arch": "co-location", + "num_replicas": 1, + "cluster_scheduler": "sticky_round_robin", + "model_name": MODEL, + "device": "h20", + "network_device": "h20_dgx", + "attn_tensor_parallel_size": config.tp, + "attn_data_parallel_size": 1, + "moe_tensor_parallel_size": config.moe_tp, + "moe_expert_parallel_size": config.moe_ep, + "num_pipeline_stages": 1, + "replica_scheduler": "vllm_v1", + "decode_cuda_graph_mode": "piecewise", + "batch_size_cap": config.mns, + "max_tokens_in_batch": 8192, + "long_prefill_token_threshold": 0, + "block_size": 16, + "num_blocks_mode": "explicit", + "num_blocks": int(resolved["num_gpu_blocks"]), + "gpu_memory_utilization": 0.80, + "non_kv_cache_overhead_bytes": 0, + "trace_max_tokens": 40960, + "cache_dir": str(cache / config.name), + "enable_prefix_caching": prefix, + "enable_dummy_mode": False, + "linear_op_input_file": str(paths["linear"]), + "atten_input_file": str(paths["attention"]), + "moe_input_file": str(paths["moe"]), + "linear_op_kernel_only_input_file": str(paths["linear_kernel"]), + "atten_kernel_only_input_file": str(paths["attention_kernel"]), + "moe_kernel_only_input_file": str(paths["moe_kernel"]), + "prediction_max_prefill_chunk_size": 8192, + "prediction_max_tokens_per_request": 40960, + "prediction_max_batch_size": max(int(v) for v in resolved["capture_sizes"]), + "no_cache": True, + } + + +def main() -> None: + args = parse_args() + for name in ("frontier_source", "replayserve_root", "profile_root", "python_deps", "output_root", "runtime_contract", "allreduce_csv"): + setattr(args, name, getattr(args, name).resolve()) + paths = profile_paths(args.profile_root) + contract = json.loads(args.runtime_contract.read_text())["configs"] + trace_by_tp = {} + for specification in args.trace_tp: + raw_tp, separator, path = specification.partition("=") + if not separator: + raise ValueError(f"trace-tp must be TP=PATH: {specification}") + tp = int(raw_tp) + trace_by_tp[tp] = Q30.parse_trace( + f"eval={path}", rate_contract="trace-window", prefix_caching=args.prefix_caching + ) + if set(trace_by_tp) != {4, 8}: + raise ValueError("trace-tp must provide exactly TP4 and TP8") + selected = list(GRID) + if args.config: + wanted = set(args.config) + selected = [config for config in GRID if config.name in wanted] + if {config.name for config in selected} != wanted: + raise ValueError(f"unknown configs: {wanted - {config.name for config in selected}}") + + builder = BASE.load_module( + "qwen235_frontier_builder", args.replayserve_root / "tools/run_frontier_sweep.py" + ) + environment = os.environ.copy() + environment.update( + { + "PYTHONPATH": ":".join([str(args.python_deps), str(args.frontier_source)]), + "CUDA_VISIBLE_DEVICES": "", + "NVIDIA_VISIBLE_DEVICES": "void", + "WANDB_DISABLED": "true", + "VIDUR_DISABLE_WANDB": "1", + "FRONTIER_LOG_LEVEL": "WARNING", + "PYTHONDONTWRITEBYTECODE": "1", + } + ) + results = [] + for config in selected: + config_knobs = knobs(config, paths, contract, args.output_root / "cache", args.prefix_caching) + for trace in (trace_by_tp[config.tp],): + run_dir = args.output_root / "runs" / config.name / trace["label"] + result_path = run_dir / "result.json" + if args.resume and result_path.is_file(): + results.append(json.loads(result_path.read_text())) + continue + run_dir.mkdir(parents=True, exist_ok=True) + command = builder.build_frontier_command( + python_bin="/usr/bin/python3", + trace_file=trace["path"], + metrics_root=run_dir / "metrics", + run_id=f"qwen235_v020_{config.name}_{trace['label']}", + knobs=config_knobs, + ) + command.extend(["--cudagraph_capture_sizes", *(str(v) for v in contract[config.name]["capture_sizes"])]) + command = BASE.configure_cc_command( + command, + backend="vidur", + allreduce_csv=args.allreduce_csv, + cache=args.output_root / "cc-cache", + ) + BASE.write_json(run_dir / "command.json", command) + started = time.time() + with (run_dir / "stdout.log").open("w") as stdout, (run_dir / "stderr.log").open("w") as stderr: + completed = subprocess.run( + command, + cwd=args.frontier_source, + env=environment, + stdout=stdout, + stderr=stderr, + timeout=args.timeout_seconds, + check=False, + ) + result = { + "config": asdict(config) | {"name": config.name}, + "trace": Q30.trace_manifest_entry(trace), + "elapsed_seconds": time.time() - started, + "returncode": completed.returncode, + } + if completed.returncode == 0: + metrics = Q30.find_request_metrics(run_dir) + result.update(status="completed", metrics=Q30.score(metrics, trace["shapes"]), request_metrics_sha256=BASE.sha256(metrics)) + else: + stderr_text = (run_dir / "stderr.log").read_text(errors="replace") + result.update(status="failed", failure_class=Q30.classify_frontier_failure(stderr_text)) + BASE.write_json(result_path, result) + results.append(result) + print(json.dumps({"config": config.name, "trace": trace["label"], "status": result["status"]}, sort_keys=True), flush=True) + + manifest = { + "schema": "qwen235-v020-frontier-latency-surface-v1", + "frontier_commit": subprocess.check_output(["git", "-C", str(args.frontier_source), "rev-parse", "HEAD"], text=True).strip(), + "profiles": {name: BASE.sha256(path) for name, path in paths.items()}, + "runtime_contract_sha256": BASE.sha256(args.runtime_contract), + "prefix_caching": args.prefix_caching, + "results": results, + } + BASE.write_json(args.output_root / "frontier_surface.json", manifest) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/run_frontier_qwen30_exact_trace_surface.py b/runs/frontier-fidelity-envelope-v1/run_frontier_qwen30_exact_trace_surface.py new file mode 100644 index 0000000..a2c39a0 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_frontier_qwen30_exact_trace_surface.py @@ -0,0 +1,734 @@ +#!/usr/bin/env python3 +"""Freeze Frontier's Qwen30 exact production-trace response surface.""" + +from __future__ import annotations + +import argparse +import csv +import importlib.util +import json +import math +import os +import subprocess +import sys +import time +from pathlib import Path +from typing import Any + + +TARGET_PASS_RATE = 0.95 +TPOT_SLOS_MS = (50.0, 100.0, 150.0, 180.0) +WINDOW_SECONDS = 600.0 +GRAPH_CAPTURE_SIZES_BY_MNS = { + 8: (1, 2, 4, 8, 16), + 16: (1, 2, 4, 8, 16, 24, 32), + 32: (1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64), + 64: (1, 2, 4, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120, 128), +} +REAL_NUM_BLOCKS_BY_CONFIG = { + (1, 8): 20137, + (1, 16): 20128, + (1, 32): 20108, + (1, 64): 20069, + (2, 8): 76639, + (2, 16): 76620, + (2, 32): 76583, + (2, 64): 76505, + (4, 8): 191930, + (4, 16): 191882, + (4, 32): 191786, + (4, 64): 191589, +} +KERNEL_DECODE_KV_CONTEXTS = (128, 1024, 2048, 4096, 8192, 16384, 32768, 40960) +BASE_RUNNER = ( + Path(__file__).resolve().parents[1] + / "frontier-phase-factorial-v0/run_frontier_qwen30_prefill_surface.py" +) + + +def load_base(): + spec = importlib.util.spec_from_file_location("qwen30_prefill_surface_base", BASE_RUNNER) + if spec is None or spec.loader is None: + raise ImportError(BASE_RUNNER) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +BASE = load_base() + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--frontier-source", type=Path, required=True) + parser.add_argument("--replayserve-root", type=Path, required=True) + parser.add_argument("--profile-root", type=Path, required=True) + parser.add_argument("--kernel-profile-root", type=Path) + parser.add_argument("--python-deps", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument( + "--trace", + action="append", + required=True, + help="Frozen trace anchor as LABEL=PATH; repeat in increasing load order.", + ) + parser.add_argument("--config", action="append") + parser.add_argument( + "--rate-contract", + choices=("trace-window", "uniform-spacing"), + default="trace-window", + ) + parser.add_argument( + "--prefix-caching", + action=argparse.BooleanOptionalAction, + default=True, + ) + parser.add_argument( + "--cc-backend", choices=("analytical", "vidur"), default="vidur" + ) + parser.add_argument("--allreduce-csv", type=Path) + parser.add_argument("--timeout-seconds", type=float, default=1800.0) + parser.add_argument("--predictor-training-job-threads", type=int, default=1) + parser.add_argument( + "--decode-cuda-graph-mode", + choices=("none", "full_decode_only", "piecewise"), + default="none", + ) + parser.add_argument( + "--align-real-graph-runtime", + action="store_true", + help="Use real observed capture lists and per-(TP,MNS) KV blocks.", + ) + parser.add_argument( + "--fresh-predictor-cache", + action="store_true", + help="Disable Frontier predictor cache reuse for this profile family.", + ) + parser.add_argument("--resume", action="store_true") + parser.add_argument("--continue-on-failure", action="store_true") + return parser.parse_args() + + +def ttft_limit_ms(input_tokens: int) -> float: + return 1000.0 + 1000.0 * input_tokens / 8000.0 + + +def percentile(values: list[float], fraction: float) -> float | None: + if not values: + return None + ordered = sorted(values) + return ordered[math.ceil(fraction * len(ordered)) - 1] + + +def manifest_offered_rate(path: Path) -> tuple[float | None, str | None]: + manifest_path = path.with_name("manifest.json") + if not manifest_path.is_file(): + return None, None + manifest = json.loads(manifest_path.read_text()) + if manifest.get("public_csv") != str(path): + return None, None + rate = manifest.get("global_offered_request_rate") + if not isinstance(rate, (int, float)) or not math.isfinite(rate) or rate <= 0: + raise ValueError(f"invalid global_offered_request_rate in {manifest_path}") + return float(rate), str(manifest_path) + + +def parse_trace( + specification: str, *, rate_contract: str = "trace-window", prefix_caching: bool +) -> dict[str, Any]: + if "=" not in specification: + raise ValueError(f"trace must be LABEL=PATH: {specification}") + label, raw_path = specification.split("=", 1) + if not label or "/" in label: + raise ValueError(f"invalid trace label: {label!r}") + path = Path(raw_path).resolve() + if not path.is_file(): + raise FileNotFoundError(path) + with path.open(newline="") as source: + rows = list(csv.DictReader(source)) + if not rows: + raise ValueError(f"empty trace: {path}") + required = { + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", + } + if not required.issubset(rows[0]): + raise ValueError(f"trace columns missing: {required - set(rows[0])}") + arrivals = [float(row["arrived_at"]) for row in rows] + if any(not math.isfinite(value) or value < 0 for value in arrivals): + raise ValueError(f"invalid arrival in {path}") + if any(right < left for left, right in zip(arrivals, arrivals[1:])): + raise ValueError(f"arrival order drift in {path}") + if rate_contract == "trace-window": + offered_request_rate, rate_manifest = manifest_offered_rate(path) + if offered_request_rate is None: + offered_request_rate = len(rows) / WINDOW_SECONDS + rate_source = "legacy_fixed_600_second_window" + else: + rate_source = f"manifest:{rate_manifest}" + elif rate_contract == "uniform-spacing": + if len(rows) < 2 or arrivals[-1] <= arrivals[0]: + raise ValueError("uniform-spacing traces require at least two arrivals") + intervals = [right - left for left, right in zip(arrivals, arrivals[1:])] + expected_interval = (arrivals[-1] - arrivals[0]) / (len(arrivals) - 1) + if any(abs(value - expected_interval) > 1e-9 for value in intervals): + raise ValueError(f"non-uniform fixed trace: {path}") + offered_request_rate = 1.0 / expected_interval + rate_source = "uniform_spacing" + else: + raise ValueError(f"unknown rate contract: {rate_contract}") + shapes = [ + (int(row["num_prefill_tokens"]), int(row["num_decode_tokens"])) + for row in rows + ] + if any(prefill <= 0 or decode <= 0 or prefill + decode > 40960 for prefill, decode in shapes): + raise ValueError(f"out-of-contract shape in {path}") + if prefix_caching: + for index, (row, (prefill, _)) in enumerate(zip(rows, shapes, strict=True)): + block_ids = [value for value in row["block_hash_ids"].split("|") if value] + if len(block_ids) != prefill // 16: + raise ValueError( + "Frontier prefix-cache trace must expose only complete " + f"16-token blocks: row={index}, ISL={prefill}, " + f"ids={len(block_ids)}, expected={prefill // 16}" + ) + return { + "label": label, + "path": path, + "sha256": BASE.sha256(path), + "requests": len(rows), + "offered_request_rate": offered_request_rate, + "offered_request_rate_source": rate_source, + "first_arrival_s": arrivals[0], + "last_arrival_s": arrivals[-1], + "shapes": shapes, + } + + +def find_request_metrics(run_dir: Path) -> Path: + matches = list((run_dir / "metrics").rglob("request_metrics.csv")) + if len(matches) != 1: + raise RuntimeError(f"expected one request_metrics.csv, got {matches}") + return matches[0] + + +def trace_manifest_entry(trace: dict[str, Any]) -> dict[str, Any]: + return { + key: str(value) if isinstance(value, Path) else value + for key, value in trace.items() + if key != "shapes" + } + + +def classify_frontier_failure(stderr: str) -> str: + if "Sequential simulation ended with non-empty scheduler state" in stderr: + return "scheduler_stall" + return "frontier_error" + + +def score(path: Path, expected_shapes: list[tuple[int, int]]) -> dict[str, Any]: + with path.open(newline="") as source: + rows = list(csv.DictReader(source)) + if len(rows) != len(expected_shapes): + raise ValueError(f"request count mismatch: {len(rows)} != {len(expected_shapes)}") + rows.sort(key=lambda row: int(row["Request Id"])) + request_metrics = [] + for index, (row, expected) in enumerate(zip(rows, expected_shapes, strict=True)): + if int(row["Request Id"]) != index: + raise ValueError("Frontier request ID/order drift") + shape = ( + int(float(row["request_num_prefill_tokens"])), + int(float(row["request_num_decode_tokens"])), + ) + if shape != expected: + raise ValueError(f"request shape drift at {index}: {shape} != {expected}") + ttft = float(row["ttft"]) + e2e = float(row["request_e2e_time"]) + tpot = (e2e - ttft) / (shape[1] - 1) if shape[1] > 1 else None + values = [ttft, e2e] + ([] if tpot is None else [tpot]) + if not all(math.isfinite(value) and value >= 0 for value in values): + raise ValueError(f"invalid latency at request {index}") + request_metrics.append( + { + "request_id": index, + "input_tokens": shape[0], + "output_tokens": shape[1], + "ttft_ms": ttft, + "ttft_limit_ms": ttft_limit_ms(shape[0]), + "tpot_ms": tpot, + "e2e_ms": e2e, + } + ) + slos = {} + for limit in TPOT_SLOS_MS: + passed = sum( + row["ttft_ms"] <= row["ttft_limit_ms"] + and (row["tpot_ms"] is None or row["tpot_ms"] <= limit) + for row in request_metrics + ) + pass_rate = passed / len(request_metrics) + slos[f"tpot_{int(limit)}ms"] = { + "passed": passed, + "pass_rate": pass_rate, + "feasible": pass_rate >= TARGET_PASS_RATE, + } + ttfts = [float(row["ttft_ms"]) for row in request_metrics] + tpots = [float(row["tpot_ms"]) for row in request_metrics if row["tpot_ms"] is not None] + e2es = [float(row["e2e_ms"]) for row in request_metrics] + def summary(values: list[float]) -> dict[str, float | None]: + if not values: + return {"mean": None, "p50": None, "p90": None, "p95": None} + return { + "mean": sum(values) / len(values), + "p50": percentile(values, 0.50), + "p90": percentile(values, 0.90), + "p95": percentile(values, 0.95), + } + + ttft_summary = summary(ttfts) + tpot_summary = summary(tpots) + e2e_summary = summary(e2es) + return { + **{f"ttft_{name}_ms": value for name, value in ttft_summary.items()}, + **{f"tpot_{name}_ms": value for name, value in tpot_summary.items()}, + **{f"e2e_{name}_ms": value for name, value in e2e_summary.items()}, + "slos": slos, + } + + +def kernel_profile_paths(root: Path) -> dict[str, Path]: + paths = { + "linear": root / "linear_op.csv", + "attention": root / "attention.csv", + "moe": root / "moe.csv", + "manifest": root / "manifest.json", + } + missing = [str(path) for path in paths.values() if not path.is_file()] + if missing: + raise FileNotFoundError(missing) + return paths + + +def validate_kernel_profile(paths: dict[str, Path]) -> dict[str, Any]: + manifest = json.loads(paths["manifest"].read_text()) + outputs = manifest.get("outputs", {}) + for filename, name in ( + ("linear_op.csv", "linear"), + ("attention.csv", "attention"), + ("moe.csv", "moe"), + ): + if outputs.get(filename) != BASE.sha256(paths[name]): + raise ValueError(f"kernel-only profile hash mismatch for {filename}") + + with paths["linear"].open(newline="") as source: + linear_rows = list(csv.DictReader(source)) + with paths["attention"].open(newline="") as source: + attention_rows = list(csv.DictReader(source)) + with paths["moe"].open(newline="") as source: + moe_rows = list(csv.DictReader(source)) + for label, rows in (("linear", linear_rows), ("attention", attention_rows), ("moe", moe_rows)): + if not rows or {row.get("measurement_type") for row in rows} != {"KERNEL_ONLY"}: + raise ValueError(f"{label} lacks an exclusive KERNEL_ONLY measurement family") + + required_buckets = set(GRAPH_CAPTURE_SIZES_BY_MNS[64]) + coverage: dict[str, Any] = {} + for tp in (1, 2, 4): + linear_tokens = { + int(float(row["num_tokens"])) + for row in linear_rows + if int(float(row["num_tensor_parallel_workers"])) == tp + } + moe_tokens = { + int(float(row["num_tokens"])) + for row in moe_rows + if int(float(row["num_tensor_parallel_workers"])) == tp + } + attention_pairs = { + (int(float(row["batch_size"])), int(float(row["kv_cache_size"]))) + for row in attention_rows + if int(float(row["num_tensor_parallel_workers"])) == tp + and row["is_prefill"].lower() == "false" + and row.get("is_true_mixed_batch", "").lower() != "true" + } + missing_linear = required_buckets - linear_tokens + missing_moe = required_buckets - moe_tokens + missing_attention = { + (bucket, kv) + for bucket in required_buckets + for kv in KERNEL_DECODE_KV_CONTEXTS + if (bucket, kv) not in attention_pairs + } + if missing_linear or missing_moe or missing_attention: + raise ValueError( + f"kernel-only profile coverage TP{tp}: linear={sorted(missing_linear)}, " + f"moe={sorted(missing_moe)}, attention={sorted(missing_attention)}" + ) + coverage[str(tp)] = { + "linear_tokens": sorted(linear_tokens), + "moe_tokens": sorted(moe_tokens), + "attention_decode_pairs": len(attention_pairs), + } + return {"manifest": manifest, "coverage": coverage} + + +def main() -> None: + args = parse_args() + if args.predictor_training_job_threads <= 0: + raise ValueError("predictor training job threads must be positive") + for name in ( + "frontier_source", + "replayserve_root", + "profile_root", + "python_deps", + "output_root", + ): + setattr(args, name, getattr(args, name).resolve()) + if args.allreduce_csv is not None: + args.allreduce_csv = args.allreduce_csv.resolve() + if args.kernel_profile_root is not None: + args.kernel_profile_root = args.kernel_profile_root.resolve() + if args.decode_cuda_graph_mode == "none": + raise ValueError("--kernel-profile-root requires a non-none graph mode") + traces = [ + parse_trace( + specification, + rate_contract=args.rate_contract, + prefix_caching=args.prefix_caching, + ) + for specification in args.trace + ] + if len({trace["label"] for trace in traces}) != len(traces): + raise ValueError("trace labels must be unique") + if any( + right["offered_request_rate"] <= left["offered_request_rate"] + for left, right in zip(traces, traces[1:]) + ): + raise ValueError("trace anchors must be supplied in increasing load order") + + selected = list(BASE.GRID) + if args.config: + wanted = set(args.config) + selected = [config for config in BASE.GRID if config.name in wanted] + if {config.name for config in selected} != wanted: + raise ValueError(f"unknown configs: {wanted - {config.name for config in selected}}") + paths = BASE.profile_paths(args.profile_root) + coverage = BASE.validate_profile(paths) + kernel_paths = None + kernel_coverage = None + if args.kernel_profile_root is not None: + kernel_paths = kernel_profile_paths(args.kernel_profile_root) + kernel_coverage = validate_kernel_profile(kernel_paths) + builder = BASE.load_module( + "qwen30_exact_trace_frontier_builder", + args.replayserve_root / "tools/run_frontier_sweep.py", + ) + frontier_head = subprocess.run( + ["git", "-C", str(args.frontier_source), "rev-parse", "HEAD"], + check=True, + text=True, + stdout=subprocess.PIPE, + ).stdout.strip() + environment = os.environ.copy() + pythonpath = [str(args.python_deps), str(args.frontier_source)] + if environment.get("PYTHONPATH"): + pythonpath.append(environment["PYTHONPATH"]) + environment.update( + { + "PYTHONPATH": ":".join(pythonpath), + "CUDA_VISIBLE_DEVICES": "", + "NVIDIA_VISIBLE_DEVICES": "void", + "WANDB_DISABLED": "true", + "VIDUR_DISABLE_WANDB": "1", + "FRONTIER_LOG_LEVEL": "WARNING", + "PYTHONDONTWRITEBYTECODE": "1", + } + ) + + config_results = [] + for config in selected: + loads = [] + config_knobs = BASE.knobs(config, paths, args.output_root / "cache") + config_knobs["enable_prefix_caching"] = args.prefix_caching + config_knobs["prediction_max_tokens_per_request"] = 40960 + config_knobs["decode_cuda_graph_mode"] = args.decode_cuda_graph_mode + config_knobs["no_cache"] = args.fresh_predictor_cache + if args.align_real_graph_runtime: + config_knobs["num_blocks"] = REAL_NUM_BLOCKS_BY_CONFIG[(config.tp, config.mns)] + # These are predictor-grid limits, not scheduler knobs. Keep the + # grid inside the observed vLLM graph/runtime envelope: no batch + # can use a capture bucket above this cell's list and MBT is 8192. + config_knobs["prediction_max_batch_size"] = max( + GRAPH_CAPTURE_SIZES_BY_MNS[config.mns] + ) + config_knobs["prediction_max_prefill_chunk_size"] = 8192 + if kernel_paths is not None: + config_knobs.update( + { + "linear_op_kernel_only_input_file": str(kernel_paths["linear"]), + "atten_kernel_only_input_file": str(kernel_paths["attention"]), + "moe_kernel_only_input_file": str(kernel_paths["moe"]), + } + ) + for trace in traces: + run_dir = args.output_root / "runs" / config.name / trace["label"] + result_path = run_dir / "result.json" + if args.resume and result_path.is_file(): + loads.append(json.loads(result_path.read_text())) + continue + run_dir.mkdir(parents=True, exist_ok=True) + command = builder.build_frontier_command( + python_bin="/usr/bin/python3", + trace_file=trace["path"], + metrics_root=run_dir / "metrics", + run_id=f"qwen30_trace_{config.name}_{trace['label']}", + knobs=config_knobs, + ) + command.extend( + [ + "--random_forrest_execution_time_predictor_config_num_training_job_threads", + str(args.predictor_training_job_threads), + ] + ) + if args.align_real_graph_runtime: + command.extend( + [ + "--cudagraph_capture_sizes", + *(str(size) for size in GRAPH_CAPTURE_SIZES_BY_MNS[config.mns]), + ] + ) + command = BASE.configure_cc_command( + command, + backend=args.cc_backend, + allreduce_csv=args.allreduce_csv, + cache=args.output_root / "cc-cache", + ) + BASE.write_json(run_dir / "command.json", command) + started = time.time() + with (run_dir / "stdout.log").open("w") as stdout, ( + run_dir / "stderr.log" + ).open("w") as stderr: + completed = subprocess.run( + command, + cwd=args.frontier_source, + env=environment, + stdout=stdout, + stderr=stderr, + timeout=args.timeout_seconds, + check=False, + ) + if completed.returncode != 0: + stderr_path = run_dir / "stderr.log" + result = { + "status": "frontier_failed", + "failure_kind": classify_frontier_failure( + stderr_path.read_text(errors="replace") + ), + "returncode": completed.returncode, + "config": { + "tp": config.tp, + "mns": config.mns, + "name": config.name, + }, + "trace_label": trace["label"], + "offered_request_rate": trace["offered_request_rate"], + "offered_request_rate_per_gpu": ( + trace["offered_request_rate"] / config.tp + ), + "request_count": trace["requests"], + "elapsed_seconds": time.time() - started, + "trace_sha256": trace["sha256"], + "stderr_sha256": BASE.sha256(stderr_path), + } + BASE.write_json(result_path, result) + loads.append(result) + print( + json.dumps( + { + "config": config.name, + "trace": trace["label"], + "status": result["status"], + "failure_kind": result["failure_kind"], + }, + sort_keys=True, + ), + flush=True, + ) + if not args.continue_on_failure: + raise RuntimeError( + f"Frontier failed for {config.name}/{trace['label']}: " + f"{completed.returncode}" + ) + continue + metrics = find_request_metrics(run_dir) + result = { + "status": "completed", + "config": {"tp": config.tp, "mns": config.mns, "name": config.name}, + "trace_label": trace["label"], + "offered_request_rate": trace["offered_request_rate"], + "offered_request_rate_per_gpu": trace["offered_request_rate"] / config.tp, + "request_count": trace["requests"], + "elapsed_seconds": time.time() - started, + "trace_sha256": trace["sha256"], + "request_metrics_sha256": BASE.sha256(metrics), + "score": score(metrics, trace["shapes"]), + } + BASE.write_json(result_path, result) + loads.append(result) + print( + json.dumps( + { + "config": config.name, + "trace": trace["label"], + "rate": trace["offered_request_rate"], + "primary": result["score"]["slos"]["tpot_150ms"], + }, + sort_keys=True, + ), + flush=True, + ) + config_results.append( + { + "config": {"tp": config.tp, "mns": config.mns, "name": config.name}, + "loads": loads, + } + ) + + rankings = {} + for slo in (f"tpot_{int(value)}ms" for value in TPOT_SLOS_MS): + records = [] + for item in config_results: + completed_loads = [ + load for load in item["loads"] if load["status"] == "completed" + ] + invalid_loads = [ + load for load in item["loads"] if load["status"] != "completed" + ] + feasible = [ + load["offered_request_rate"] + for load in completed_loads + if load["score"]["slos"][slo]["feasible"] + ] + capacity = max(feasible, default=None) + records.append( + { + "config": item["config"], + "ranking_valid": not invalid_loads, + "invalid_loads": [ + { + "trace_label": load["trace_label"], + "failure_kind": load.get("failure_kind", "unknown"), + } + for load in invalid_loads + ], + "maximum_tested_feasible_request_rate": capacity, + "maximum_tested_feasible_request_rate_per_gpu": ( + capacity / item["config"]["tp"] if capacity is not None else None + ), + "lower_censored": capacity is None and not invalid_loads, + "upper_censored": ( + capacity == traces[-1]["offered_request_rate"] + and not invalid_loads + ), + } + ) + records.sort( + key=lambda row: ( + -( + row["maximum_tested_feasible_request_rate_per_gpu"] + if row["maximum_tested_feasible_request_rate_per_gpu"] is not None + else -1 + ), + row["config"]["name"], + ) + ) + rankings[slo] = records + + has_invalid_cells = any( + load["status"] != "completed" + for item in config_results + for load in item["loads"] + ) + if selected != list(BASE.GRID): + manifest_status = "partial_not_decision_bearing" + elif has_invalid_cells: + manifest_status = "frozen_with_invalid_cells" + else: + manifest_status = "frozen_before_real" + manifest = { + "schema": "frontier-qwen30-exact-trace-surface-v1", + "status": manifest_status, + "contract": { + "window_seconds": ( + WINDOW_SECONDS if args.rate_contract == "trace-window" else None + ), + "rate_contract": args.rate_contract, + "prefix_caching": args.prefix_caching, + "arrival": "original_trace_timestamp_and_order", + "input_output": "exact_source_values", + "ttft_slo": "1000ms + 1000ms * input_tokens / 8000", + "tpot_slos_ms": TPOT_SLOS_MS, + "primary_tpot_slo_ms": 150.0, + "target_pass_rate": TARGET_PASS_RATE, + "predictor_training_job_threads": args.predictor_training_job_threads, + "decode_cuda_graph_mode": args.decode_cuda_graph_mode, + "align_real_graph_runtime": args.align_real_graph_runtime, + "fresh_predictor_cache": args.fresh_predictor_cache, + }, + "frontier": { + "source": str(args.frontier_source), + "git_head": frontier_head, + "git_status_short": subprocess.run( + ["git", "-C", str(args.frontier_source), "status", "--short"], + check=True, + text=True, + stdout=subprocess.PIPE, + ).stdout, + }, + "profiles": { + "root": str(args.profile_root), + "coverage": coverage, + "sha256": {name: BASE.sha256(path) for name, path in paths.items()}, + }, + "kernel_only_profiles": ( + None + if kernel_paths is None + else { + "root": str(args.kernel_profile_root), + "coverage": kernel_coverage, + "sha256": { + name: BASE.sha256(path) for name, path in kernel_paths.items() + }, + } + ), + "runtime_alignment": { + "capture_sizes_by_mns": ( + GRAPH_CAPTURE_SIZES_BY_MNS if args.align_real_graph_runtime else None + ), + "num_blocks_by_config": ( + { + f"tp{tp}_mns{mns}": blocks + for (tp, mns), blocks in REAL_NUM_BLOCKS_BY_CONFIG.items() + } + if args.align_real_graph_runtime + else None + ), + }, + "collective": { + "backend": args.cc_backend, + "allreduce_csv": str(args.allreduce_csv) if args.allreduce_csv else None, + "allreduce_csv_sha256": BASE.sha256(args.allreduce_csv) if args.allreduce_csv else None, + }, + "traces": [trace_manifest_entry(trace) for trace in traces], + "config_results": config_results, + "rankings": rankings, + } + BASE.write_json(args.output_root / "frontier_trace_surface_frozen.json", manifest) + print(args.output_root / "frontier_trace_surface_frozen.json") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_attention.sh b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_attention.sh new file mode 100644 index 0000000..4d6ca07 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_attention.sh @@ -0,0 +1,75 @@ +#!/usr/bin/env bash + +set -euo pipefail + +TP="${TP:?TP must be 1, 2, or 4}" +case "${TP}" in + 1|2|4) ;; + *) echo "invalid TP=${TP}" >&2; exit 1 ;; +esac + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +RUN_DIR="$(pwd -P)" +PROFILE_DIR="${PROFILE_DIR:-${RUN_DIR%/runs/frontier-fidelity-envelope-v1}/runs/frontier-qwen30-vllm020-profile-v1}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +PROFILE_HOME="${PROFILE_HOME:-/tmp/wjh}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-t1-dash0-deadc4a}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +CAPTURE_BUCKETS="${CAPTURE_BUCKETS:-1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128}" +KV_CONTEXTS="${KV_CONTEXTS:-128 1024 2048 4096 8192 16384 32768 40960}" +WARMUP_ITERS="${WARMUP_ITERS:-3}" +REPEATS="${REPEATS:-5}" +export HOME="${PROFILE_HOME}" +export XDG_CACHE_HOME="${PROFILE_HOME}/.cache" +export VLLM_CACHE_ROOT="${XDG_CACHE_HOME}/vllm" + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/attention-tp${TP}.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?a fleet-allocated GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +BATCH_SPECS=() +for bucket in ${CAPTURE_BUCKETS}; do + prefix="" + if [[ "${bucket}" -ne 1 ]]; then + prefix="${bucket}" + fi + for context in ${KV_CONTEXTS}; do + BATCH_SPECS+=("${prefix}q1s${context}") + done +done + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} role=FA3-decode-kernel-only tp=${TP} buckets='${CAPTURE_BUCKETS}' kv_contexts='${KV_CONTEXTS}' method=Frontier-RecordFunctionTracer output=${OUTPUT_ROOT} expected_wall=10-35m expected_gpu_cap=1.0_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader +test "$(git -C "${FRONTIER_SOURCE}" rev-parse HEAD)" = "deadc4a321f0baaa534c6ebd17f974123733cdc2" +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +test -f "${MODEL_ROOT}/config.json" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm.commit" +printf '%s\n' "${BATCH_SPECS[@]}" > "${OUTPUT_ROOT}/provenance/batch-specs.txt" +sha256sum "${PROFILE_DIR}/profile_vllm020_flashattn.py" "${RUN_DIR}/run_graph_kernel_only_attention.sh" > "${OUTPUT_ROOT}/provenance/source.sha256" + +timeout --signal=TERM --kill-after=30s 2400 \ + "${VENV_ROOT}/bin/python" "${PROFILE_DIR}/profile_vllm020_flashattn.py" \ + --vllm-source "${VLLM_SOURCE}" \ + --frontier-source "${FRONTIER_SOURCE}" \ + --model "${MODEL_ROOT}" \ + --output "${OUTPUT_ROOT}/raw/attention-tp${TP}.json" \ + --tp "${TP}" \ + --batch-specs "${BATCH_SPECS[@]}" \ + --warmup-iters "${WARMUP_ITERS}" \ + --repeats "${REPEATS}" \ + --profile-kv-update \ + --profile-method record_function + +test -s "${OUTPUT_ROOT}/raw/attention-tp${TP}.json" +sha256sum "${OUTPUT_ROOT}/raw/attention-tp${TP}.json" "${OUTPUT_ROOT}/provenance"/* > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "GRAPH_KERNEL_ONLY_ATTENTION_COMPLETE tp=${TP} rows=${#BATCH_SPECS[@]}" diff --git a/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_linear.sh b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_linear.sh new file mode 100644 index 0000000..5764f97 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_linear.sh @@ -0,0 +1,48 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +RUN_DIR="$(pwd -P)" +PROFILE_DIR="${PROFILE_DIR:-${RUN_DIR%/runs/frontier-fidelity-envelope-v1}/runs/frontier-qwen30-vllm020-profile-v1}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +PROFILE_HOME="${PROFILE_HOME:-/tmp/wjh}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-t1-dash0-deadc4a}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +TOKENS_TEXT="${TOKENS:-1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128}" +read -r -a TOKENS <<< "${TOKENS_TEXT}" +export MODEL_ROOT +export HOME="${PROFILE_HOME}" +export XDG_CACHE_HOME="${PROFILE_HOME}/.cache" +export VLLM_CACHE_ROOT="${XDG_CACHE_HOME}/vllm" + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/profiles" +exec > >(tee -a "${OUTPUT_ROOT}/logs/linear.log") 2>&1 +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?a fleet-allocated GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} role=linear-kernel-only tp=1,2,4 tokens='${TOKENS[*]}' method=Frontier-RecordFunctionTracer output=${OUTPUT_ROOT} expected_wall=15-35m expected_gpu_cap=1.0_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +test "$(git -C "${FRONTIER_SOURCE}" rev-parse HEAD)" = "deadc4a321f0baaa534c6ebd17f974123733cdc2" +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +sha256sum "${PROFILE_DIR}/frontier_vllm020_compat.py" "${RUN_DIR}/run_graph_kernel_only_linear.sh" > "${OUTPUT_ROOT}/provenance/source.sha256" + +cd "${FRONTIER_SOURCE}" +timeout --signal=TERM --kill-after=30s 2400 \ + "${VENV_ROOT}/bin/python" "${PROFILE_DIR}/frontier_vllm020_compat.py" \ + --disable_ray --num_gpus 1 --output_dir "${OUTPUT_ROOT}/profiles" \ + --device h20 --models qwen3-a3b-30b-moe \ + --num_tensor_parallel_workers 1 2 4 --max_tokens 128 \ + --num_tokens_list "${TOKENS[@]}" --profile_method record_function \ + --precision BF16 --is_moe --yes + +find "${OUTPUT_ROOT}/profiles" -name linear_op_kernel_only.csv -type f -size +0c -print -quit > "${OUTPUT_ROOT}/provenance/linear-path.txt" +test -s "${OUTPUT_ROOT}/provenance/linear-path.txt" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "GRAPH_KERNEL_ONLY_LINEAR_COMPLETE" diff --git a/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_moe.sh b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_moe.sh new file mode 100644 index 0000000..27bb472 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_graph_kernel_only_moe.sh @@ -0,0 +1,52 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +RUN_DIR="$(pwd -P)" +PROFILE_DIR="${PROFILE_DIR:-${RUN_DIR%/runs/frontier-fidelity-envelope-v1}/runs/frontier-qwen30-vllm020-profile-v1}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +PROFILE_HOME="${PROFILE_HOME:-/tmp/wjh}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-t1-dash0-deadc4a}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +TOKENS="${TOKENS:-1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128}" +export HOME="${PROFILE_HOME}" +export XDG_CACHE_HOME="${PROFILE_HOME}/.cache" +export VLLM_CACHE_ROOT="${XDG_CACHE_HOME}/vllm" + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/moe-router.log") 2>&1 +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?a fleet-allocated GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} role=MoE+router-kernel-only tp=1,2,4 tokens='${TOKENS}' method=Frontier-RecordFunctionTracer output=${OUTPUT_ROOT} expected_wall=10-30m expected_gpu_cap=1.0_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +test "$(git -C "${FRONTIER_SOURCE}" rev-parse HEAD)" = "deadc4a321f0baaa534c6ebd17f974123733cdc2" +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +printf '%s\n' ${TOKENS} > "${OUTPUT_ROOT}/provenance/tokens.txt" +sha256sum "${PROFILE_DIR}/profile_vllm020_moe.py" "${PROFILE_DIR}/profile_vllm020_router.py" "${RUN_DIR}/run_graph_kernel_only_moe.sh" > "${OUTPUT_ROOT}/provenance/source.sha256" + +timeout --signal=TERM --kill-after=30s 2400 \ + "${VENV_ROOT}/bin/python" "${PROFILE_DIR}/profile_vllm020_moe.py" \ + --vllm-source "${VLLM_SOURCE}" --frontier-source "${FRONTIER_SOURCE}" \ + --model "${MODEL_ROOT}" --output "${OUTPUT_ROOT}/raw/moe.json" \ + --tp 1 2 4 --num-tokens ${TOKENS} --routing-modes uniform_random_logits \ + --warmup-iters 3 --repeats 5 --profile-method record_function --check-reference + +timeout --signal=TERM --kill-after=30s 1800 \ + "${VENV_ROOT}/bin/python" "${PROFILE_DIR}/profile_vllm020_router.py" \ + --vllm-source "${VLLM_SOURCE}" --frontier-source "${FRONTIER_SOURCE}" \ + --model "${MODEL_ROOT}" --output "${OUTPUT_ROOT}/raw/router.json" \ + --num-tokens ${TOKENS} --warmup-iters 3 --repeats 5 --profile-method record_function + +test -s "${OUTPUT_ROOT}/raw/moe.json" +test -s "${OUTPUT_ROOT}/raw/router.json" +sha256sum "${OUTPUT_ROOT}/raw"/*.json "${OUTPUT_ROOT}/provenance"/* > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "GRAPH_KERNEL_ONLY_MOE_COMPLETE" diff --git a/runs/frontier-fidelity-envelope-v1/run_q235_after_q30.sh b/runs/frontier-fidelity-envelope-v1/run_q235_after_q30.sh new file mode 100644 index 0000000..90fe528 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_q235_after_q30.sh @@ -0,0 +1,48 @@ +#!/usr/bin/env bash + +set -euo pipefail + +Q30_ROOT="${Q30_ROOT:?Q30_ROOT is required}" +Q30_SESSION="${Q30_SESSION:-q30_fixed_pressure_20260719}" +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +PROFILE_ROOT="${PROFILE_ROOT:?PROFILE_ROOT is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-q235-v020-5b953f5}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" + +# FlashInfer's TensorRT-LLM MoE runtime invokes `nvcc` by name when it +# autotunes a previously unseen FP8 grouped-GEMM shape. +export PATH="/usr/local/cuda/bin:${PATH}" +command -v nvcc >/dev/null + +mkdir -p "${OUTPUT_ROOT}/supervisor" "${PROFILE_ROOT}" +exec > >(tee -a "${OUTPUT_ROOT}/supervisor/controller.log") 2>&1 +echo "Q235_DEFERRED_LAUNCH_ECHO dependency=${Q30_ROOT}:Q30_FIXED_PRESSURE_CAMPAIGN_COMPLETE profile_model=Qwen3-235B-A22B-FP8 profile_backends={TP4/EP1:Triton,TP8/EP8:FlashInfer-CUTLASS} profile_cost=2-7_H20-GPUh experiment_cases={Fixed-PD,Fixed-PO,Trace-PD,Trace-PO} configs={TP4/EP1,TP8/EP8}xMNS{64,128} requests=129 trials=3 expected_campaign_wall=10-30h expected_campaign_cost=90-220_H20-GPUh profile_output=${PROFILE_ROOT} campaign_output=${OUTPUT_ROOT}" +date -u +SUPERVISOR_START_UTC=%Y-%m-%dT%H:%M:%SZ + +while tmux has-session -t "${Q30_SESSION}" 2>/dev/null; do + printf 'WAIT_Q30_UTC=%s results=%s\n' "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \ + "$(find "${Q30_ROOT}/real" -path '*/results/result.json' 2>/dev/null | wc -l)" + sleep 30 +done +grep -q 'Q30_FIXED_PRESSURE_CAMPAIGN_COMPLETE' "${Q30_ROOT}/controller.log" || { + echo 'ERROR: Q30 session ended without completion marker; refusing Q235 launch' >&2 + exit 1 +} +nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | awk '$1 > 16 {exit 1}' + +mkdir -p "${OUTPUT_ROOT}/profile-smoke" +env CUDA_VISIBLE_DEVICES=0 PYTHONPATH="${FRONTIER_SOURCE}" \ + "${VENV_ROOT}/bin/python" "${RUNNER_DIR}/smoke_qwen235_v020_frontier_moe.py" \ + --frontier-source "${FRONTIER_SOURCE}" \ + --output "${OUTPUT_ROOT}/profile-smoke/moe.json" + +OUTPUT_ROOT="${PROFILE_ROOT}" FRONTIER_SOURCE="${FRONTIER_SOURCE}" \ + VENV_ROOT="${VENV_ROOT}" RUNNER_DIR="${RUNNER_DIR}" \ + bash "${RUNNER_DIR}/run_qwen235_v020_profiles.sh" + +CAMPAIGN_ROOT="${OUTPUT_ROOT}" PROFILE_ROOT="${PROFILE_ROOT}/frozen" \ + FRONTIER_SOURCE="${FRONTIER_SOURCE}" VENV_ROOT="${VENV_ROOT}" RUNNER_DIR="${RUNNER_DIR}" \ + bash "${RUNNER_DIR}/run_qwen235_v020_campaign.sh" +date -u +SUPERVISOR_END_UTC=%Y-%m-%dT%H:%M:%SZ +echo Q235_AFTER_Q30_COMPLETE diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_campaign.sh b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_campaign.sh new file mode 100644 index 0000000..0663035 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_campaign.sh @@ -0,0 +1,86 @@ +#!/usr/bin/env bash + +set -euo pipefail + +CAMPAIGN_ROOT="${CAMPAIGN_ROOT:?CAMPAIGN_ROOT is required}" +PROFILE_ROOT="${PROFILE_ROOT:?PROFILE_ROOT is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-q235-v020-5b953f5}" +REPLAYSERVE_ROOT="${REPLAYSERVE_ROOT:-/home/admin/cpfs/wjh/replayserve}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8}" +PYTHON_DEPS="${PYTHON_DEPS:-${VENV_ROOT}/lib/python3.12/site-packages}" +ALLREDUCE_CSV="${ALLREDUCE_CSV:-${RUNNER_DIR}/profiles/measured-allreduce.csv}" +BASE_PUBLIC=/home/admin/cpfs/wjh/aituner/fidelity-envelope-private/trace-exact-v1/public/u0p01/frontier.csv +BASE_PRIVATE=/home/admin/cpfs/wjh/aituner/fidelity-envelope-private/trace-exact-v1/private/u0p01/real_requests.jsonl +FIXED_PER_GPU_RATE="${FIXED_PER_GPU_RATE:-0.2}" +REQUESTS=129 + +mkdir -p "${CAMPAIGN_ROOT}"/{provenance,traces,real,sim,analysis} +exec > >(tee -a "${CAMPAIGN_ROOT}/controller.log") 2>&1 +echo "Q235_CAMPAIGN_LAUNCH_ECHO host=dash0 model=Qwen3-235B-A22B-FP8 engine=vLLM-0.20.0+cu129 simulator=Frontier-5b953f5-piecewise cases={Fixed-PD:4096x256@0.2rps/gpu,Fixed-PO:4096x1@0.2rps/gpu,Trace-PD:u0p01-original-OSL,Trace-PO:u0p01-OSL1} requests=129 transform=t_prime=t/TP configs={TP4/EP1,TP8/EP8}xMNS{64,128} MBT=8192 trials=3 fresh_server=true metrics=mean,p90(TTFT,TPOT-if-PD,E2E) SLO=not_scored expected_wall=10-30h expected_cost=90-220_H20-GPUh profile=${PROFILE_ROOT} output=${CAMPAIGN_ROOT}" +date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ +sha256sum "${BASH_SOURCE[0]}" "${BASE_PUBLIC}" "${BASE_PRIVATE}" "${MODEL_ROOT}/config.json" \ + "${PROFILE_ROOT}/manifest.json" "${ALLREDUCE_CSV}" > "${CAMPAIGN_ROOT}/provenance/input.sha256" +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${CAMPAIGN_ROOT}/provenance/frontier.commit" +git -C "${RUNNER_DIR}" rev-parse HEAD > "${CAMPAIGN_ROOT}/provenance/aituner.commit" + +NORMALIZER="${RUNNER_DIR}/../simulator-tuning-latency-matrix-v0/materialize_qwen30_tp_normalized_trace.py" +MATERIALIZER="${RUNNER_DIR}/prepare_qwen30_latency_case.py" +for tp in 4 8; do + "${VENV_ROOT}/bin/python" "${NORMALIZER}" --base-public-csv "${BASE_PUBLIC}" \ + --base-private-jsonl "${BASE_PRIVATE}" --tp "${tp}" \ + --output-root "${CAMPAIGN_ROOT}/traces/trace-pd/tp${tp}" + "${VENV_ROOT}/bin/python" "${MATERIALIZER}" trace \ + --base-public "${CAMPAIGN_ROOT}/traces/trace-pd/tp${tp}/public/frontier.csv" \ + --base-private "${CAMPAIGN_ROOT}/traces/trace-pd/tp${tp}/private/real_requests.jsonl" \ + --output-tokens 1 --tp "${tp}" --output-root "${CAMPAIGN_ROOT}/traces/trace-po/tp${tp}" + for spec in fixed-pd:256 fixed-po:1; do + case_name="${spec%%:*}"; osl="${spec##*:}" + "${VENV_ROOT}/bin/python" "${MATERIALIZER}" fixed --model "${MODEL_ROOT}" \ + --input-tokens 4096 --output-tokens "${osl}" --requests "${REQUESTS}" \ + --per-gpu-rate "${FIXED_PER_GPU_RATE}" --tp "${tp}" \ + --output-root "${CAMPAIGN_ROOT}/traces/${case_name}/tp${tp}" + done +done + +run_real() { + local case_name="$1" prefix="$2" port="$3" + CASE_NAME="${case_name}" PREFIX_CACHING="${prefix}" \ + TRACE_ROOT="${CAMPAIGN_ROOT}/traces/${case_name}" OUTPUT_ROOT="${CAMPAIGN_ROOT}/real/${case_name}" \ + RUNNER_DIR="${RUNNER_DIR}" VENV_ROOT="${VENV_ROOT}" MODEL_ROOT="${MODEL_ROOT}" \ + BASE_PORT="${port}" bash "${RUNNER_DIR}/run_qwen235_v020_real_surface.sh" +} +run_real fixed-pd false 9300 +run_real fixed-po false 9400 +run_real trace-pd true 9500 +run_real trace-po true 9600 + +"${VENV_ROOT}/bin/python" "${RUNNER_DIR}/extract_qwen235_v020_runtime_contract.py" \ + --case-root "${CAMPAIGN_ROOT}/real/fixed-pd" \ + --output "${CAMPAIGN_ROOT}/provenance/runtime-contract.json" + +run_sim() { + local case_name="$1" prefix_flag="$2" + "${VENV_ROOT}/bin/python" "${RUNNER_DIR}/run_frontier_qwen235_v020_surface.py" \ + --frontier-source "${FRONTIER_SOURCE}" --replayserve-root "${REPLAYSERVE_ROOT}" \ + --profile-root "${PROFILE_ROOT}" --python-deps "${PYTHON_DEPS}" \ + --output-root "${CAMPAIGN_ROOT}/sim/${case_name}" \ + --runtime-contract "${CAMPAIGN_ROOT}/provenance/runtime-contract.json" \ + --trace-tp "4=${CAMPAIGN_ROOT}/traces/${case_name}/tp4/public/frontier.csv" \ + --trace-tp "8=${CAMPAIGN_ROOT}/traces/${case_name}/tp8/public/frontier.csv" \ + "${prefix_flag}" --allreduce-csv "${ALLREDUCE_CSV}" --resume +} +run_sim fixed-pd --no-prefix-caching +run_sim fixed-po --no-prefix-caching +run_sim trace-pd --prefix-caching +run_sim trace-po --prefix-caching + +"${VENV_ROOT}/bin/python" "${RUNNER_DIR}/analyze_qwen235_v020_campaign.py" \ + --campaign-root "${CAMPAIGN_ROOT}" \ + --json-output "${CAMPAIGN_ROOT}/analysis/comparison.json" \ + --markdown-output "${CAMPAIGN_ROOT}/analysis/comparison.md" +find "${CAMPAIGN_ROOT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 \ + | sort -z | xargs -0 sha256sum > "${CAMPAIGN_ROOT}/provenance/artifacts.sha256" +date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ +echo Q235_V020_CAMPAIGN_COMPLETE diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh new file mode 100644 index 0000000..ed775b3 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_profiles.sh @@ -0,0 +1,72 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-q235-v020-5b953f5}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +MODEL=Qwen3-235B-A22B +TOKENS=(1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128 136 144 152 160 168 176 184 192 200 208 216 224 232 240 248 256 512 1024 2048 4096 8192) +BATCHES=(1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128 136 144 152 160 168 176 184 192 200 208 216 224 232 240 248 256) +KV=(128 1024 2048 4096 8192 16384 32768 40960) + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" +exec > >(tee -a "${OUTPUT_ROOT}/controller.log") 2>&1 + +echo "Q235_PROFILE_LAUNCH_ECHO host=dash0 model=${MODEL} vllm=0.20.0 frontier=$(git -C "${FRONTIER_SOURCE}" rev-parse HEAD) device=H20 methods={CUDA_EVENT,KERNEL_ONLY} attention_tp={4,8} moe_paths={TP4/EP1:Triton,TP1/EP8:FlashInfer-CUTLASS} token_points=${#TOKENS[@]} batch_points=${#BATCHES[@]} kv_points=${#KV[@]} parallel_gpus=6 expected_wall=20-75m expected_cost=2-7_H20-GPUh output=${OUTPUT_ROOT}" +date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ +nvidia-smi --query-gpu=index,memory.used --format=csv,noheader,nounits \ + | awk '$2 > 16 {exit 1}' +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +"${VENV_ROOT}/bin/vllm" --version > "${OUTPUT_ROOT}/provenance/vllm.version" +# Frontier's model-config loader resolves data/config/models from the current +# source checkout, so all profiling entrypoints must run from this directory. +cd "${FRONTIER_SOURCE}" + +common_profile() { + local gpu="$1" method="$2" root="$3" + env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \ + "${VENV_ROOT}/bin/python" -m frontier.profiling.linear_op.main \ + --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \ + --num_tensor_parallel_workers 1 4 8 --attn_tp 4 8 --ffn_tp 1 4 \ + --max_tokens 8192 --num_tokens_list "${TOKENS[@]}" \ + --profile_method "${method}" --yes + env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \ + "${VENV_ROOT}/bin/python" -m frontier.profiling.attention.main \ + --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \ + --num_tensor_parallel_workers 4 8 --max_model_len 40960 --max_seq_len 8192 \ + --batch_size_list "${BATCHES[@]}" --decode_kv_cache_size_list "${KV[@]}" \ + --fixed_chunked_prefill_size 8192 --attention_backend FLASHINFER \ + --profile_method "${method}" --yes +} + +moe_profile() { + local gpu="$1" method="$2" tp="$3" ep="$4" root="$5" + env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \ + "${VENV_ROOT}/bin/python" -m frontier.profiling.moe.main \ + --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \ + --num_tensor_parallel_workers "${tp}" --expert_parallel_sizes "${ep}" \ + --max_tokens 8192 --num_tokens_list "${TOKENS[@]}" --load_distributions uniform \ + --num_samples_per_distribution 1 --profile_method "${method}" --yes +} + +declare -a pids=() +common_profile 0 cuda_event "${OUTPUT_ROOT}/cuda-common" > "${OUTPUT_ROOT}/logs/cuda-common.log" 2>&1 & pids+=("$!") +common_profile 1 record_function "${OUTPUT_ROOT}/kernel-common" > "${OUTPUT_ROOT}/logs/kernel-common.log" 2>&1 & pids+=("$!") +moe_profile 2 cuda_event 4 1 "${OUTPUT_ROOT}/cuda-moe-tp4" > "${OUTPUT_ROOT}/logs/cuda-moe-tp4.log" 2>&1 & pids+=("$!") +moe_profile 3 cuda_event 1 8 "${OUTPUT_ROOT}/cuda-moe-ep8" > "${OUTPUT_ROOT}/logs/cuda-moe-ep8.log" 2>&1 & pids+=("$!") +moe_profile 4 record_function 4 1 "${OUTPUT_ROOT}/kernel-moe-tp4" > "${OUTPUT_ROOT}/logs/kernel-moe-tp4.log" 2>&1 & pids+=("$!") +moe_profile 5 record_function 1 8 "${OUTPUT_ROOT}/kernel-moe-ep8" > "${OUTPUT_ROOT}/logs/kernel-moe-ep8.log" 2>&1 & pids+=("$!") +failed=0 +for pid in "${pids[@]}"; do wait "${pid}" || failed=1; done +[[ "${failed}" -eq 0 ]] || { tail -n 80 "${OUTPUT_ROOT}"/logs/*.log; exit 1; } + +"${VENV_ROOT}/bin/python" "${RUNNER_DIR}/assemble_qwen235_v020_profiles.py" \ + --cuda-common "${OUTPUT_ROOT}/cuda-common" \ + --cuda-moe-tp4 "${OUTPUT_ROOT}/cuda-moe-tp4" --cuda-moe-ep8 "${OUTPUT_ROOT}/cuda-moe-ep8" \ + --kernel-common "${OUTPUT_ROOT}/kernel-common" \ + --kernel-moe-tp4 "${OUTPUT_ROOT}/kernel-moe-tp4" --kernel-moe-ep8 "${OUTPUT_ROOT}/kernel-moe-ep8" \ + --output-root "${OUTPUT_ROOT}/frozen" +date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ +echo Q235_V020_PROFILES_COMPLETE diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_real_surface.sh b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_real_surface.sh new file mode 100644 index 0000000..b7fd213 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_real_surface.sh @@ -0,0 +1,119 @@ +#!/usr/bin/env bash + +# Three fresh-server trials for the four-cell Qwen235 vLLM 0.20 surface. + +set -euo pipefail + +OUT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +TRACE_ROOT="${TRACE_ROOT:?TRACE_ROOT is required}" +CASE_NAME="${CASE_NAME:?CASE_NAME is required}" +PREFIX_CACHING="${PREFIX_CACHING:?PREFIX_CACHING is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +RUNNER="${RUNNER_DIR}/run_qwen30_exact_trace_real_anchor.sh" +CLIENT="${RUNNER_DIR}/qwen30_exact_trace_client.py" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8}" +SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-600}" +IDLE_GPU_MEMORY_TOLERANCE_MIB="${IDLE_GPU_MEMORY_TOLERANCE_MIB:-16}" +RESUME_VALID_CELLS="${RESUME_VALID_CELLS:-true}" +PORT="${BASE_PORT:-9300}" +REQUEST_COUNT="$(wc -l < "${TRACE_ROOT}/tp4/private/real_requests.jsonl")" + +[[ "$(wc -l < "${TRACE_ROOT}/tp8/private/real_requests.jsonl")" == "${REQUEST_COUNT}" ]] || { + echo 'ERROR: TP-specific request counts differ' >&2 + exit 1 +} +case "${CASE_NAME}" in fixed-pd|fixed-po|trace-pd|trace-po) ;; *) exit 2 ;; esac +case "${PREFIX_CACHING}" in true|false) ;; *) exit 2 ;; esac + +mkdir -p "${OUT}/provenance" +declare -a WAVE_PIDS=() + +has_valid_result() { + local result="$1" + [[ -s "${result}" ]] || return 1 + "${VENV_ROOT}/bin/python" - "${result}" "${REQUEST_COUNT}" <<'PY' +import json, sys +s = json.load(open(sys.argv[1])).get("summary", {}) +raise SystemExit(0 if s.get("completed") == int(sys.argv[2]) and s.get("failed") == 0 else 1) +PY +} + +preflight_gpus() { + nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader \ + | tee -a "${OUT}/controller.log" + nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits \ + | awk -v tolerance="${IDLE_GPU_MEMORY_TOLERANCE_MIB}" '$1 > tolerance {exit 1}' +} + +assert_no_server() { + ! pgrep -fa 'qwen3-235b-v020-eval' > "${OUT}/provenance/unexpected_server_processes.txt" +} + +wait_for_wave() { + local failed=0 pid + for pid in "${WAVE_PIDS[@]}"; do wait "${pid}" || failed=1; done + WAVE_PIDS=() + [[ "${failed}" -eq 0 ]] +} + +launch_config() { + local trial="$1" tp="$2" mns="$3" gpus="$4" ep=false ep_size=1 + if [[ "${tp}" == 8 ]]; then ep=true; ep_size=8; fi + local config="tp${tp}_ep${ep_size}_mns${mns}" run_out="${OUT}/real/${config}/trial${trial}" + local requests="${TRACE_ROOT}/tp${tp}/private/real_requests.jsonl" port="${PORT}" + PORT=$((PORT + 1)) + if [[ "${RESUME_VALID_CELLS}" == true ]] && has_valid_result "${run_out}/results/result.json"; then + printf 'CONFIG_REUSED_VALID case=%s trial=%s config=%s\n' "${CASE_NAME}" "${trial}" "${config}" + return 0 + fi + mkdir -p "${run_out}" + ( + cd "${RUNNER_DIR}" + set +e + env HOME=/tmp/wjh XDG_CACHE_HOME=/tmp/wjh/.cache \ + VLLM_CACHE_ROOT=/tmp/wjh/.cache/vllm CUDA_VISIBLE_DEVICES="${gpus}" \ + TP="${tp}" MNS="${mns}" TRACE_LABEL="${CASE_NAME}/tp${tp}-normalized" \ + PREFIX_CACHING="${PREFIX_CACHING}" ENABLE_EXPERT_PARALLEL="${ep}" \ + MODEL_QUANTIZATION=fp8 DISABLE_CUSTOM_ALL_REDUCE=true GPU_MEMORY_UTILIZATION=0.80 \ + SERVED_MODEL=qwen3-235b-v020-eval SERVER_PORT="${port}" OUTPUT_ROOT="${run_out}" \ + REQUESTS_FILE="${requests}" VENV_ROOT="${VENV_ROOT}" MODEL_ROOT="${MODEL_ROOT}" \ + SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS}" CLIENT_TIMEOUT_SECONDS=6000 \ + EXACT_TRACE_CLIENT="${CLIENT}" \ + timeout --signal=TERM --kill-after=60s 7800 bash "${RUNNER}" \ + > "${run_out}/launcher.stdout.log" 2> "${run_out}/launcher.stderr.log" + status="$?" + printf '%s\n' "${status}" > "${run_out}/launcher.exit_code" + exit "${status}" + ) & + WAVE_PIDS+=("$!") +} + +run_trial() { + local trial="$1" order="$2" + IFS=',' read -r -a mnss <<< "${order}" + preflight_gpus && assert_no_server + printf 'WAVE_START case=%s trial=%s tp=4 mns=%s,%s\n' \ + "${CASE_NAME}" "${trial}" "${mnss[0]}" "${mnss[1]}" | tee -a "${OUT}/controller.log" + launch_config "${trial}" 4 "${mnss[0]}" '0,1,2,3' + launch_config "${trial}" 4 "${mnss[1]}" '4,5,6,7' + wait_for_wave && preflight_gpus && assert_no_server + for mns in "${mnss[@]}"; do + printf 'WAVE_START case=%s trial=%s tp=8 mns=%s\n' \ + "${CASE_NAME}" "${trial}" "${mns}" | tee -a "${OUT}/controller.log" + launch_config "${trial}" 8 "${mns}" '0,1,2,3,4,5,6,7' + wait_for_wave && preflight_gpus && assert_no_server + done +} + +{ + printf '%s\n' "Q235_REAL_LAUNCH_ECHO host=dash0 model=Qwen3-235B-A22B-FP8 engine=vLLM-0.20.0+cu129 checkpoint=FP8 cases=${CASE_NAME} trace=${TRACE_ROOT}/tp{4,8} requests=${REQUEST_COUNT} transform=t_prime=t/TP prefix=${PREFIX_CACHING} surface={TP4/EP1,TP8/EP8}xMNS{64,128} MBT=8192 trials=3 fresh_server=true metrics=mean,p90(TTFT,TPOT-if-PD,E2E) SLO=not_scored output=${OUT}/real" + date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ + sha256sum "${BASH_SOURCE[0]}" "${RUNNER}" "${CLIENT}" "${MODEL_ROOT}/config.json" \ + > "${OUT}/provenance/real-input.sha256" + run_trial 1 '64,128' + run_trial 2 '128,64' + run_trial 3 '64,128' + date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ + printf '%s\n' 'Q235_REAL_SURFACE_COMPLETE' +} 2>&1 | tee -a "${OUT}/controller.log" diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_smoke.sh b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_smoke.sh new file mode 100644 index 0000000..f2affde --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen235_v020_smoke.sh @@ -0,0 +1,120 @@ +#!/usr/bin/env bash + +# Compatibility gate only: this is not a simulator-fidelity measurement. + +set -euo pipefail + +OUT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +TP="${TP:?TP is required}" +MNS="${MNS:?MNS is required}" +MBT="${MBT:?MBT is required}" +SERVER_PORT="${SERVER_PORT:?SERVER_PORT is required}" +ENABLE_EXPERT_PARALLEL="${ENABLE_EXPERT_PARALLEL:?ENABLE_EXPERT_PARALLEL is required}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8}" +CLIENT="${CLIENT:?CLIENT is required}" +READY_TIMEOUT_SECONDS="${READY_TIMEOUT_SECONDS:-1800}" +TRTLLM_DG_CACHE_DIR="${TRTLLM_DG_CACHE_DIR:-${OUT}/deep_gemm}" +TRTLLM_DG_NVCC_COMPILER="${TRTLLM_DG_NVCC_COMPILER:-/usr/local/cuda/bin/nvcc}" +DEEP_GEMM_JIT_DEBUG="${DEEP_GEMM_JIT_DEBUG:-false}" +SERVED_MODEL="qwen3-235b-v020-smoke" +SERVER_PID="" + +mkdir -p "${OUT}/logs" "${OUT}/provenance" "${OUT}/results" +exec > >(tee -a "${OUT}/logs/controller.log") 2>&1 + +cleanup() { + if [[ -n "${SERVER_PID}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then + kill -TERM -- "-${SERVER_PID}" 2>/dev/null || true + for _ in $(seq 1 30); do + kill -0 "${SERVER_PID}" 2>/dev/null || break + sleep 1 + done + kill -KILL -- "-${SERVER_PID}" 2>/dev/null || true + fi + SERVER_PID="" +} +trap cleanup EXIT INT TERM + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?GPU allocation is required}" +if [[ "${#GPU_IDS[@]}" -ne "${TP}" ]]; then + echo "ERROR: expected TP=${TP} GPUs, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi +case "${ENABLE_EXPERT_PARALLEL}" in + true) EP_FLAG=(--enable-expert-parallel) ;; + false) EP_FLAG=() ;; + *) echo "ERROR: ENABLE_EXPERT_PARALLEL must be true or false" >&2; exit 1 ;; +esac +if ! [[ "${READY_TIMEOUT_SECONDS}" =~ ^[1-9][0-9]*$ ]]; then + echo "ERROR: READY_TIMEOUT_SECONDS must be a positive integer" >&2 + exit 1 +fi +if [[ ! -x "${TRTLLM_DG_NVCC_COMPILER}" ]]; then + echo "ERROR: TRTLLM_DG_NVCC_COMPILER is not executable: ${TRTLLM_DG_NVCC_COMPILER}" >&2 + exit 1 +fi +case "${DEEP_GEMM_JIT_DEBUG}" in + true|false) ;; + *) echo "ERROR: DEEP_GEMM_JIT_DEBUG must be true or false" >&2; exit 1 ;; +esac + +echo "QWEN235_V020_SMOKE_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 checkpoint=FP8_compute_kv=auto config=TP${TP}_EP${ENABLE_EXPERT_PARALLEL}_MNS${MNS}_MBT${MBT} prefix=false requests=1 shape=2048x1 ready_timeout=${READY_TIMEOUT_SECONDS}s deep_gemm_cache=${TRTLLM_DG_CACHE_DIR} deep_gemm_nvcc=${TRTLLM_DG_NVCC_COMPILER} deep_gemm_jit_debug=${DEEP_GEMM_JIT_DEBUG} role=compatibility_gate_not_latency_measurement output=${OUT} expected_wall=20-35m expected_gpu_cap=$((${TP} * 35 / 60))_H20h" +date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ +sha256sum "${BASH_SOURCE[0]}" "${CLIENT}" "${MODEL_ROOT}/config.json" > "${OUT}/provenance/input.sha256" +"${VENV_ROOT}/bin/vllm" --version > "${OUT}/provenance/vllm.version" +"${VENV_ROOT}/bin/python" - <<'PY' > "${OUT}/provenance/runtime.json" +import importlib.metadata as metadata +import json +import vllm +print(json.dumps({"vllm": metadata.version("vllm"), "vllm_file": vllm.__file__}, sort_keys=True)) +PY + +export TOKENIZERS_PARALLELISM=false +export VLLM_USE_V1=1 +export TORCH_CUDA_ARCH_LIST=9.0 +export HF_HUB_OFFLINE=1 +export TRANSFORMERS_OFFLINE=1 +export HOME=/tmp/wjh +export XDG_CACHE_HOME=/tmp/wjh/.cache +export VLLM_CACHE_ROOT=/tmp/wjh/.cache/vllm +export TRTLLM_DG_CACHE_DIR +export TRTLLM_DG_NVCC_COMPILER +if [[ "${DEEP_GEMM_JIT_DEBUG}" == true ]]; then + export TRTLLM_DG_JIT_DEBUG=1 +fi +mkdir -p "${TRTLLM_DG_CACHE_DIR}" +setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \ + --host 127.0.0.1 --port "${SERVER_PORT}" --served-model-name "${SERVED_MODEL}" \ + --tensor-parallel-size "${TP}" --disable-custom-all-reduce --quantization fp8 \ + --gpu-memory-utilization 0.80 --kv-cache-dtype auto --max-model-len 40960 \ + --max-num-batched-tokens "${MBT}" --max-num-seqs "${MNS}" \ + --no-enable-prefix-caching --enable-chunked-prefill --no-enable-log-requests \ + "${EP_FLAG[@]}" > "${OUT}/logs/server.log" 2>&1 & +SERVER_PID=$! +READY=0 +READY_DEADLINE=$((SECONDS + READY_TIMEOUT_SECONDS)) +while (( SECONDS < READY_DEADLINE )); do + if curl -fsS --max-time 2 "http://127.0.0.1:${SERVER_PORT}/v1/models" > "${OUT}/results/models.json" 2>/dev/null; then + READY=1 + break + fi + if ! kill -0 "${SERVER_PID}" 2>/dev/null; then + tail -200 "${OUT}/logs/server.log" + exit 1 + fi + sleep 3 +done +if [[ "${READY}" -ne 1 ]]; then + tail -200 "${OUT}/logs/server.log" + exit 1 +fi + +"${VENV_ROOT}/bin/python" "${CLIENT}" --port "${SERVER_PORT}" --served-model "${SERVED_MODEL}" \ + --model-path "${MODEL_ROOT}" --rate 1 --requests 1 --input-tokens 2048 --output-tokens 1 \ + --output "${OUT}/results/one_request.json" +grep -E -i 'quant|fp8|gpu blocks|num_gpu_blocks|kv cache|expert.parallel|expert parallel' "${OUT}/logs/server.log" > "${OUT}/provenance/resolved-runtime.log" || true +cleanup +find "${OUT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 | sort -z | xargs -0 sha256sum > "${OUT}/provenance/artifacts.sha256" +date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ +echo 'QWEN235_V020_SMOKE_COMPLETE' diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen30_exact_trace_real_anchor.sh b/runs/frontier-fidelity-envelope-v1/run_qwen30_exact_trace_real_anchor.sh new file mode 100755 index 0000000..3ef6c21 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen30_exact_trace_real_anchor.sh @@ -0,0 +1,140 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +REQUESTS_FILE="${REQUESTS_FILE:?REQUESTS_FILE is required}" +TP="${TP:?TP is required}" +MNS="${MNS:?MNS is required}" +TRACE_LABEL="${TRACE_LABEL:?TRACE_LABEL is required}" +SERVER_PORT="${SERVER_PORT:?SERVER_PORT is required}" +PREFIX_CACHING="${PREFIX_CACHING:-true}" +ENABLE_EXPERT_PARALLEL="${ENABLE_EXPERT_PARALLEL:-false}" +MODEL_QUANTIZATION="${MODEL_QUANTIZATION:-}" +DISABLE_CUSTOM_ALL_REDUCE="${DISABLE_CUSTOM_ALL_REDUCE:-false}" +GPU_MEMORY_UTILIZATION="${GPU_MEMORY_UTILIZATION:-0.92}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:-/tmp/wjh/flashinfer-workspace-vllm020-profiler-v1}" +SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-120}" +CLIENT_TIMEOUT_SECONDS="${CLIENT_TIMEOUT_SECONDS:-1800}" +SERVED_MODEL="${SERVED_MODEL:-qwen3-30b-exact-trace}" +EXACT_TRACE_CLIENT="${EXACT_TRACE_CLIENT:-qwen30_exact_trace_client.py}" +SERVER_PID="" + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/results" \ + "${FLASHINFER_WORKSPACE_BASE}" +exec > >(tee -a "${OUTPUT_ROOT}/logs/controller.log") 2>&1 + +cleanup() { + if [[ -n "${SERVER_PID}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then + kill -TERM -- "-${SERVER_PID}" 2>/dev/null || true + for _ in $(seq 1 30); do + kill -0 "${SERVER_PID}" 2>/dev/null || break + sleep 1 + done + kill -KILL -- "-${SERVER_PID}" 2>/dev/null || true + fi + SERVER_PID="" +} +trap cleanup EXIT INT TERM + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU allocation is required}" +if [[ "${#GPU_IDS[@]}" -ne "${TP}" ]]; then + echo "ERROR: expected ${TP} GPUs, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +case "${PREFIX_CACHING}" in + true) + PREFIX_CACHING_FLAG="--enable-prefix-caching" + ;; + false) + PREFIX_CACHING_FLAG="--no-enable-prefix-caching" + ;; + *) + echo "ERROR: PREFIX_CACHING must be true or false, got ${PREFIX_CACHING}" >&2 + exit 1 + ;; +esac + +case "${ENABLE_EXPERT_PARALLEL}" in + true) EP_FLAG=(--enable-expert-parallel) ;; + false) EP_FLAG=() ;; + *) echo "ERROR: ENABLE_EXPERT_PARALLEL must be true or false" >&2; exit 1 ;; +esac +case "${DISABLE_CUSTOM_ALL_REDUCE}" in + true) CUSTOM_AR_FLAG=(--disable-custom-all-reduce) ;; + false) CUSTOM_AR_FLAG=() ;; + *) echo "ERROR: DISABLE_CUSTOM_ALL_REDUCE must be true or false" >&2; exit 1 ;; +esac +QUANT_FLAG=() +if [[ -n "${MODEL_QUANTIZATION}" ]]; then + QUANT_FLAG=(--quantization "${MODEL_QUANTIZATION}") +fi + +REQUEST_COUNT="$(wc -l < "${REQUESTS_FILE}")" +echo "EXACT_TRACE_REAL_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 quantization=${MODEL_QUANTIZATION:-none} config=TP${TP}_EP${ENABLE_EXPERT_PARALLEL}_MNS${MNS}_MBT8192 trace=${TRACE_LABEL} requests=${REQUEST_COUNT} source=${REQUESTS_FILE} arrivals=manifest prefix=${PREFIX_CACHING} block=16 metrics=TTFT,TPOT-if-OSL-gt-1,E2E flashinfer_workspace=${FLASHINFER_WORKSPACE_BASE} output=${OUTPUT_ROOT} ready_budget_s=$((SERVER_READY_ATTEMPTS * 3)) client_timeout_s=${CLIENT_TIMEOUT_SECONDS}" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +sha256sum qwen30_exact_trace_client.py run_qwen30_exact_trace_real_anchor.sh \ + ../frontier-phase-factorial-v0/qwen30_prefill_client.py \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +sha256sum "${REQUESTS_FILE}" > "${OUTPUT_ROOT}/provenance/requests.sha256" +sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model.sha256" +nvidia-smi --query-gpu=index,name,uuid,driver_version --format=csv,noheader \ + > "${OUTPUT_ROOT}/provenance/gpus.csv" + +export TOKENIZERS_PARALLELISM=false +export VLLM_USE_V1=1 +export TORCH_CUDA_ARCH_LIST=9.0 +export HF_HUB_OFFLINE=1 +export TRANSFORMERS_OFFLINE=1 +export FLASHINFER_WORKSPACE_BASE +ulimit -n 65536 + +setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \ + --host 127.0.0.1 --port "${SERVER_PORT}" --served-model-name "${SERVED_MODEL}" \ + --tensor-parallel-size "${TP}" --gpu-memory-utilization "${GPU_MEMORY_UTILIZATION}" \ + --max-model-len 40960 --max-num-batched-tokens 8192 --max-num-seqs "${MNS}" \ + "${PREFIX_CACHING_FLAG}" --enable-chunked-prefill --no-enable-log-requests \ + "${CUSTOM_AR_FLAG[@]}" "${QUANT_FLAG[@]}" "${EP_FLAG[@]}" \ + > "${OUTPUT_ROOT}/logs/server.log" 2>&1 & +SERVER_PID=$! +READY=0 +for _ in $(seq 1 "${SERVER_READY_ATTEMPTS}"); do + if curl -fsS --max-time 2 "http://127.0.0.1:${SERVER_PORT}/v1/models" \ + > "${OUTPUT_ROOT}/results/models.json" 2>/dev/null; then + READY=1 + break + fi + if ! kill -0 "${SERVER_PID}" 2>/dev/null; then + tail -200 "${OUTPUT_ROOT}/logs/server.log" + exit 1 + fi + sleep 3 +done +if [[ "${READY}" -ne 1 ]]; then + echo "ERROR server did not become ready within $((SERVER_READY_ATTEMPTS * 3)) seconds" >&2 + tail -200 "${OUTPUT_ROOT}/logs/server.log" + exit 1 +fi + +# Warm execution kernels with four prefix-disjoint requests, then start the +# measured replay from an empty scheduler queue. The unrelated 2K-token KV +# footprint remains explicit and negligible relative to the configured cache. +"${VENV_ROOT}/bin/python" ../frontier-phase-factorial-v0/qwen30_prefill_client.py \ + --port "${SERVER_PORT}" --served-model "${SERVED_MODEL}" \ + --model-path "${MODEL_ROOT}" --rate 1 --requests 4 --input-tokens 512 \ + --output-tokens 1 --output "${OUTPUT_ROOT}/results/warmup.json" + +"${VENV_ROOT}/bin/python" "${EXACT_TRACE_CLIENT}" \ + --port "${SERVER_PORT}" --requests-file "${REQUESTS_FILE}" \ + --served-model "${SERVED_MODEL}" \ + --output "${OUTPUT_ROOT}/results/result.json" --tpot-slo-ms 150 \ + --timeout-seconds "${CLIENT_TIMEOUT_SECONDS}" + +cleanup +find "${OUTPUT_ROOT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 \ + | sort -z | xargs -0 sha256sum > "${OUTPUT_ROOT}/provenance/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "QWEN30_EXACT_TRACE_REAL_ANCHOR_COMPLETE" diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pd_pressure_probe.sh b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pd_pressure_probe.sh new file mode 100644 index 0000000..b8ada40 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pd_pressure_probe.sh @@ -0,0 +1,255 @@ +#!/usr/bin/env bash + +# Real-only, pressure-matching probe for the next Fixed-PD workload. This +# intentionally profiles one anchor, then freezes the workload before any +# Frontier-vs-real selection comparison. + +set -euo pipefail + +OUT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +CLIENT="${CLIENT:-${RUNNER_DIR}/../frontier-phase-factorial-v0/qwen30_prefill_client.py}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +# Reuse the validated vLLM 0.20 H20 kernel cache. A per-output workspace +# starts costly FlashInfer MoE JIT compilation and changes startup behavior +# without changing the serving configuration being profiled. +FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:?FLASHINFER_WORKSPACE_BASE is required}" +GPU_IDS="${GPU_IDS:-0,1,2,3}" +TP="${TP:-4}" +MNS="${MNS:-64}" +REQUESTS="${REQUESTS:-257}" +GLOBAL_RATES="${GLOBAL_RATES:-4 4.5 5 6}" +SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-180}" +PORT="${PORT:-8930}" +SERVED_MODEL="qwen3-30b-fixed-pd-pressure" +SERVER_PID="" + +[[ "${TP}" == "4" ]] || { echo 'ERROR: this calibrated probe is TP4-only' >&2; exit 1; } +[[ "${MNS}" == "64" ]] || { echo 'ERROR: this calibrated probe is MNS64-only' >&2; exit 1; } +[[ "${REQUESTS}" =~ ^[1-9][0-9]*$ ]] || { echo 'ERROR: REQUESTS must be positive' >&2; exit 1; } +[[ -f "${CLIENT}" ]] || { echo "ERROR: client missing: ${CLIENT}" >&2; exit 1; } +[[ -f "${MODEL_ROOT}/config.json" ]] || { echo "ERROR: model missing: ${MODEL_ROOT}" >&2; exit 1; } +[[ -d "${FLASHINFER_WORKSPACE_BASE}" ]] || { echo "ERROR: FlashInfer workspace missing: ${FLASHINFER_WORKSPACE_BASE}" >&2; exit 1; } +read -r -a RATE_VALUES <<< "${GLOBAL_RATES}" +[[ "${#RATE_VALUES[@]}" -eq 4 ]] || { echo 'ERROR: GLOBAL_RATES must contain exactly four rates' >&2; exit 1; } +for rate in "${RATE_VALUES[@]}"; do + awk -v value="${rate}" 'BEGIN {exit !(value > 0)}' || { echo "ERROR: invalid rate: ${rate}" >&2; exit 1; } +done + +mkdir -p "${OUT}/provenance" "${OUT}/trials" "${FLASHINFER_WORKSPACE_BASE}" +exec > >(tee -a "${OUT}/controller.log") 2>&1 + +# Match the validated Qwen30 real-surface runner's file-descriptor budget for +# vLLM's multiprocessing shared-memory transport. The failed r1 probe had the +# default remote limit (1024) and stalled before KV-cache creation; r2 tests +# whether this runner-contract difference is causal. +ulimit -n 65536 + +cleanup_server() { + if [[ -n "${SERVER_PID}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then + kill -TERM -- "-${SERVER_PID}" 2>/dev/null || true + for _ in $(seq 1 30); do + kill -0 "${SERVER_PID}" 2>/dev/null || break + sleep 1 + done + kill -KILL -- "-${SERVER_PID}" 2>/dev/null || true + fi + SERVER_PID="" +} +trap cleanup_server EXIT +trap 'cleanup_server; exit 130' INT +trap 'cleanup_server; exit 143' TERM + +assert_idle() { + nvidia-smi --query-gpu=index,memory.used,utilization.gpu --format=csv,noheader + nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits \ + | awk '$1 > 16 {exit 1}' +} + +wait_ready() { + local target="$1" + for _ in $(seq 1 "${SERVER_READY_ATTEMPTS}"); do + if curl -fsS --max-time 2 "http://127.0.0.1:${PORT}/v1/models" > "${target}/models.json" 2>/dev/null; then + return 0 + fi + if ! kill -0 "${SERVER_PID}" 2>/dev/null; then + tail -200 "${target}/server.log" >&2 || true + return 1 + fi + sleep 3 + done + echo "ERROR: vLLM did not become ready in $((SERVER_READY_ATTEMPTS * 3)) seconds" >&2 + return 1 +} + +start_server() { + local target="$1" + export TOKENIZERS_PARALLELISM=false + export VLLM_USE_V1=1 + export TORCH_CUDA_ARCH_LIST=9.0 + export HF_HUB_OFFLINE=1 + export TRANSFORMERS_OFFLINE=1 + export FLASHINFER_WORKSPACE_BASE + export HOME=/tmp/wjh + export XDG_CACHE_HOME=/tmp/wjh/.cache + export VLLM_CACHE_ROOT=/tmp/wjh/.cache/vllm + export CUDA_VISIBLE_DEVICES="${GPU_IDS}" + setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \ + --host 127.0.0.1 --port "${PORT}" --served-model-name "${SERVED_MODEL}" \ + --tensor-parallel-size "${TP}" --gpu-memory-utilization 0.92 \ + --max-model-len 40960 --max-num-batched-tokens 8192 --max-num-seqs "${MNS}" \ + --no-enable-prefix-caching --enable-chunked-prefill --no-enable-log-requests \ + > "${target}/server.log" 2>&1 & + SERVER_PID=$! + wait_ready "${target}" +} + +run_client() { + local target="$1" rate="$2" + timeout --signal=TERM --kill-after=60s 1800 \ + "${VENV_ROOT}/bin/python" "${CLIENT}" \ + --port "${PORT}" --served-model "${SERVED_MODEL}" --model-path "${MODEL_ROOT}" \ + --rate "${rate}" --requests "${REQUESTS}" --input-tokens 4096 --output-tokens 256 \ + --timeout-seconds 1200 --output "${target}/result.json" +} + +warmup_server() { + local target="$1" + timeout --signal=TERM --kill-after=60s 600 \ + "${VENV_ROOT}/bin/python" "${CLIENT}" \ + --port "${PORT}" --served-model "${SERVED_MODEL}" --model-path "${MODEL_ROOT}" \ + --rate 1 --requests 4 --input-tokens 512 --output-tokens 1 \ + --timeout-seconds 300 --output "${target}/result.json" +} + +analyze() { + "${VENV_ROOT}/bin/python" - "${OUT}" "${GLOBAL_RATES}" <<'PY' +import json +import math +import statistics +import sys +from pathlib import Path + +root = Path(sys.argv[1]) +rates = [float(value) for value in sys.argv[2].split()] +target = {"ttft_ms": 245.9527667526406, "tpot_ms": 13.178025610291787} + +def p90(values): + return sorted(values)[math.ceil(0.9 * len(values)) - 1] + +rows = [] +for rate in rates: + label = f"r{rate:g}" + trial_means = {"ttft_ms": [], "tpot_ms": [], "e2e_ms": []} + pooled = {key: [] for key in trial_means} + for trial in range(1, 4): + path = root / "trials" / f"trial{trial}" / label / "result.json" + payload = json.loads(path.read_text()) + workload = payload["workload"] + if (float(workload["offered_request_rate"]) != rate or workload["request_count"] != 257 + or workload["input_tokens"] != 4096 or workload["output_tokens"] != 256 + or workload["prefix_caching"] is not False): + raise ValueError(f"workload drift: {path}") + requests = payload["requests"] + if len(requests) != 257 or any(not request["success"] for request in requests): + raise ValueError(f"incomplete client result: {path}") + for key in pooled: + values = [float(request[key]) for request in requests] + pooled[key].extend(values) + trial_means[key].append(statistics.mean(values)) + row = { + "global_rate": rate, + "per_gpu_rate": rate / 4.0, + "requests_per_trial": 257, + "trials": 3, + "metrics": { + key: { + "pooled_mean_ms": statistics.mean(values), + "pooled_p90_ms": p90(values), + "trial_mean_stdev_ms": statistics.stdev(trial_means[key]), + } + for key, values in pooled.items() + }, + } + row["inflight_proxy"] = rate * row["metrics"]["e2e_ms"]["pooled_mean_ms"] / 1000.0 + row["relative_distance"] = math.sqrt(sum( + ((row["metrics"][key]["pooled_mean_ms"] - target[key]) / target[key]) ** 2 + for key in target + )) + rows.append(row) + +winner = min(rows, key=lambda row: (row["relative_distance"], row["global_rate"])) +payload = { + "schema": "qwen30-fixed-pd-pressure-probe-v1", + "target_trace_pd_tp4_mns64": target, + "decision_rule": "minimum Euclidean distance of relative mean TTFT and TPOT errors", + "rates": rows, + "recommended_global_rate": winner["global_rate"], + "recommended_per_gpu_rate": winner["per_gpu_rate"], +} +(root / "pressure-analysis.json").write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") +lines = [ + "# Fixed-PD pressure probe", + "", + "| Global / per-GPU rps | TTFT mean / p90 (ms) | TPOT mean / p90 (ms) | E2E mean / p90 (ms) | In-flight proxy | Relative distance |", + "|---|---:|---:|---:|---:|---:|", +] +for row in rows: + metric = row["metrics"] + lines.append( + f"| {row['global_rate']:g} / {row['per_gpu_rate']:g} | " + f"{metric['ttft_ms']['pooled_mean_ms']:.2f} / {metric['ttft_ms']['pooled_p90_ms']:.2f} | " + f"{metric['tpot_ms']['pooled_mean_ms']:.2f} / {metric['tpot_ms']['pooled_p90_ms']:.2f} | " + f"{metric['e2e_ms']['pooled_mean_ms']:.2f} / {metric['e2e_ms']['pooled_p90_ms']:.2f} | " + f"{row['inflight_proxy']:.2f} | {row['relative_distance']:.3f} |" + ) +lines.extend([ + "", + f"Recommended frozen rate: **{winner['global_rate']:g} global rps / {winner['per_gpu_rate']:g} rps per GPU**.", + "Selection uses only pooled mean TTFT and TPOT; p90 and in-flight proxy are audit outputs.", +]) +(root / "pressure-analysis.md").write_text("\n".join(lines) + "\n") +print(json.dumps(payload, sort_keys=True)) +PY +} + +{ + echo "FIXED_PD_PRESSURE_PROBE_LAUNCH_ECHO host=$(hostname) model=${MODEL_ROOT} engine=vLLM-0.20.0+cu129 dtype=BF16 config=TP${TP}_MNS${MNS}_MBT8192 gpus=${GPU_IDS} prefix=false shape=4096_to_256 requests_per_rate=${REQUESTS} global_rates={${GLOBAL_RATES}} rate_contract=global_rate_divided_by_TP flashinfer_workspace=${FLASHINFER_WORKSPACE_BASE} trials=3 fresh_server=true metric_target=TracePD_TP4_MNS64_meanTTFT245.95ms_meanTPOT13.18ms expected_wall=12-20m expected_cost=0.8-1.4_H20-GPUh output=${OUT}" + date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ + assert_idle + sha256sum "${BASH_SOURCE[0]}" "${CLIENT}" "${MODEL_ROOT}/config.json" > "${OUT}/provenance/input.sha256" + "${VENV_ROOT}/bin/vllm" --version > "${OUT}/provenance/vllm.version" + "${VENV_ROOT}/bin/python" -c 'import torch, transformers, vllm; print(f"torch={torch.__version__}"); print(f"transformers={transformers.__version__}"); print(f"vllm={vllm.__version__}")' > "${OUT}/provenance/runtime.versions" + ulimit -n > "${OUT}/provenance/open-file-limit" + readlink -f "${FLASHINFER_WORKSPACE_BASE}" > "${OUT}/provenance/flashinfer-workspace" + nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader > "${OUT}/provenance/gpus.before.csv" + + declare -a ORDERS=( + "${RATE_VALUES[*]}" + "${RATE_VALUES[3]} ${RATE_VALUES[2]} ${RATE_VALUES[1]} ${RATE_VALUES[0]}" + "${RATE_VALUES[1]} ${RATE_VALUES[3]} ${RATE_VALUES[0]} ${RATE_VALUES[2]}" + ) + for trial in 1 2 3; do + trial_root="${OUT}/trials/trial${trial}" + mkdir -p "${trial_root}" + echo "TRIAL_START trial=${trial} order=${ORDERS[$((trial - 1))]}" + start_server "${trial_root}" + warmup_server "${trial_root}/warmup" + for rate in ${ORDERS[$((trial - 1))]}; do + rate_root="${trial_root}/r${rate}" + mkdir -p "${rate_root}" + echo "RATE_START trial=${trial} global_rate=${rate} per_gpu_rate=$(awk -v value="${rate}" 'BEGIN {printf "%.3f", value / 4}')" + run_client "${rate_root}" "${rate}" + echo "RATE_COMPLETE trial=${trial} global_rate=${rate}" + done + cleanup_server + assert_idle + echo "TRIAL_COMPLETE trial=${trial}" + done + analyze + find "${OUT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 | sort -z | xargs -0 sha256sum > "${OUT}/provenance/artifacts.sha256" + nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader > "${OUT}/provenance/gpus.after.csv" + date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ + echo 'FIXED_PD_PRESSURE_PROBE_COMPLETE' +} >> "${OUT}/controller.log" 2>&1 diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pressure_campaign.sh b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pressure_campaign.sh new file mode 100644 index 0000000..2f0ade6 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_pressure_campaign.sh @@ -0,0 +1,138 @@ +#!/usr/bin/env bash + +# Frozen Qwen30 Fixed-PD/Fixed-PO Frontier-versus-real campaign. The pressure +# calibration artifacts are deliberately not reused as evaluation results. + +set -euo pipefail + +CAMPAIGN_ROOT="${CAMPAIGN_ROOT:?CAMPAIGN_ROOT is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-t1-dash0-deadc4a}" +REPLAYSERVE_ROOT="${REPLAYSERVE_ROOT:-/home/admin/cpfs/wjh/replayserve}" +PYTHON_DEPS="${PYTHON_DEPS:-${VENV_ROOT}/lib/python3.12/site-packages}" +PROFILE_ROOT="${PROFILE_ROOT:-/home/admin/cpfs/wjh/aituner/aituner-graph-piecewise-bdc357d/runs/frontier-fidelity-envelope-v1/profiles/profile-v4-trace-final}" +KERNEL_PROFILE_ROOT="${KERNEL_PROFILE_ROOT:-/home/admin/cpfs/wjh/aituner/graph-piecewise-qwen30-20260717/full/frozen-kernel-only}" +ALLREDUCE_CSV="${ALLREDUCE_CSV:-/home/admin/cpfs/wjh/aituner/aituner-graph-piecewise-bdc357d/runs/frontier-fidelity-envelope-v1/profiles/measured-allreduce.csv}" +FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:?FLASHINFER_WORKSPACE_BASE is required}" +REQUESTS="${REQUESTS:-257}" +PER_GPU_RATE="${PER_GPU_RATE:-1.125}" + +[[ "${REQUESTS}" == "257" ]] || { echo 'ERROR: frozen campaign requires REQUESTS=257' >&2; exit 1; } +[[ "${PER_GPU_RATE}" == "1.125" ]] || { echo 'ERROR: frozen campaign requires PER_GPU_RATE=1.125' >&2; exit 1; } + +MATERIALIZER="${RUNNER_DIR}/prepare_qwen30_latency_case.py" +REAL_RUNNER="${RUNNER_DIR}/run_qwen30_latency_case_real_surface.sh" +SIM_RUNNER="${RUNNER_DIR}/run_frontier_qwen30_exact_trace_surface.py" +REAL_AUDITOR="${RUNNER_DIR}/audit_qwen30_latency_case.py" +COMPARATOR="${RUNNER_DIR}/analyze_qwen30_latency_case.py" + +mkdir -p "${CAMPAIGN_ROOT}/provenance" "${CAMPAIGN_ROOT}/traces" \ + "${CAMPAIGN_ROOT}/real" "${CAMPAIGN_ROOT}/sim" "${CAMPAIGN_ROOT}/analysis" +exec > >(tee -a "${CAMPAIGN_ROOT}/controller.log") 2>&1 + +printf '%s\n' "Q30_FIXED_PRESSURE_CAMPAIGN_LAUNCH_ECHO host=dash0 model=Qwen3-30B-A3B engine=vLLM-0.20.0+cu129 dtype=BF16 cases={fixed-pd:4096_to_256,fixed-po:4096_to_1} prefix=false requests=${REQUESTS} per_gpu_rate=${PER_GPU_RATE} global_rates={TP1:1.125,TP2:2.25,TP4:4.5} surface_per_case=TP{1,2,4}xMNS{8,16,32,64} real_trials=3 fresh_server=true simulator=Frontier-deadc4a_piecewise_graph-kernel-only metrics=mean,p90(TTFT,TPOT-if-PD,E2E) SLO=not_scored flashinfer_workspace=${FLASHINFER_WORKSPACE_BASE} expected_wall=2-5h expected_cost=12-24_H20-GPUh output=${CAMPAIGN_ROOT}" +date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ + +sha256sum "${BASH_SOURCE[0]}" "${MATERIALIZER}" "${REAL_RUNNER}" \ + "${SIM_RUNNER}" "${REAL_AUDITOR}" "${COMPARATOR}" \ + "${MODEL_ROOT}/config.json" "${PROFILE_ROOT}/manifest.json" \ + "${KERNEL_PROFILE_ROOT}/manifest.json" "${ALLREDUCE_CSV}" \ + > "${CAMPAIGN_ROOT}/provenance/input.sha256" +git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${CAMPAIGN_ROOT}/provenance/frontier.commit" +git -C "${RUNNER_DIR}" rev-parse HEAD > "${CAMPAIGN_ROOT}/provenance/aituner.commit" +"${VENV_ROOT}/bin/vllm" --version > "${CAMPAIGN_ROOT}/provenance/vllm.version" +nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader \ + > "${CAMPAIGN_ROOT}/provenance/gpus.before.csv" + +prepare_case() { + local case_name="$1" output_tokens="$2" + local tp + for tp in 1 2 4; do + "${VENV_ROOT}/bin/python" "${MATERIALIZER}" fixed \ + --model "${MODEL_ROOT}" --input-tokens 4096 --output-tokens "${output_tokens}" \ + --requests "${REQUESTS}" --per-gpu-rate "${PER_GPU_RATE}" --tp "${tp}" \ + --output-root "${CAMPAIGN_ROOT}/traces/${case_name}/tp${tp}" + done +} + +run_real_case() { + local case_name="$1" base_port="$2" + CASE_NAME="${case_name}" PREFIX_CACHING=false \ + TRACE_ROOT="${CAMPAIGN_ROOT}/traces/${case_name}" \ + OUTPUT_ROOT="${CAMPAIGN_ROOT}/real/${case_name}" \ + RUNNER_DIR="${RUNNER_DIR}" VENV_ROOT="${VENV_ROOT}" MODEL_ROOT="${MODEL_ROOT}" \ + FLASHINFER_SHARED_WORKSPACE="${FLASHINFER_WORKSPACE_BASE}" \ + BASE_PORT="${base_port}" RESUME_VALID_CELLS=true \ + bash "${REAL_RUNNER}" +} + +run_sim_tp() { + local case_name="$1" tp="$2" + local sim_root="${CAMPAIGN_ROOT}/sim/${case_name}" + local -a configs=() + local mns + for mns in 8 16 32 64; do + configs+=(--config "tp${tp}_mns${mns}") + done + /usr/bin/python3 "${SIM_RUNNER}" \ + --frontier-source "${FRONTIER_SOURCE}" --replayserve-root "${REPLAYSERVE_ROOT}" \ + --profile-root "${PROFILE_ROOT}" --kernel-profile-root "${KERNEL_PROFILE_ROOT}" \ + --python-deps "${PYTHON_DEPS}" --output-root "${sim_root}" \ + --trace "tp${tp}=${CAMPAIGN_ROOT}/traces/${case_name}/tp${tp}/public/frontier.csv" \ + "${configs[@]}" --rate-contract uniform-spacing --no-prefix-caching \ + --cc-backend vidur --allreduce-csv "${ALLREDUCE_CSV}" \ + --timeout-seconds 3600 --predictor-training-job-threads 4 \ + --decode-cuda-graph-mode piecewise --align-real-graph-runtime \ + --fresh-predictor-cache --resume --continue-on-failure +} + +run_sim_case() { + local case_name="$1" failed=0 pid + local -a pids=() + mkdir -p "${CAMPAIGN_ROOT}/sim/${case_name}" + for tp in 1 2 4; do + run_sim_tp "${case_name}" "${tp}" \ + > "${CAMPAIGN_ROOT}/sim/${case_name}/launcher-tp${tp}.log" 2>&1 & + pids+=("$!") + done + for pid in "${pids[@]}"; do + wait "${pid}" || failed=1 + done + [[ "${failed}" -eq 0 ]] || return 1 +} + +analyze_case() { + local case_name="$1" + local analysis_root="${CAMPAIGN_ROOT}/analysis/${case_name}" + mkdir -p "${analysis_root}" + "${VENV_ROOT}/bin/python" "${REAL_AUDITOR}" \ + --case-root "${CAMPAIGN_ROOT}/real/${case_name}" \ + --traces-root "${CAMPAIGN_ROOT}/traces/${case_name}" \ + --json-output "${analysis_root}/real-audit.json" \ + --markdown-output "${analysis_root}/real-audit.md" + "${VENV_ROOT}/bin/python" "${COMPARATOR}" \ + --sim-root "${CAMPAIGN_ROOT}/sim/${case_name}" \ + --real-audit "${analysis_root}/real-audit.json" \ + --json-output "${analysis_root}/comparison.json" \ + --markdown-output "${analysis_root}/comparison.md" +} + +prepare_case fixed-pd 256 +prepare_case fixed-po 1 + +run_real_case fixed-pd 9000 +run_real_case fixed-po 9100 + +run_sim_case fixed-pd +analyze_case fixed-pd +run_sim_case fixed-po +analyze_case fixed-po + +nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader \ + > "${CAMPAIGN_ROOT}/provenance/gpus.after.csv" +find "${CAMPAIGN_ROOT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 \ + | sort -z | xargs -0 sha256sum > "${CAMPAIGN_ROOT}/provenance/artifacts.sha256" +date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ +printf '%s\n' 'Q30_FIXED_PRESSURE_CAMPAIGN_COMPLETE' diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_runtime_preflight.sh b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_runtime_preflight.sh new file mode 100644 index 0000000..1844faa --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen30_fixed_runtime_preflight.sh @@ -0,0 +1,150 @@ +#!/usr/bin/env bash + +# Capture vLLM's resolved runtime state for Fixed-P/PD before their Frontier +# surfaces are frozen. No benchmark request is issued in this preflight. + +set -euo pipefail + +OUT="${1:?output root is required}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:-${OUT}/flashinfer-shared-workspace}" +TPS="${TPS:-4 2 1}" +PORT=8600 + +mkdir -p "${OUT}/runs" "${OUT}/provenance" "${FLASHINFER_WORKSPACE_BASE}" +exec > >(tee -a "${OUT}/controller.log") 2>&1 + +echo "FIXED_RUNTIME_PREFLIGHT_LAUNCH_ECHO host=$(hostname) model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 tps=${TPS}xMNS{8,16,32,64} MBT=8192 prefix=false chunked_prefill=true requests=0 resolved_state=graph_capture_and_kv_blocks output=${OUT} expected_wall=8-20m expected_gpu_cap=2_H20h" +date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ +sha256sum "${BASH_SOURCE[0]}" "${MODEL_ROOT}/config.json" > "${OUT}/provenance/input.sha256" +"${VENV_ROOT}/bin/vllm" --version > "${OUT}/provenance/vllm.version" + +wait_for_idle() { + for _ in $(seq 1 30); do + if nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | awk '$1 != 0 {exit 1}'; then + return 0 + fi + sleep 2 + done + return 1 +} + +if ! wait_for_idle; then + echo 'ERROR: a GPU is not idle before runtime preflight' >&2 + exit 1 +fi + +declare -a PIDS=() + +cleanup_pid() { + local pid="$1" + if kill -0 "${pid}" 2>/dev/null; then + kill -TERM -- "-${pid}" 2>/dev/null || true + for _ in $(seq 1 30); do + kill -0 "${pid}" 2>/dev/null || break + sleep 1 + done + kill -KILL -- "-${pid}" 2>/dev/null || true + fi +} + +cleanup_all() { + local pid + for pid in "${PIDS[@]:-}"; do + cleanup_pid "${pid}" + done +} +trap cleanup_all EXIT INT TERM + +launch_one() { + local tp="$1" mns="$2" gpus="$3" port="$4" + local name="tp${tp}_mns${mns}" + local root="${OUT}/runs/${name}" + mkdir -p "${root}" + ( + export HOME=/tmp/wjh + export XDG_CACHE_HOME=/tmp/wjh/.cache + export VLLM_CACHE_ROOT=/tmp/wjh/.cache/vllm + export TOKENIZERS_PARALLELISM=false + export VLLM_USE_V1=1 + export TORCH_CUDA_ARCH_LIST=9.0 + export HF_HUB_OFFLINE=1 + export TRANSFORMERS_OFFLINE=1 + export CUDA_VISIBLE_DEVICES="${gpus}" + setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \ + --host 127.0.0.1 --port "${port}" --served-model-name qwen3-30b-fixed-preflight \ + --tensor-parallel-size "${tp}" --gpu-memory-utilization 0.92 \ + --max-model-len 40960 --max-num-batched-tokens 8192 --max-num-seqs "${mns}" \ + --no-enable-prefix-caching --enable-chunked-prefill --no-enable-log-requests \ + > "${root}/server.log" 2>&1 & + local server_pid=$! + local ready=0 + for _ in $(seq 1 150); do + if curl -fsS --max-time 2 "http://127.0.0.1:${port}/v1/models" > "${root}/models.json" 2>/dev/null; then + ready=1 + break + fi + if ! kill -0 "${server_pid}" 2>/dev/null; then + break + fi + sleep 2 + done + grep -E -i 'cuda graph|cudagraph|capture|gpu blocks|num_gpu_blocks|kv cache' "${root}/server.log" > "${root}/resolved-state.log" || true + if [[ "${ready}" -ne 1 ]]; then + echo "status=failed" > "${root}/status" + cleanup_pid "${server_pid}" + exit 1 + fi + echo "status=ready" > "${root}/status" + cleanup_pid "${server_pid}" + ) & + PIDS+=("$!") +} + +wait_wave() { + local failed=0 pid + for pid in "${PIDS[@]}"; do + wait "${pid}" || failed=1 + done + PIDS=() + if [[ "${failed}" -ne 0 ]]; then + echo 'ERROR: one or more runtime preflight servers failed' >&2 + exit 1 + fi + if ! wait_for_idle; then + echo 'ERROR: a GPU remains allocated after runtime preflight wave' >&2 + exit 1 + fi +} + +for tp in ${TPS}; do + echo "PREFLIGHT_WAVE_START tp=${tp}" + case "${tp}" in + 4) + launch_one 4 8 '0,1,2,3' "${PORT}"; PORT=$((PORT + 1)) + launch_one 4 16 '4,5,6,7' "${PORT}"; PORT=$((PORT + 1)) + wait_wave + launch_one 4 32 '0,1,2,3' "${PORT}"; PORT=$((PORT + 1)) + launch_one 4 64 '4,5,6,7' "${PORT}"; PORT=$((PORT + 1)) + ;; + 2) + launch_one 2 8 '0,1' "${PORT}"; PORT=$((PORT + 1)) + launch_one 2 16 '2,3' "${PORT}"; PORT=$((PORT + 1)) + launch_one 2 32 '4,5' "${PORT}"; PORT=$((PORT + 1)) + launch_one 2 64 '6,7' "${PORT}"; PORT=$((PORT + 1)) + ;; + 1) + launch_one 1 8 '0' "${PORT}"; PORT=$((PORT + 1)) + launch_one 1 16 '1' "${PORT}"; PORT=$((PORT + 1)) + launch_one 1 32 '2' "${PORT}"; PORT=$((PORT + 1)) + launch_one 1 64 '3' "${PORT}"; PORT=$((PORT + 1)) + ;; + esac + wait_wave + echo "PREFLIGHT_WAVE_COMPLETE tp=${tp}" +done + +find "${OUT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 | sort -z | xargs -0 sha256sum > "${OUT}/provenance/artifacts.sha256" +date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ +echo 'FIXED_RUNTIME_PREFLIGHT_COMPLETE' diff --git a/runs/frontier-fidelity-envelope-v1/run_qwen30_latency_case_real_surface.sh b/runs/frontier-fidelity-envelope-v1/run_qwen30_latency_case_real_surface.sh new file mode 100644 index 0000000..e6a68ff --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/run_qwen30_latency_case_real_surface.sh @@ -0,0 +1,229 @@ +#!/usr/bin/env bash + +# One three-trial Qwen30 latency-selection surface. The caller supplies a +# frozen trace root and runs different cases sequentially to keep GPU use +# exclusive. + +set -euo pipefail + +OUT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +TRACE_ROOT="${TRACE_ROOT:?TRACE_ROOT is required}" +CASE_NAME="${CASE_NAME:?CASE_NAME is required}" +PREFIX_CACHING="${PREFIX_CACHING:?PREFIX_CACHING is required}" +RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}" +RUNNER="${RUNNER_DIR}/run_qwen30_exact_trace_real_anchor.sh" +CLIENT="${RUNNER_DIR}/qwen30_exact_trace_client.py" +PREFILL_CLIENT="${RUNNER_DIR}/../frontier-phase-factorial-v0/qwen30_prefill_client.py" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +FLASHINFER_SHARED_WORKSPACE="${FLASHINFER_SHARED_WORKSPACE:-${OUT}/flashinfer-shared-workspace}" +SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-300}" +IDLE_GPU_MEMORY_TOLERANCE_MIB="${IDLE_GPU_MEMORY_TOLERANCE_MIB:-16}" +RESUME_VALID_CELLS="${RESUME_VALID_CELLS:-false}" +PORT="${BASE_PORT:-8300}" +REQUEST_COUNT="$(wc -l < "${TRACE_ROOT}/tp1/private/real_requests.jsonl")" + +for tp in 2 4; do + [[ "$(wc -l < "${TRACE_ROOT}/tp${tp}/private/real_requests.jsonl")" == "${REQUEST_COUNT}" ]] || { + echo "ERROR: TP-specific request counts differ" >&2 + exit 1 + } +done + +case "${PREFIX_CACHING}" in + true|false) ;; + *) echo "ERROR: PREFIX_CACHING must be true or false" >&2; exit 1 ;; +esac + +case "${CASE_NAME}" in + trace-po|fixed-pd|fixed-po) ;; + *) echo "ERROR: unsupported CASE_NAME=${CASE_NAME}" >&2; exit 1 ;; +esac + +case "${RESUME_VALID_CELLS}" in + true|false) ;; + *) echo "ERROR: RESUME_VALID_CELLS must be true or false" >&2; exit 1 ;; +esac + +[[ "${IDLE_GPU_MEMORY_TOLERANCE_MIB}" =~ ^[0-9]+$ ]] || { + echo "ERROR: IDLE_GPU_MEMORY_TOLERANCE_MIB must be a non-negative integer" >&2 + exit 1 +} + +for path in "${RUNNER}" "${CLIENT}" "${PREFILL_CLIENT}" "${MODEL_ROOT}/config.json"; do + [[ -f "${path}" ]] || { echo "ERROR: required path is missing: ${path}" >&2; exit 1; } +done +mkdir -p "${OUT}" + +declare -a WAVE_PIDS=() + +has_valid_result() { + local result="$1" expected_requests="$2" + [[ -s "${result}" ]] || return 1 + "${VENV_ROOT}/bin/python" - "${result}" "${expected_requests}" <<'PY' +import json +import sys + +with open(sys.argv[1]) as source: + summary = json.load(source).get("summary", {}) +raise SystemExit( + 0 + if summary.get("completed") == int(sys.argv[2]) and summary.get("failed") == 0 + else 1 +) +PY +} + +wait_for_wave() { + local failed=0 pid + for pid in "${WAVE_PIDS[@]}"; do + if ! wait "${pid}"; then + failed=1 + fi + done + WAVE_PIDS=() + if [[ "${failed}" -ne 0 ]]; then + printf '%s\n' 'ERROR one or more real-serving runs failed; aborting surface.' \ + | tee -a "${OUT}/controller.log" >&2 + return 1 + fi +} + +preflight_gpus() { + nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader \ + | tee -a "${OUT}/controller.log" + if nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits \ + | awk -v tolerance="${IDLE_GPU_MEMORY_TOLERANCE_MIB}" '$1 > tolerance {exit 1}'; then + return 0 + fi + printf 'ERROR GPU memory exceeds idle tolerance (%s MiB) before a real-serving wave; refusing to start.\n' \ + "${IDLE_GPU_MEMORY_TOLERANCE_MIB}" \ + | tee -a "${OUT}/controller.log" >&2 + return 1 +} + +assert_no_our_server() { + if pgrep -fa 'qwen3-30b-exact-trace' > "${OUT}/provenance/unexpected_server_processes.txt"; then + printf '%s\n' 'ERROR a prior Qwen30 server remains; refusing the next wave.' \ + | tee -a "${OUT}/controller.log" >&2 + return 1 + fi +} + +launch_config() { + local trial="$1" tp="$2" mns="$3" gpus="$4" + local config="tp${tp}_mns${mns}" + local run_out="${OUT}/real/${config}/trial${trial}" + local requests="${TRACE_ROOT}/tp${tp}/private/real_requests.jsonl" + local request_count + local port="${PORT}" + PORT=$((PORT + 1)) + request_count="$(wc -l < "${requests}")" + if [[ "${RESUME_VALID_CELLS}" == true ]] \ + && has_valid_result "${run_out}/results/result.json" "${request_count}"; then + printf 'CONFIG_REUSED_VALID case=%s trial=%s tp=%s mns=%s\n' \ + "${CASE_NAME}" "${trial}" "${tp}" "${mns}" + return 0 + fi + mkdir -p "${run_out}" + ( + cd "${RUNNER_DIR}" + set +e + env \ + HOME=/tmp/wjh \ + XDG_CACHE_HOME=/tmp/wjh/.cache \ + VLLM_CACHE_ROOT=/tmp/wjh/.cache/vllm \ + CUDA_VISIBLE_DEVICES="${gpus}" \ + TP="${tp}" \ + MNS="${mns}" \ + TRACE_LABEL="${CASE_NAME}/tp${tp}-normalized" \ + PREFIX_CACHING="${PREFIX_CACHING}" \ + SERVER_PORT="${port}" \ + OUTPUT_ROOT="${run_out}" \ + REQUESTS_FILE="${requests}" \ + VENV_ROOT="${VENV_ROOT}" \ + MODEL_ROOT="${MODEL_ROOT}" \ + FLASHINFER_WORKSPACE_BASE="${FLASHINFER_SHARED_WORKSPACE}" \ + SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS}" \ + EXACT_TRACE_CLIENT="${CLIENT}" \ + timeout --signal=TERM --kill-after=60s 4200 bash "${RUNNER}" \ + > "${run_out}/launcher.stdout.log" 2> "${run_out}/launcher.stderr.log" + local_status="$?" + printf '%s\n' "${local_status}" > "${run_out}/launcher.exit_code" + exit "${local_status}" + ) & + WAVE_PIDS+=("$!") +} + +run_tp_wave() { + local trial="$1" tp="$2" order="$3" + IFS=',' read -r -a mnss <<< "${order}" + WAVE_PIDS=() + preflight_gpus + assert_no_our_server + printf 'WAVE_START case=%s trial=%s tp=%s mns=%s\n' "${CASE_NAME}" "${trial}" "${tp}" "${order}" \ + | tee -a "${OUT}/controller.log" + case "${tp}" in + 4) + launch_config "${trial}" 4 "${mnss[0]}" '0,1,2,3' + launch_config "${trial}" 4 "${mnss[1]}" '4,5,6,7' + wait_for_wave + preflight_gpus + assert_no_our_server + launch_config "${trial}" 4 "${mnss[2]}" '0,1,2,3' + launch_config "${trial}" 4 "${mnss[3]}" '4,5,6,7' + ;; + 2) + launch_config "${trial}" 2 "${mnss[0]}" '0,1' + launch_config "${trial}" 2 "${mnss[1]}" '2,3' + launch_config "${trial}" 2 "${mnss[2]}" '4,5' + launch_config "${trial}" 2 "${mnss[3]}" '6,7' + ;; + 1) + launch_config "${trial}" 1 "${mnss[0]}" '0' + launch_config "${trial}" 1 "${mnss[1]}" '1' + launch_config "${trial}" 1 "${mnss[2]}" '2' + launch_config "${trial}" 1 "${mnss[3]}" '3' + ;; + esac + wait_for_wave + preflight_gpus + assert_no_our_server + printf 'WAVE_COMPLETE case=%s trial=%s tp=%s\n' "${CASE_NAME}" "${trial}" "${tp}" \ + | tee -a "${OUT}/controller.log" +} + +{ + printf '%s\n' "REAL_LAUNCH_ECHO host=dash0 model=Qwen3-30B-A3B engine=vLLM-0.20.0+cu129 dtype=BF16 case=${CASE_NAME} traces=${TRACE_ROOT}/tp{1,2,4}/private/real_requests.jsonl prefix=${PREFIX_CACHING} requests=${REQUEST_COUNT} transform=t_prime=t/TP surface=TP{1,2,4}xMNS{8,16,32,64} trials=3 fresh_server=true resume_valid_cells=${RESUME_VALID_CELLS} idle_gpu_memory_tolerance_mib=${IDLE_GPU_MEMORY_TOLERANCE_MIB} metrics=mean,p90(TTFT,TPOT-if-OSL-gt-1,E2E) expected_cost=13_H20-GPUh_nominal__41_H20-GPUh_max output=${OUT}/real" + date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ + mkdir -p "${OUT}/provenance" "${FLASHINFER_SHARED_WORKSPACE}" + sha256sum "${BASH_SOURCE[0]}" "${RUNNER}" "${CLIENT}" "${PREFILL_CLIENT}" \ + "${MODEL_ROOT}/config.json" > "${OUT}/provenance/real-input.sha256" + for tp in 1 2 4; do + sha256sum "${TRACE_ROOT}/tp${tp}/public/manifest.json" \ + "${TRACE_ROOT}/tp${tp}/private/real_requests.jsonl" \ + >> "${OUT}/provenance/real-input.sha256" + done + "${VENV_ROOT}/bin/vllm" --version > "${OUT}/provenance/vllm.version" + "${VENV_ROOT}/bin/python" -c 'import torch, transformers, vllm; print(f"torch={torch.__version__}"); print(f"transformers={transformers.__version__}"); print(f"vllm={vllm.__version__}")' \ + > "${OUT}/provenance/runtime.versions" + nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader \ + > "${OUT}/provenance/gpus.before.csv" + + run_tp_wave 1 4 '8,16,32,64' + run_tp_wave 1 2 '8,16,32,64' + run_tp_wave 1 1 '8,16,32,64' + run_tp_wave 2 1 '64,32,16,8' + run_tp_wave 2 2 '64,32,16,8' + run_tp_wave 2 4 '64,32,16,8' + run_tp_wave 3 2 '16,32,64,8' + run_tp_wave 3 4 '16,32,64,8' + run_tp_wave 3 1 '16,32,64,8' + + nvidia-smi --query-gpu=index,name,uuid,driver_version,memory.total --format=csv,noheader \ + > "${OUT}/provenance/gpus.after.csv" + find "${OUT}/real" -type f ! -path '*/provenance/artifacts.sha256' -print0 \ + | sort -z | xargs -0 sha256sum > "${OUT}/provenance/real-artifacts.sha256" + date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ + printf '%s\n' 'REAL_SURFACE_COMPLETE' +} >> "${OUT}/controller.log" 2>&1 diff --git a/runs/frontier-fidelity-envelope-v1/smoke-report.md b/runs/frontier-fidelity-envelope-v1/smoke-report.md new file mode 100644 index 0000000..f55c31b --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/smoke-report.md @@ -0,0 +1,15 @@ +# Frontier measured-collective smoke + +日期：2026-07-17。设备：local CPU（Frontier simulation only）。Frontier commit：`d9cfeb6d8791fbf2f295dd9744c56a666171776e`，沿用既有 dirty compatibility patch set；A2 额外 patch SHA256 为 `35cc6be846589faf8cb5fa3ce5fdfe0aee8f086ba7dbb5dbcdc677148f19a3c8`。 + +固定 cell：Qwen3-30B-A3B BF16 profile-v2，`TP2/MNS8/MBT8192`，`ISL=2048/OSL=1`，64 requests，8 req/s，prefix off，TTFT SLO 1256 ms。 + +| variant | CC path for 2048-token payload | TTFT p50/p95/max (ms) | pass rate | +|---|---|---:|---:| +| A0 analytical | analytical | 122.2406276 | 1.0 | +| A1 native Vidur | `4,194,304 > 100,000` elements，lookup miss 后 analytical fallback | 122.2406276 | 1.0 | +| A2 measured + direct miss | measured random-forest estimator；exact payload memoized | 120.8898909 | 1.0 | + +A1 与 A0 的所有 TTFT 数值完全一致，验证 measured CSV 虽成功加载和训练，但没有参与该 payload 的 prediction。A2 model 对该 payload 的预测为 `0.09877793 ms`；同一个 TP2/2048 row 的 measured critical-path median 是 `0.10891200 ms`。A2 相对 A1 的 E2E delta 为 `-1.3507367 ms`（`-1.105%`），与每层多次 collective 的累计量级一致，因此通过“execution path 必须变化”的 smoke gate。 + +这个 smoke 只证明 profile consumption；不证明 ranking 已恢复。下一步必须运行完整 A2 TP×MNS surface，并与已冻结 real capacity比较。 diff --git a/runs/frontier-fidelity-envelope-v1/smoke_qwen235_v020_frontier_moe.py b/runs/frontier-fidelity-envelope-v1/smoke_qwen235_v020_frontier_moe.py new file mode 100644 index 0000000..5157422 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/smoke_qwen235_v020_frontier_moe.py @@ -0,0 +1,84 @@ +#!/usr/bin/env python3 +"""One-cell gate for the two Qwen235 vLLM 0.20 MoE runtime backends.""" + +from __future__ import annotations + +import argparse +import json +import sys +from pathlib import Path + +import torch + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--frontier-source", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def routing(tokens: int, experts: int = 128, topk: int = 8): + ids = torch.arange(tokens * topk, device="cuda", dtype=torch.int64) + ids = (ids % experts).view(tokens, topk) + weights = torch.full((tokens, topk), 1.0 / topk, device="cuda") + return weights, ids + + +def main() -> None: + args = parse_args() + sys.path.insert(0, str(args.frontier_source.resolve())) + from frontier.profiling.moe.moe_vllm_kernel import profile_fused_moe_kernel + + weights, ids = routing(8) + cells = [] + cells.append( + { + "name": "tp4_ep1_triton", + "stats": profile_fused_moe_kernel( + num_tokens=8, + num_experts=128, + hidden_dim=4096, + expert_hidden_dim=1536, + top_k=8, + topk_weights=weights, + topk_ids=ids, + tensor_parallel_size=4, + use_fp8=True, + block_shape=[128, 128], + warmup_steps=1, + active_steps=2, + ), + } + ) + expert_map = torch.full((128,), -1, device="cuda", dtype=torch.int32) + expert_map[:16] = torch.arange(16, device="cuda", dtype=torch.int32) + cells.append( + { + "name": "tp1_ep8_flashinfer_cutlass", + "stats": profile_fused_moe_kernel( + num_tokens=8, + num_experts=16, + hidden_dim=4096, + expert_hidden_dim=1536, + top_k=8, + topk_weights=weights, + topk_ids=ids, + tensor_parallel_size=1, + use_fp8=True, + block_shape=[128, 128], + warmup_steps=1, + active_steps=2, + global_num_experts=128, + expert_map=expert_map, + ), + } + ) + payload = {"schema": "qwen235-v020-frontier-moe-smoke-v1", "cells": cells} + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps(payload, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-fidelity-envelope-v1/test_fidelity_envelope.py b/runs/frontier-fidelity-envelope-v1/test_fidelity_envelope.py new file mode 100644 index 0000000..ac0f3e0 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/test_fidelity_envelope.py @@ -0,0 +1,247 @@ +#!/usr/bin/env python3 + +from __future__ import annotations + +import csv +import importlib.util +import json +import sys +import tempfile +import unittest +from pathlib import Path + + +ROOT = Path(__file__).parent +REPO_ROOT = ROOT.parents[1] + + +def load(name: str): + path = ROOT / name + spec = importlib.util.spec_from_file_location(path.stem, path) + assert spec and spec.loader + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +class FidelityEnvelopeTest(unittest.TestCase): + def test_block_identities_are_parent_sensitive_and_prefix_stable(self) -> None: + module = load("prepare_exact_trace.py") + prefix = list(range(32)) + left = module.block_identities(prefix + [100, 101], 16) + right = module.block_identities(prefix + [200, 201], 16) + self.assertEqual(left[:2], right[:2]) + self.assertNotEqual(left[2], right[2]) + changed_parent = module.block_identities([999] + prefix[1:] + [100, 101], 16) + self.assertNotEqual(left[0], changed_parent[0]) + self.assertNotEqual(left[1], changed_parent[1]) + + def test_root_sessions_follow_parent_chain(self) -> None: + module = load("prepare_exact_trace.py") + rows = [ + {"chat_id": 10, "parent_chat_id": -1}, + {"chat_id": 11, "parent_chat_id": 10}, + {"chat_id": 12, "parent_chat_id": 11}, + {"chat_id": 20, "parent_chat_id": -1}, + ] + self.assertEqual(module.root_sessions(rows), {10: 10, 11: 10, 12: 10, 20: 20}) + + def test_materialize_allreduce(self) -> None: + module = load("materialize_frontier_allreduce.py") + rows = [] + for tp in (2, 4): + for tokens in (1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192): + rows.append( + { + "tensor_parallel_size": tp, + "num_tokens": tokens, + "hidden_dim": 2048, + "payload_bytes": tokens * 2048 * 2, + "critical_path_median_ms": tp + tokens / 1000, + } + ) + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + source = root / "allreduce.json" + source.write_text( + json.dumps( + { + "schema_version": "qwen30_vllm020_allreduce_frozen.v1", + "rows": rows, + } + ) + ) + output = root / "all_reduce.csv" + manifest = module.convert(source, output) + self.assertEqual(manifest["rows"], 24) + with output.open(newline="") as handle: + converted = list(csv.DictReader(handle)) + self.assertEqual(converted[0]["num_workers"], "2") + self.assertEqual(converted[0]["size"], "4096") + self.assertEqual(converted[-1]["num_workers"], "4") + self.assertEqual(converted[-1]["size"], str(8192 * 2048 * 2)) + self.assertEqual( + converted[-1]["time_stats.all_reduce.median"], + str(4 + 8192 / 1000), + ) + + def test_batch_profile_retains_one_single_request_anchor_per_tp(self) -> None: + runner = REPO_ROOT / "runs/frontier-phase-factorial-v0/run_frontier_qwen30_prefill_surface.py" + spec = importlib.util.spec_from_file_location("qwen30_surface_runner", runner) + assert spec and spec.loader + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + profile = ROOT / "profiles/profile-v3-batch-final" + coverage = module.validate_profile(module.profile_paths(profile)) + self.assertEqual( + coverage["attention"], + { + "1": {"exact_prefill_2048_rows": 1, "profile_batch_size": 1}, + "2": {"exact_prefill_2048_rows": 1, "profile_batch_size": 1}, + "4": {"exact_prefill_2048_rows": 1, "profile_batch_size": 1}, + }, + ) + + def test_exact_trace_parser_and_joint_slo_score(self) -> None: + module = load("run_frontier_qwen30_exact_trace_surface.py") + self.assertEqual( + module.classify_frontier_failure( + "Sequential simulation ended with non-empty scheduler state" + ), + "scheduler_stall", + ) + self.assertEqual( + module.classify_frontier_failure("unexpected failure"), + "frontier_error", + ) + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + trace = root / "trace.csv" + with trace.open("w", newline="") as handle: + writer = csv.DictWriter( + handle, + fieldnames=[ + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", + ], + lineterminator="\n", + ) + writer.writeheader() + writer.writerows( + [ + { + "arrived_at": 1.0, + "num_prefill_tokens": 512, + "num_decode_tokens": 1, + "session_id": 1, + "block_hash_ids": "10|11", + }, + { + "arrived_at": 2.0, + "num_prefill_tokens": 2048, + "num_decode_tokens": 2, + "session_id": 2, + "block_hash_ids": "20|21", + }, + ] + ) + parsed = module.parse_trace(f"u0p01={trace}") + self.assertEqual(parsed["requests"], 2) + self.assertEqual(parsed["shapes"], [(512, 1), (2048, 2)]) + json.dumps(module.trace_manifest_entry(parsed)) + uniform = module.parse_trace( + f"r1={trace}", rate_contract="uniform-spacing" + ) + self.assertEqual(uniform["offered_request_rate"], 1.0) + + metrics = root / "request_metrics.csv" + with metrics.open("w", newline="") as handle: + writer = csv.DictWriter( + handle, + fieldnames=[ + "Request Id", + "request_num_prefill_tokens", + "request_num_decode_tokens", + "ttft", + "request_e2e_time", + ], + lineterminator="\n", + ) + writer.writeheader() + writer.writerows( + [ + { + "Request Id": 1, + "request_num_prefill_tokens": 2048, + "request_num_decode_tokens": 2, + "ttft": 1200, + "request_e2e_time": 1300, + }, + { + "Request Id": 0, + "request_num_prefill_tokens": 512, + "request_num_decode_tokens": 1, + "ttft": 1000, + "request_e2e_time": 1000, + }, + ] + ) + scored = module.score(metrics, parsed["shapes"]) + self.assertEqual(scored["slos"]["tpot_150ms"]["passed"], 2) + self.assertTrue(scored["slos"]["tpot_150ms"]["feasible"]) + + def test_fixed_trace_materializer_uses_disjoint_prefixes(self) -> None: + module = load("prepare_fixed_frontier_traces.py") + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + manifest = module.materialize( + root, + input_tokens=32, + output_tokens=4, + requests=2, + rates=[2.0], + block_size=16, + ) + with (root / "r2.csv").open(newline="") as handle: + rows = list(csv.DictReader(handle)) + self.assertEqual(rows[0]["block_hash_ids"], "1|2") + self.assertEqual(rows[1]["block_hash_ids"], "3|4") + self.assertEqual(rows[1]["arrived_at"], "0.500000000000") + self.assertEqual(manifest["contract"]["prefix_caching"], False) + steady = module.materialize( + root / "steady", + input_tokens=32, + output_tokens=4, + requests=2, + rates=[2.0], + block_size=16, + duration_seconds=10.0, + ) + self.assertEqual(steady["anchors"][0]["requests"], 20) + + def test_exact_real_client_uses_the_prepared_row_vector_contract(self) -> None: + client = load("qwen30_exact_trace_client.py") + prepare = load("prepare_exact_trace.py") + rows = [ + { + "source_index": 7, + "arrived_at": 1.25, + "input_length": 32, + "output_length": 4, + "session_id": 3, + "runtime_block_ids": [11, 12], + } + ] + digest = client.hashlib.sha256() + prepare.update_digest(digest, [7, 1.25, 32, 4, 3, [11, 12]]) + self.assertEqual(client.row_vector_sha256(rows), digest.hexdigest()) + self.assertEqual(client.ttft_slo_ms(512), 1064.0) + + +if __name__ == "__main__": + unittest.main() diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1-steady10s/manifest.json b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1-steady10s/manifest.json new file mode 100644 index 0000000..744a672 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1-steady10s/manifest.json @@ -0,0 +1,57 @@ +{ + "anchors": [ + { + "label": "r16", + "path": "r16.csv", + "rate": 16.0, + "requests": 160, + "sha256": "931604bdf085b1e0c2c70ffed9d10cbaeb6bfb3523976ce25c9bb3faf5fb20a9" + }, + { + "label": "r32", + "path": "r32.csv", + "rate": 32.0, + "requests": 320, + "sha256": "3f839823181c75149184df43326df4139c1f00cbc607dd305323d34e92467140" + }, + { + "label": "r64", + "path": "r64.csv", + "rate": 64.0, + "requests": 640, + "sha256": "b5bea5cf977f6a875a863a135eca70e643653b07ab92233f933f8b5e44e4a8c7" + }, + { + "label": "r128", + "path": "r128.csv", + "rate": 128.0, + "requests": 1280, + "sha256": "ea94212548621233172e430649e523f494064d035a7609d0b9ec3b1930222a3b" + }, + { + "label": "r256", + "path": "r256.csv", + "rate": 256.0, + "requests": 2560, + "sha256": "1e580f6ca30317f1f1ff81a86f318347341607cf1afcfef7ba53123335f02295" + }, + { + "label": "r512", + "path": "r512.csv", + "rate": 512.0, + "requests": 5120, + "sha256": "992ad52e6e6c08bcb0843d59ff7d529d637ad8ed812aca96d1318cb1ecfeb71b" + } + ], + "contract": { + "arrival": "open_loop_uniform", + "block_size": 16, + "input_tokens": 512, + "minimum_requests_per_anchor": 64, + "minimum_steady_duration_seconds": 10.0, + "output_tokens": 1, + "prefix_caching": false, + "prefix_relation": "all request block identities are disjoint" + }, + "schema": "qwen30-fixed-frontier-traces-v1" +} diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/manifest.json b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/manifest.json new file mode 100644 index 0000000..f765bce --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/manifest.json @@ -0,0 +1,50 @@ +{ + "anchors": [ + { + "label": "r16", + "path": "r16.csv", + "rate": 16.0, + "sha256": "dd00640d57184b1607fb466d2735bd4edf42e585ac6d6f2d6c6363c76b34aac8" + }, + { + "label": "r32", + "path": "r32.csv", + "rate": 32.0, + "sha256": "de448fa237c367c4910570586f8574a5020c3186a2adc1db5daa26acf0d2b9e1" + }, + { + "label": "r64", + "path": "r64.csv", + "rate": 64.0, + "sha256": "c796965d7afa1a9df4b8ba523412cef2e4dc7ddfc3cef6bce7e91265b20aaf2d" + }, + { + "label": "r128", + "path": "r128.csv", + "rate": 128.0, + "sha256": "30706d59e020ad0dcb670d367201de77ca7337b5fb41da0efd5af8c1915b2f04" + }, + { + "label": "r256", + "path": "r256.csv", + "rate": 256.0, + "sha256": "f93c2390f084c4c60e7ca0d3096c7bdf024b7b286fd7ac72f4dc21d3d2af8cbf" + }, + { + "label": "r512", + "path": "r512.csv", + "rate": 512.0, + "sha256": "e2612362defdb10f876eb478a642ef78a9f6101833b7cc4ff1f0b02b8a453767" + } + ], + "contract": { + "arrival": "open_loop_uniform", + "block_size": 16, + "input_tokens": 512, + "output_tokens": 1, + "prefix_caching": false, + "prefix_relation": "all request block identities are disjoint", + "requests": 64 + }, + "schema": "qwen30-fixed-frontier-traces-v1" +} diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r128.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r128.csv new file mode 100644 index 0000000..aff1039 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r128.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.007812500000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.015625000000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.023437500000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.031250000000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.039062500000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.046875000000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.054687500000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.062500000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.070312500000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.078125000000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.085937500000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.093750000000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.101562500000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.109375000000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.117187500000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.125000000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +0.132812500000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +0.140625000000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +0.148437500000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.156250000000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +0.164062500000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +0.171875000000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +0.179687500000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.187500000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +0.195312500000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +0.203125000000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +0.210937500000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.218750000000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +0.226562500000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +0.234375000000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +0.242187500000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +0.250000000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +0.257812500000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +0.265625000000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +0.273437500000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +0.281250000000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +0.289062500000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +0.296875000000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +0.304687500000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +0.312500000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +0.320312500000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +0.328125000000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +0.335937500000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +0.343750000000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +0.351562500000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +0.359375000000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +0.367187500000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +0.375000000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +0.382812500000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +0.390625000000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +0.398437500000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +0.406250000000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +0.414062500000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +0.421875000000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +0.429687500000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +0.437500000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +0.445312500000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +0.453125000000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +0.460937500000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +0.468750000000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +0.476562500000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +0.484375000000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +0.492187500000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r16.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r16.csv new file mode 100644 index 0000000..6123705 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r16.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.062500000000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.125000000000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.187500000000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.250000000000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.312500000000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.375000000000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.437500000000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.500000000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.562500000000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.625000000000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.687500000000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.750000000000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.812500000000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.875000000000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.937500000000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +1.000000000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +1.062500000000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +1.125000000000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +1.187500000000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +1.250000000000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +1.312500000000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +1.375000000000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +1.437500000000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +1.500000000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +1.562500000000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +1.625000000000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +1.687500000000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +1.750000000000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +1.812500000000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +1.875000000000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +1.937500000000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +2.000000000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +2.062500000000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +2.125000000000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +2.187500000000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +2.250000000000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +2.312500000000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +2.375000000000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +2.437500000000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +2.500000000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +2.562500000000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +2.625000000000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +2.687500000000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +2.750000000000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +2.812500000000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +2.875000000000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +2.937500000000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +3.000000000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +3.062500000000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +3.125000000000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +3.187500000000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +3.250000000000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +3.312500000000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +3.375000000000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +3.437500000000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +3.500000000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +3.562500000000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +3.625000000000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +3.687500000000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +3.750000000000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +3.812500000000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +3.875000000000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +3.937500000000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r256.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r256.csv new file mode 100644 index 0000000..c0b0a56 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r256.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.003906250000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.007812500000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.011718750000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.015625000000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.019531250000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.023437500000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.027343750000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.031250000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.035156250000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.039062500000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.042968750000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.046875000000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.050781250000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.054687500000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.058593750000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.062500000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +0.066406250000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +0.070312500000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +0.074218750000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.078125000000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +0.082031250000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +0.085937500000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +0.089843750000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.093750000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +0.097656250000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +0.101562500000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +0.105468750000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.109375000000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +0.113281250000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +0.117187500000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +0.121093750000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +0.125000000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +0.128906250000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +0.132812500000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +0.136718750000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +0.140625000000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +0.144531250000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +0.148437500000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +0.152343750000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +0.156250000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +0.160156250000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +0.164062500000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +0.167968750000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +0.171875000000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +0.175781250000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +0.179687500000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +0.183593750000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +0.187500000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +0.191406250000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +0.195312500000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +0.199218750000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +0.203125000000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +0.207031250000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +0.210937500000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +0.214843750000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +0.218750000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +0.222656250000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +0.226562500000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +0.230468750000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +0.234375000000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +0.238281250000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +0.242187500000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +0.246093750000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r32.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r32.csv new file mode 100644 index 0000000..22a1ea6 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r32.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.031250000000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.062500000000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.093750000000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.125000000000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.156250000000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.187500000000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.218750000000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.250000000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.281250000000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.312500000000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.343750000000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.375000000000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.406250000000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.437500000000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.468750000000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.500000000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +0.531250000000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +0.562500000000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +0.593750000000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.625000000000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +0.656250000000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +0.687500000000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +0.718750000000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.750000000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +0.781250000000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +0.812500000000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +0.843750000000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.875000000000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +0.906250000000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +0.937500000000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +0.968750000000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +1.000000000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +1.031250000000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +1.062500000000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +1.093750000000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +1.125000000000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +1.156250000000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +1.187500000000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +1.218750000000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +1.250000000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +1.281250000000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +1.312500000000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +1.343750000000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +1.375000000000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +1.406250000000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +1.437500000000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +1.468750000000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +1.500000000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +1.531250000000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +1.562500000000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +1.593750000000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +1.625000000000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +1.656250000000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +1.687500000000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +1.718750000000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +1.750000000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +1.781250000000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +1.812500000000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +1.843750000000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +1.875000000000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +1.906250000000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +1.937500000000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +1.968750000000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r512.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r512.csv new file mode 100644 index 0000000..bdbfbee --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r512.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.001953125000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.003906250000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.005859375000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.007812500000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.009765625000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.011718750000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.013671875000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.015625000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.017578125000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.019531250000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.021484375000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.023437500000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.025390625000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.027343750000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.029296875000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.031250000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +0.033203125000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +0.035156250000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +0.037109375000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.039062500000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +0.041015625000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +0.042968750000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +0.044921875000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.046875000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +0.048828125000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +0.050781250000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +0.052734375000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.054687500000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +0.056640625000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +0.058593750000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +0.060546875000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +0.062500000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +0.064453125000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +0.066406250000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +0.068359375000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +0.070312500000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +0.072265625000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +0.074218750000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +0.076171875000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +0.078125000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +0.080078125000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +0.082031250000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +0.083984375000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +0.085937500000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +0.087890625000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +0.089843750000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +0.091796875000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +0.093750000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +0.095703125000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +0.097656250000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +0.099609375000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +0.101562500000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +0.103515625000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +0.105468750000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +0.107421875000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +0.109375000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +0.111328125000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +0.113281250000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +0.115234375000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +0.117187500000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +0.119140625000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +0.121093750000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +0.123046875000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r64.csv b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r64.csv new file mode 100644 index 0000000..db33eff --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f1-isl512-osl1/r64.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,512,1,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32 +0.015625000000,512,1,1,33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64 +0.031250000000,512,1,2,65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96 +0.046875000000,512,1,3,97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.062500000000,512,1,4,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160 +0.078125000000,512,1,5,161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192 +0.093750000000,512,1,6,193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224 +0.109375000000,512,1,7,225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.125000000000,512,1,8,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288 +0.140625000000,512,1,9,289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320 +0.156250000000,512,1,10,321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352 +0.171875000000,512,1,11,353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.187500000000,512,1,12,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416 +0.203125000000,512,1,13,417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448 +0.218750000000,512,1,14,449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480 +0.234375000000,512,1,15,481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.250000000000,512,1,16,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544 +0.265625000000,512,1,17,545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576 +0.281250000000,512,1,18,577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608 +0.296875000000,512,1,19,609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.312500000000,512,1,20,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672 +0.328125000000,512,1,21,673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704 +0.343750000000,512,1,22,705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736 +0.359375000000,512,1,23,737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.375000000000,512,1,24,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800 +0.390625000000,512,1,25,801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832 +0.406250000000,512,1,26,833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864 +0.421875000000,512,1,27,865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.437500000000,512,1,28,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928 +0.453125000000,512,1,29,929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960 +0.468750000000,512,1,30,961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992 +0.484375000000,512,1,31,993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +0.500000000000,512,1,32,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056 +0.515625000000,512,1,33,1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088 +0.531250000000,512,1,34,1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120 +0.546875000000,512,1,35,1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +0.562500000000,512,1,36,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184 +0.578125000000,512,1,37,1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216 +0.593750000000,512,1,38,1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248 +0.609375000000,512,1,39,1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +0.625000000000,512,1,40,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312 +0.640625000000,512,1,41,1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344 +0.656250000000,512,1,42,1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376 +0.671875000000,512,1,43,1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +0.687500000000,512,1,44,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440 +0.703125000000,512,1,45,1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472 +0.718750000000,512,1,46,1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504 +0.734375000000,512,1,47,1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +0.750000000000,512,1,48,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568 +0.765625000000,512,1,49,1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600 +0.781250000000,512,1,50,1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632 +0.796875000000,512,1,51,1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +0.812500000000,512,1,52,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696 +0.828125000000,512,1,53,1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728 +0.843750000000,512,1,54,1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760 +0.859375000000,512,1,55,1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +0.875000000000,512,1,56,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824 +0.890625000000,512,1,57,1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856 +0.906250000000,512,1,58,1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888 +0.921875000000,512,1,59,1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +0.937500000000,512,1,60,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952 +0.953125000000,512,1,61,1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984 +0.968750000000,512,1,62,1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016 +0.984375000000,512,1,63,2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/manifest.json b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/manifest.json new file mode 100644 index 0000000..6cac8e9 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/manifest.json @@ -0,0 +1,50 @@ +{ + "anchors": [ + { + "label": "r0p5", + "path": "r0p5.csv", + "rate": 0.5, + "sha256": "f8564275c4d7b4f2fa28c375b97ebf38634f623abd68ea88ba2a436061ecbb6d" + }, + { + "label": "r1", + "path": "r1.csv", + "rate": 1.0, + "sha256": "5e029026dc0e7dab6801dfa263975c2a9c2720d844ca41367b2aafd373bfea9e" + }, + { + "label": "r2", + "path": "r2.csv", + "rate": 2.0, + "sha256": "215997a1e318e4aea365e6c9cb87d36620342e128e2cae3e27ac3aef48908900" + }, + { + "label": "r4", + "path": "r4.csv", + "rate": 4.0, + "sha256": "d69a8a9d2b946e4c567ac77b75738bb07ec91b189651a23f3a323f90977cda8d" + }, + { + "label": "r8", + "path": "r8.csv", + "rate": 8.0, + "sha256": "51341a2520c0c2e14f5d7c110083d861b75d7b712ec628b44d07f785bc22a4d2" + }, + { + "label": "r16", + "path": "r16.csv", + "rate": 16.0, + "sha256": "f6d7522e766e0c049118245a9f7c0686fe4e6dc597a871ffc7cc447e0af90aaa" + } + ], + "contract": { + "arrival": "open_loop_uniform", + "block_size": 16, + "input_tokens": 2048, + "output_tokens": 128, + "prefix_caching": false, + "prefix_relation": "all request block identities are disjoint", + "requests": 64 + }, + "schema": "qwen30-fixed-frontier-traces-v1" +} diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r0p5.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r0p5.csv new file mode 100644 index 0000000..1969175 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r0p5.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +2.000000000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +4.000000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +6.000000000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +8.000000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +10.000000000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +12.000000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +14.000000000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +16.000000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +18.000000000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +20.000000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +22.000000000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +24.000000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +26.000000000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +28.000000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +30.000000000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +32.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +34.000000000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +36.000000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +38.000000000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +40.000000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +42.000000000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +44.000000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +46.000000000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +48.000000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +50.000000000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +52.000000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +54.000000000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +56.000000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +58.000000000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +60.000000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +62.000000000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +64.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +66.000000000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +68.000000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +70.000000000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +72.000000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +74.000000000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +76.000000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +78.000000000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +80.000000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +82.000000000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +84.000000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +86.000000000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +88.000000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +90.000000000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +92.000000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +94.000000000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +96.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +98.000000000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +100.000000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +102.000000000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +104.000000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +106.000000000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +108.000000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +110.000000000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +112.000000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +114.000000000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +116.000000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +118.000000000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +120.000000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +122.000000000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +124.000000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +126.000000000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r1.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r1.csv new file mode 100644 index 0000000..2fa8c5e --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r1.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +1.000000000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +2.000000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +3.000000000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +4.000000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +5.000000000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +6.000000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +7.000000000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +8.000000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +9.000000000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +10.000000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +11.000000000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +12.000000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +13.000000000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +14.000000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +15.000000000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +16.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +17.000000000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +18.000000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +19.000000000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +20.000000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +21.000000000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +22.000000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +23.000000000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +24.000000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +25.000000000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +26.000000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +27.000000000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +28.000000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +29.000000000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +30.000000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +31.000000000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +32.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +33.000000000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +34.000000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +35.000000000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +36.000000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +37.000000000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +38.000000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +39.000000000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +40.000000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +41.000000000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +42.000000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +43.000000000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +44.000000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +45.000000000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +46.000000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +47.000000000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +48.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +49.000000000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +50.000000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +51.000000000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +52.000000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +53.000000000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +54.000000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +55.000000000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +56.000000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +57.000000000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +58.000000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +59.000000000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +60.000000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +61.000000000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +62.000000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +63.000000000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r16.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r16.csv new file mode 100644 index 0000000..5d32a58 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r16.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.062500000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.125000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.187500000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.250000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.312500000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.375000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.437500000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +0.500000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +0.562500000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +0.625000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +0.687500000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +0.750000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +0.812500000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +0.875000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +0.937500000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +1.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +1.062500000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +1.125000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +1.187500000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +1.250000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +1.312500000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +1.375000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +1.437500000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +1.500000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +1.562500000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +1.625000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +1.687500000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +1.750000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +1.812500000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +1.875000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +1.937500000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +2.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +2.062500000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +2.125000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +2.187500000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +2.250000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +2.312500000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +2.375000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +2.437500000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +2.500000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +2.562500000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +2.625000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +2.687500000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +2.750000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +2.812500000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +2.875000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +2.937500000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +3.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +3.062500000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +3.125000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +3.187500000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +3.250000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +3.312500000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +3.375000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +3.437500000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +3.500000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +3.562500000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +3.625000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +3.687500000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +3.750000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +3.812500000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +3.875000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +3.937500000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r2.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r2.csv new file mode 100644 index 0000000..5a55e4e --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r2.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.500000000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +1.000000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +1.500000000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +2.000000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +2.500000000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +3.000000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +3.500000000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +4.000000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +4.500000000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +5.000000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +5.500000000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +6.000000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +6.500000000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +7.000000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +7.500000000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +8.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +8.500000000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +9.000000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +9.500000000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +10.000000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +10.500000000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +11.000000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +11.500000000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +12.000000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +12.500000000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +13.000000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +13.500000000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +14.000000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +14.500000000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +15.000000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +15.500000000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +16.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +16.500000000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +17.000000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +17.500000000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +18.000000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +18.500000000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +19.000000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +19.500000000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +20.000000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +20.500000000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +21.000000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +21.500000000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +22.000000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +22.500000000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +23.000000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +23.500000000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +24.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +24.500000000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +25.000000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +25.500000000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +26.000000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +26.500000000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +27.000000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +27.500000000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +28.000000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +28.500000000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +29.000000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +29.500000000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +30.000000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +30.500000000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +31.000000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +31.500000000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r4.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r4.csv new file mode 100644 index 0000000..859ffcd --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r4.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.250000000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.500000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.750000000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +1.000000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +1.250000000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +1.500000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +1.750000000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +2.000000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +2.250000000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +2.500000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +2.750000000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +3.000000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +3.250000000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +3.500000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +3.750000000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +4.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +4.250000000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +4.500000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +4.750000000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +5.000000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +5.250000000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +5.500000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +5.750000000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +6.000000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +6.250000000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +6.500000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +6.750000000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +7.000000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +7.250000000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +7.500000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +7.750000000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +8.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +8.250000000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +8.500000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +8.750000000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +9.000000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +9.250000000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +9.500000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +9.750000000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +10.000000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +10.250000000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +10.500000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +10.750000000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +11.000000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +11.250000000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +11.500000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +11.750000000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +12.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +12.250000000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +12.500000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +12.750000000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +13.000000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +13.250000000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +13.500000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +13.750000000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +14.000000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +14.250000000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +14.500000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +14.750000000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +15.000000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +15.250000000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +15.500000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +15.750000000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r8.csv b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r8.csv new file mode 100644 index 0000000..5469301 --- /dev/null +++ b/runs/frontier-fidelity-envelope-v1/workloads/f2-isl2048-osl128/r8.csv @@ -0,0 +1,65 @@ +arrived_at,num_prefill_tokens,num_decode_tokens,session_id,block_hash_ids +0.000000000000,2048,128,0,1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30|31|32|33|34|35|36|37|38|39|40|41|42|43|44|45|46|47|48|49|50|51|52|53|54|55|56|57|58|59|60|61|62|63|64|65|66|67|68|69|70|71|72|73|74|75|76|77|78|79|80|81|82|83|84|85|86|87|88|89|90|91|92|93|94|95|96|97|98|99|100|101|102|103|104|105|106|107|108|109|110|111|112|113|114|115|116|117|118|119|120|121|122|123|124|125|126|127|128 +0.125000000000,2048,128,1,129|130|131|132|133|134|135|136|137|138|139|140|141|142|143|144|145|146|147|148|149|150|151|152|153|154|155|156|157|158|159|160|161|162|163|164|165|166|167|168|169|170|171|172|173|174|175|176|177|178|179|180|181|182|183|184|185|186|187|188|189|190|191|192|193|194|195|196|197|198|199|200|201|202|203|204|205|206|207|208|209|210|211|212|213|214|215|216|217|218|219|220|221|222|223|224|225|226|227|228|229|230|231|232|233|234|235|236|237|238|239|240|241|242|243|244|245|246|247|248|249|250|251|252|253|254|255|256 +0.250000000000,2048,128,2,257|258|259|260|261|262|263|264|265|266|267|268|269|270|271|272|273|274|275|276|277|278|279|280|281|282|283|284|285|286|287|288|289|290|291|292|293|294|295|296|297|298|299|300|301|302|303|304|305|306|307|308|309|310|311|312|313|314|315|316|317|318|319|320|321|322|323|324|325|326|327|328|329|330|331|332|333|334|335|336|337|338|339|340|341|342|343|344|345|346|347|348|349|350|351|352|353|354|355|356|357|358|359|360|361|362|363|364|365|366|367|368|369|370|371|372|373|374|375|376|377|378|379|380|381|382|383|384 +0.375000000000,2048,128,3,385|386|387|388|389|390|391|392|393|394|395|396|397|398|399|400|401|402|403|404|405|406|407|408|409|410|411|412|413|414|415|416|417|418|419|420|421|422|423|424|425|426|427|428|429|430|431|432|433|434|435|436|437|438|439|440|441|442|443|444|445|446|447|448|449|450|451|452|453|454|455|456|457|458|459|460|461|462|463|464|465|466|467|468|469|470|471|472|473|474|475|476|477|478|479|480|481|482|483|484|485|486|487|488|489|490|491|492|493|494|495|496|497|498|499|500|501|502|503|504|505|506|507|508|509|510|511|512 +0.500000000000,2048,128,4,513|514|515|516|517|518|519|520|521|522|523|524|525|526|527|528|529|530|531|532|533|534|535|536|537|538|539|540|541|542|543|544|545|546|547|548|549|550|551|552|553|554|555|556|557|558|559|560|561|562|563|564|565|566|567|568|569|570|571|572|573|574|575|576|577|578|579|580|581|582|583|584|585|586|587|588|589|590|591|592|593|594|595|596|597|598|599|600|601|602|603|604|605|606|607|608|609|610|611|612|613|614|615|616|617|618|619|620|621|622|623|624|625|626|627|628|629|630|631|632|633|634|635|636|637|638|639|640 +0.625000000000,2048,128,5,641|642|643|644|645|646|647|648|649|650|651|652|653|654|655|656|657|658|659|660|661|662|663|664|665|666|667|668|669|670|671|672|673|674|675|676|677|678|679|680|681|682|683|684|685|686|687|688|689|690|691|692|693|694|695|696|697|698|699|700|701|702|703|704|705|706|707|708|709|710|711|712|713|714|715|716|717|718|719|720|721|722|723|724|725|726|727|728|729|730|731|732|733|734|735|736|737|738|739|740|741|742|743|744|745|746|747|748|749|750|751|752|753|754|755|756|757|758|759|760|761|762|763|764|765|766|767|768 +0.750000000000,2048,128,6,769|770|771|772|773|774|775|776|777|778|779|780|781|782|783|784|785|786|787|788|789|790|791|792|793|794|795|796|797|798|799|800|801|802|803|804|805|806|807|808|809|810|811|812|813|814|815|816|817|818|819|820|821|822|823|824|825|826|827|828|829|830|831|832|833|834|835|836|837|838|839|840|841|842|843|844|845|846|847|848|849|850|851|852|853|854|855|856|857|858|859|860|861|862|863|864|865|866|867|868|869|870|871|872|873|874|875|876|877|878|879|880|881|882|883|884|885|886|887|888|889|890|891|892|893|894|895|896 +0.875000000000,2048,128,7,897|898|899|900|901|902|903|904|905|906|907|908|909|910|911|912|913|914|915|916|917|918|919|920|921|922|923|924|925|926|927|928|929|930|931|932|933|934|935|936|937|938|939|940|941|942|943|944|945|946|947|948|949|950|951|952|953|954|955|956|957|958|959|960|961|962|963|964|965|966|967|968|969|970|971|972|973|974|975|976|977|978|979|980|981|982|983|984|985|986|987|988|989|990|991|992|993|994|995|996|997|998|999|1000|1001|1002|1003|1004|1005|1006|1007|1008|1009|1010|1011|1012|1013|1014|1015|1016|1017|1018|1019|1020|1021|1022|1023|1024 +1.000000000000,2048,128,8,1025|1026|1027|1028|1029|1030|1031|1032|1033|1034|1035|1036|1037|1038|1039|1040|1041|1042|1043|1044|1045|1046|1047|1048|1049|1050|1051|1052|1053|1054|1055|1056|1057|1058|1059|1060|1061|1062|1063|1064|1065|1066|1067|1068|1069|1070|1071|1072|1073|1074|1075|1076|1077|1078|1079|1080|1081|1082|1083|1084|1085|1086|1087|1088|1089|1090|1091|1092|1093|1094|1095|1096|1097|1098|1099|1100|1101|1102|1103|1104|1105|1106|1107|1108|1109|1110|1111|1112|1113|1114|1115|1116|1117|1118|1119|1120|1121|1122|1123|1124|1125|1126|1127|1128|1129|1130|1131|1132|1133|1134|1135|1136|1137|1138|1139|1140|1141|1142|1143|1144|1145|1146|1147|1148|1149|1150|1151|1152 +1.125000000000,2048,128,9,1153|1154|1155|1156|1157|1158|1159|1160|1161|1162|1163|1164|1165|1166|1167|1168|1169|1170|1171|1172|1173|1174|1175|1176|1177|1178|1179|1180|1181|1182|1183|1184|1185|1186|1187|1188|1189|1190|1191|1192|1193|1194|1195|1196|1197|1198|1199|1200|1201|1202|1203|1204|1205|1206|1207|1208|1209|1210|1211|1212|1213|1214|1215|1216|1217|1218|1219|1220|1221|1222|1223|1224|1225|1226|1227|1228|1229|1230|1231|1232|1233|1234|1235|1236|1237|1238|1239|1240|1241|1242|1243|1244|1245|1246|1247|1248|1249|1250|1251|1252|1253|1254|1255|1256|1257|1258|1259|1260|1261|1262|1263|1264|1265|1266|1267|1268|1269|1270|1271|1272|1273|1274|1275|1276|1277|1278|1279|1280 +1.250000000000,2048,128,10,1281|1282|1283|1284|1285|1286|1287|1288|1289|1290|1291|1292|1293|1294|1295|1296|1297|1298|1299|1300|1301|1302|1303|1304|1305|1306|1307|1308|1309|1310|1311|1312|1313|1314|1315|1316|1317|1318|1319|1320|1321|1322|1323|1324|1325|1326|1327|1328|1329|1330|1331|1332|1333|1334|1335|1336|1337|1338|1339|1340|1341|1342|1343|1344|1345|1346|1347|1348|1349|1350|1351|1352|1353|1354|1355|1356|1357|1358|1359|1360|1361|1362|1363|1364|1365|1366|1367|1368|1369|1370|1371|1372|1373|1374|1375|1376|1377|1378|1379|1380|1381|1382|1383|1384|1385|1386|1387|1388|1389|1390|1391|1392|1393|1394|1395|1396|1397|1398|1399|1400|1401|1402|1403|1404|1405|1406|1407|1408 +1.375000000000,2048,128,11,1409|1410|1411|1412|1413|1414|1415|1416|1417|1418|1419|1420|1421|1422|1423|1424|1425|1426|1427|1428|1429|1430|1431|1432|1433|1434|1435|1436|1437|1438|1439|1440|1441|1442|1443|1444|1445|1446|1447|1448|1449|1450|1451|1452|1453|1454|1455|1456|1457|1458|1459|1460|1461|1462|1463|1464|1465|1466|1467|1468|1469|1470|1471|1472|1473|1474|1475|1476|1477|1478|1479|1480|1481|1482|1483|1484|1485|1486|1487|1488|1489|1490|1491|1492|1493|1494|1495|1496|1497|1498|1499|1500|1501|1502|1503|1504|1505|1506|1507|1508|1509|1510|1511|1512|1513|1514|1515|1516|1517|1518|1519|1520|1521|1522|1523|1524|1525|1526|1527|1528|1529|1530|1531|1532|1533|1534|1535|1536 +1.500000000000,2048,128,12,1537|1538|1539|1540|1541|1542|1543|1544|1545|1546|1547|1548|1549|1550|1551|1552|1553|1554|1555|1556|1557|1558|1559|1560|1561|1562|1563|1564|1565|1566|1567|1568|1569|1570|1571|1572|1573|1574|1575|1576|1577|1578|1579|1580|1581|1582|1583|1584|1585|1586|1587|1588|1589|1590|1591|1592|1593|1594|1595|1596|1597|1598|1599|1600|1601|1602|1603|1604|1605|1606|1607|1608|1609|1610|1611|1612|1613|1614|1615|1616|1617|1618|1619|1620|1621|1622|1623|1624|1625|1626|1627|1628|1629|1630|1631|1632|1633|1634|1635|1636|1637|1638|1639|1640|1641|1642|1643|1644|1645|1646|1647|1648|1649|1650|1651|1652|1653|1654|1655|1656|1657|1658|1659|1660|1661|1662|1663|1664 +1.625000000000,2048,128,13,1665|1666|1667|1668|1669|1670|1671|1672|1673|1674|1675|1676|1677|1678|1679|1680|1681|1682|1683|1684|1685|1686|1687|1688|1689|1690|1691|1692|1693|1694|1695|1696|1697|1698|1699|1700|1701|1702|1703|1704|1705|1706|1707|1708|1709|1710|1711|1712|1713|1714|1715|1716|1717|1718|1719|1720|1721|1722|1723|1724|1725|1726|1727|1728|1729|1730|1731|1732|1733|1734|1735|1736|1737|1738|1739|1740|1741|1742|1743|1744|1745|1746|1747|1748|1749|1750|1751|1752|1753|1754|1755|1756|1757|1758|1759|1760|1761|1762|1763|1764|1765|1766|1767|1768|1769|1770|1771|1772|1773|1774|1775|1776|1777|1778|1779|1780|1781|1782|1783|1784|1785|1786|1787|1788|1789|1790|1791|1792 +1.750000000000,2048,128,14,1793|1794|1795|1796|1797|1798|1799|1800|1801|1802|1803|1804|1805|1806|1807|1808|1809|1810|1811|1812|1813|1814|1815|1816|1817|1818|1819|1820|1821|1822|1823|1824|1825|1826|1827|1828|1829|1830|1831|1832|1833|1834|1835|1836|1837|1838|1839|1840|1841|1842|1843|1844|1845|1846|1847|1848|1849|1850|1851|1852|1853|1854|1855|1856|1857|1858|1859|1860|1861|1862|1863|1864|1865|1866|1867|1868|1869|1870|1871|1872|1873|1874|1875|1876|1877|1878|1879|1880|1881|1882|1883|1884|1885|1886|1887|1888|1889|1890|1891|1892|1893|1894|1895|1896|1897|1898|1899|1900|1901|1902|1903|1904|1905|1906|1907|1908|1909|1910|1911|1912|1913|1914|1915|1916|1917|1918|1919|1920 +1.875000000000,2048,128,15,1921|1922|1923|1924|1925|1926|1927|1928|1929|1930|1931|1932|1933|1934|1935|1936|1937|1938|1939|1940|1941|1942|1943|1944|1945|1946|1947|1948|1949|1950|1951|1952|1953|1954|1955|1956|1957|1958|1959|1960|1961|1962|1963|1964|1965|1966|1967|1968|1969|1970|1971|1972|1973|1974|1975|1976|1977|1978|1979|1980|1981|1982|1983|1984|1985|1986|1987|1988|1989|1990|1991|1992|1993|1994|1995|1996|1997|1998|1999|2000|2001|2002|2003|2004|2005|2006|2007|2008|2009|2010|2011|2012|2013|2014|2015|2016|2017|2018|2019|2020|2021|2022|2023|2024|2025|2026|2027|2028|2029|2030|2031|2032|2033|2034|2035|2036|2037|2038|2039|2040|2041|2042|2043|2044|2045|2046|2047|2048 +2.000000000000,2048,128,16,2049|2050|2051|2052|2053|2054|2055|2056|2057|2058|2059|2060|2061|2062|2063|2064|2065|2066|2067|2068|2069|2070|2071|2072|2073|2074|2075|2076|2077|2078|2079|2080|2081|2082|2083|2084|2085|2086|2087|2088|2089|2090|2091|2092|2093|2094|2095|2096|2097|2098|2099|2100|2101|2102|2103|2104|2105|2106|2107|2108|2109|2110|2111|2112|2113|2114|2115|2116|2117|2118|2119|2120|2121|2122|2123|2124|2125|2126|2127|2128|2129|2130|2131|2132|2133|2134|2135|2136|2137|2138|2139|2140|2141|2142|2143|2144|2145|2146|2147|2148|2149|2150|2151|2152|2153|2154|2155|2156|2157|2158|2159|2160|2161|2162|2163|2164|2165|2166|2167|2168|2169|2170|2171|2172|2173|2174|2175|2176 +2.125000000000,2048,128,17,2177|2178|2179|2180|2181|2182|2183|2184|2185|2186|2187|2188|2189|2190|2191|2192|2193|2194|2195|2196|2197|2198|2199|2200|2201|2202|2203|2204|2205|2206|2207|2208|2209|2210|2211|2212|2213|2214|2215|2216|2217|2218|2219|2220|2221|2222|2223|2224|2225|2226|2227|2228|2229|2230|2231|2232|2233|2234|2235|2236|2237|2238|2239|2240|2241|2242|2243|2244|2245|2246|2247|2248|2249|2250|2251|2252|2253|2254|2255|2256|2257|2258|2259|2260|2261|2262|2263|2264|2265|2266|2267|2268|2269|2270|2271|2272|2273|2274|2275|2276|2277|2278|2279|2280|2281|2282|2283|2284|2285|2286|2287|2288|2289|2290|2291|2292|2293|2294|2295|2296|2297|2298|2299|2300|2301|2302|2303|2304 +2.250000000000,2048,128,18,2305|2306|2307|2308|2309|2310|2311|2312|2313|2314|2315|2316|2317|2318|2319|2320|2321|2322|2323|2324|2325|2326|2327|2328|2329|2330|2331|2332|2333|2334|2335|2336|2337|2338|2339|2340|2341|2342|2343|2344|2345|2346|2347|2348|2349|2350|2351|2352|2353|2354|2355|2356|2357|2358|2359|2360|2361|2362|2363|2364|2365|2366|2367|2368|2369|2370|2371|2372|2373|2374|2375|2376|2377|2378|2379|2380|2381|2382|2383|2384|2385|2386|2387|2388|2389|2390|2391|2392|2393|2394|2395|2396|2397|2398|2399|2400|2401|2402|2403|2404|2405|2406|2407|2408|2409|2410|2411|2412|2413|2414|2415|2416|2417|2418|2419|2420|2421|2422|2423|2424|2425|2426|2427|2428|2429|2430|2431|2432 +2.375000000000,2048,128,19,2433|2434|2435|2436|2437|2438|2439|2440|2441|2442|2443|2444|2445|2446|2447|2448|2449|2450|2451|2452|2453|2454|2455|2456|2457|2458|2459|2460|2461|2462|2463|2464|2465|2466|2467|2468|2469|2470|2471|2472|2473|2474|2475|2476|2477|2478|2479|2480|2481|2482|2483|2484|2485|2486|2487|2488|2489|2490|2491|2492|2493|2494|2495|2496|2497|2498|2499|2500|2501|2502|2503|2504|2505|2506|2507|2508|2509|2510|2511|2512|2513|2514|2515|2516|2517|2518|2519|2520|2521|2522|2523|2524|2525|2526|2527|2528|2529|2530|2531|2532|2533|2534|2535|2536|2537|2538|2539|2540|2541|2542|2543|2544|2545|2546|2547|2548|2549|2550|2551|2552|2553|2554|2555|2556|2557|2558|2559|2560 +2.500000000000,2048,128,20,2561|2562|2563|2564|2565|2566|2567|2568|2569|2570|2571|2572|2573|2574|2575|2576|2577|2578|2579|2580|2581|2582|2583|2584|2585|2586|2587|2588|2589|2590|2591|2592|2593|2594|2595|2596|2597|2598|2599|2600|2601|2602|2603|2604|2605|2606|2607|2608|2609|2610|2611|2612|2613|2614|2615|2616|2617|2618|2619|2620|2621|2622|2623|2624|2625|2626|2627|2628|2629|2630|2631|2632|2633|2634|2635|2636|2637|2638|2639|2640|2641|2642|2643|2644|2645|2646|2647|2648|2649|2650|2651|2652|2653|2654|2655|2656|2657|2658|2659|2660|2661|2662|2663|2664|2665|2666|2667|2668|2669|2670|2671|2672|2673|2674|2675|2676|2677|2678|2679|2680|2681|2682|2683|2684|2685|2686|2687|2688 +2.625000000000,2048,128,21,2689|2690|2691|2692|2693|2694|2695|2696|2697|2698|2699|2700|2701|2702|2703|2704|2705|2706|2707|2708|2709|2710|2711|2712|2713|2714|2715|2716|2717|2718|2719|2720|2721|2722|2723|2724|2725|2726|2727|2728|2729|2730|2731|2732|2733|2734|2735|2736|2737|2738|2739|2740|2741|2742|2743|2744|2745|2746|2747|2748|2749|2750|2751|2752|2753|2754|2755|2756|2757|2758|2759|2760|2761|2762|2763|2764|2765|2766|2767|2768|2769|2770|2771|2772|2773|2774|2775|2776|2777|2778|2779|2780|2781|2782|2783|2784|2785|2786|2787|2788|2789|2790|2791|2792|2793|2794|2795|2796|2797|2798|2799|2800|2801|2802|2803|2804|2805|2806|2807|2808|2809|2810|2811|2812|2813|2814|2815|2816 +2.750000000000,2048,128,22,2817|2818|2819|2820|2821|2822|2823|2824|2825|2826|2827|2828|2829|2830|2831|2832|2833|2834|2835|2836|2837|2838|2839|2840|2841|2842|2843|2844|2845|2846|2847|2848|2849|2850|2851|2852|2853|2854|2855|2856|2857|2858|2859|2860|2861|2862|2863|2864|2865|2866|2867|2868|2869|2870|2871|2872|2873|2874|2875|2876|2877|2878|2879|2880|2881|2882|2883|2884|2885|2886|2887|2888|2889|2890|2891|2892|2893|2894|2895|2896|2897|2898|2899|2900|2901|2902|2903|2904|2905|2906|2907|2908|2909|2910|2911|2912|2913|2914|2915|2916|2917|2918|2919|2920|2921|2922|2923|2924|2925|2926|2927|2928|2929|2930|2931|2932|2933|2934|2935|2936|2937|2938|2939|2940|2941|2942|2943|2944 +2.875000000000,2048,128,23,2945|2946|2947|2948|2949|2950|2951|2952|2953|2954|2955|2956|2957|2958|2959|2960|2961|2962|2963|2964|2965|2966|2967|2968|2969|2970|2971|2972|2973|2974|2975|2976|2977|2978|2979|2980|2981|2982|2983|2984|2985|2986|2987|2988|2989|2990|2991|2992|2993|2994|2995|2996|2997|2998|2999|3000|3001|3002|3003|3004|3005|3006|3007|3008|3009|3010|3011|3012|3013|3014|3015|3016|3017|3018|3019|3020|3021|3022|3023|3024|3025|3026|3027|3028|3029|3030|3031|3032|3033|3034|3035|3036|3037|3038|3039|3040|3041|3042|3043|3044|3045|3046|3047|3048|3049|3050|3051|3052|3053|3054|3055|3056|3057|3058|3059|3060|3061|3062|3063|3064|3065|3066|3067|3068|3069|3070|3071|3072 +3.000000000000,2048,128,24,3073|3074|3075|3076|3077|3078|3079|3080|3081|3082|3083|3084|3085|3086|3087|3088|3089|3090|3091|3092|3093|3094|3095|3096|3097|3098|3099|3100|3101|3102|3103|3104|3105|3106|3107|3108|3109|3110|3111|3112|3113|3114|3115|3116|3117|3118|3119|3120|3121|3122|3123|3124|3125|3126|3127|3128|3129|3130|3131|3132|3133|3134|3135|3136|3137|3138|3139|3140|3141|3142|3143|3144|3145|3146|3147|3148|3149|3150|3151|3152|3153|3154|3155|3156|3157|3158|3159|3160|3161|3162|3163|3164|3165|3166|3167|3168|3169|3170|3171|3172|3173|3174|3175|3176|3177|3178|3179|3180|3181|3182|3183|3184|3185|3186|3187|3188|3189|3190|3191|3192|3193|3194|3195|3196|3197|3198|3199|3200 +3.125000000000,2048,128,25,3201|3202|3203|3204|3205|3206|3207|3208|3209|3210|3211|3212|3213|3214|3215|3216|3217|3218|3219|3220|3221|3222|3223|3224|3225|3226|3227|3228|3229|3230|3231|3232|3233|3234|3235|3236|3237|3238|3239|3240|3241|3242|3243|3244|3245|3246|3247|3248|3249|3250|3251|3252|3253|3254|3255|3256|3257|3258|3259|3260|3261|3262|3263|3264|3265|3266|3267|3268|3269|3270|3271|3272|3273|3274|3275|3276|3277|3278|3279|3280|3281|3282|3283|3284|3285|3286|3287|3288|3289|3290|3291|3292|3293|3294|3295|3296|3297|3298|3299|3300|3301|3302|3303|3304|3305|3306|3307|3308|3309|3310|3311|3312|3313|3314|3315|3316|3317|3318|3319|3320|3321|3322|3323|3324|3325|3326|3327|3328 +3.250000000000,2048,128,26,3329|3330|3331|3332|3333|3334|3335|3336|3337|3338|3339|3340|3341|3342|3343|3344|3345|3346|3347|3348|3349|3350|3351|3352|3353|3354|3355|3356|3357|3358|3359|3360|3361|3362|3363|3364|3365|3366|3367|3368|3369|3370|3371|3372|3373|3374|3375|3376|3377|3378|3379|3380|3381|3382|3383|3384|3385|3386|3387|3388|3389|3390|3391|3392|3393|3394|3395|3396|3397|3398|3399|3400|3401|3402|3403|3404|3405|3406|3407|3408|3409|3410|3411|3412|3413|3414|3415|3416|3417|3418|3419|3420|3421|3422|3423|3424|3425|3426|3427|3428|3429|3430|3431|3432|3433|3434|3435|3436|3437|3438|3439|3440|3441|3442|3443|3444|3445|3446|3447|3448|3449|3450|3451|3452|3453|3454|3455|3456 +3.375000000000,2048,128,27,3457|3458|3459|3460|3461|3462|3463|3464|3465|3466|3467|3468|3469|3470|3471|3472|3473|3474|3475|3476|3477|3478|3479|3480|3481|3482|3483|3484|3485|3486|3487|3488|3489|3490|3491|3492|3493|3494|3495|3496|3497|3498|3499|3500|3501|3502|3503|3504|3505|3506|3507|3508|3509|3510|3511|3512|3513|3514|3515|3516|3517|3518|3519|3520|3521|3522|3523|3524|3525|3526|3527|3528|3529|3530|3531|3532|3533|3534|3535|3536|3537|3538|3539|3540|3541|3542|3543|3544|3545|3546|3547|3548|3549|3550|3551|3552|3553|3554|3555|3556|3557|3558|3559|3560|3561|3562|3563|3564|3565|3566|3567|3568|3569|3570|3571|3572|3573|3574|3575|3576|3577|3578|3579|3580|3581|3582|3583|3584 +3.500000000000,2048,128,28,3585|3586|3587|3588|3589|3590|3591|3592|3593|3594|3595|3596|3597|3598|3599|3600|3601|3602|3603|3604|3605|3606|3607|3608|3609|3610|3611|3612|3613|3614|3615|3616|3617|3618|3619|3620|3621|3622|3623|3624|3625|3626|3627|3628|3629|3630|3631|3632|3633|3634|3635|3636|3637|3638|3639|3640|3641|3642|3643|3644|3645|3646|3647|3648|3649|3650|3651|3652|3653|3654|3655|3656|3657|3658|3659|3660|3661|3662|3663|3664|3665|3666|3667|3668|3669|3670|3671|3672|3673|3674|3675|3676|3677|3678|3679|3680|3681|3682|3683|3684|3685|3686|3687|3688|3689|3690|3691|3692|3693|3694|3695|3696|3697|3698|3699|3700|3701|3702|3703|3704|3705|3706|3707|3708|3709|3710|3711|3712 +3.625000000000,2048,128,29,3713|3714|3715|3716|3717|3718|3719|3720|3721|3722|3723|3724|3725|3726|3727|3728|3729|3730|3731|3732|3733|3734|3735|3736|3737|3738|3739|3740|3741|3742|3743|3744|3745|3746|3747|3748|3749|3750|3751|3752|3753|3754|3755|3756|3757|3758|3759|3760|3761|3762|3763|3764|3765|3766|3767|3768|3769|3770|3771|3772|3773|3774|3775|3776|3777|3778|3779|3780|3781|3782|3783|3784|3785|3786|3787|3788|3789|3790|3791|3792|3793|3794|3795|3796|3797|3798|3799|3800|3801|3802|3803|3804|3805|3806|3807|3808|3809|3810|3811|3812|3813|3814|3815|3816|3817|3818|3819|3820|3821|3822|3823|3824|3825|3826|3827|3828|3829|3830|3831|3832|3833|3834|3835|3836|3837|3838|3839|3840 +3.750000000000,2048,128,30,3841|3842|3843|3844|3845|3846|3847|3848|3849|3850|3851|3852|3853|3854|3855|3856|3857|3858|3859|3860|3861|3862|3863|3864|3865|3866|3867|3868|3869|3870|3871|3872|3873|3874|3875|3876|3877|3878|3879|3880|3881|3882|3883|3884|3885|3886|3887|3888|3889|3890|3891|3892|3893|3894|3895|3896|3897|3898|3899|3900|3901|3902|3903|3904|3905|3906|3907|3908|3909|3910|3911|3912|3913|3914|3915|3916|3917|3918|3919|3920|3921|3922|3923|3924|3925|3926|3927|3928|3929|3930|3931|3932|3933|3934|3935|3936|3937|3938|3939|3940|3941|3942|3943|3944|3945|3946|3947|3948|3949|3950|3951|3952|3953|3954|3955|3956|3957|3958|3959|3960|3961|3962|3963|3964|3965|3966|3967|3968 +3.875000000000,2048,128,31,3969|3970|3971|3972|3973|3974|3975|3976|3977|3978|3979|3980|3981|3982|3983|3984|3985|3986|3987|3988|3989|3990|3991|3992|3993|3994|3995|3996|3997|3998|3999|4000|4001|4002|4003|4004|4005|4006|4007|4008|4009|4010|4011|4012|4013|4014|4015|4016|4017|4018|4019|4020|4021|4022|4023|4024|4025|4026|4027|4028|4029|4030|4031|4032|4033|4034|4035|4036|4037|4038|4039|4040|4041|4042|4043|4044|4045|4046|4047|4048|4049|4050|4051|4052|4053|4054|4055|4056|4057|4058|4059|4060|4061|4062|4063|4064|4065|4066|4067|4068|4069|4070|4071|4072|4073|4074|4075|4076|4077|4078|4079|4080|4081|4082|4083|4084|4085|4086|4087|4088|4089|4090|4091|4092|4093|4094|4095|4096 +4.000000000000,2048,128,32,4097|4098|4099|4100|4101|4102|4103|4104|4105|4106|4107|4108|4109|4110|4111|4112|4113|4114|4115|4116|4117|4118|4119|4120|4121|4122|4123|4124|4125|4126|4127|4128|4129|4130|4131|4132|4133|4134|4135|4136|4137|4138|4139|4140|4141|4142|4143|4144|4145|4146|4147|4148|4149|4150|4151|4152|4153|4154|4155|4156|4157|4158|4159|4160|4161|4162|4163|4164|4165|4166|4167|4168|4169|4170|4171|4172|4173|4174|4175|4176|4177|4178|4179|4180|4181|4182|4183|4184|4185|4186|4187|4188|4189|4190|4191|4192|4193|4194|4195|4196|4197|4198|4199|4200|4201|4202|4203|4204|4205|4206|4207|4208|4209|4210|4211|4212|4213|4214|4215|4216|4217|4218|4219|4220|4221|4222|4223|4224 +4.125000000000,2048,128,33,4225|4226|4227|4228|4229|4230|4231|4232|4233|4234|4235|4236|4237|4238|4239|4240|4241|4242|4243|4244|4245|4246|4247|4248|4249|4250|4251|4252|4253|4254|4255|4256|4257|4258|4259|4260|4261|4262|4263|4264|4265|4266|4267|4268|4269|4270|4271|4272|4273|4274|4275|4276|4277|4278|4279|4280|4281|4282|4283|4284|4285|4286|4287|4288|4289|4290|4291|4292|4293|4294|4295|4296|4297|4298|4299|4300|4301|4302|4303|4304|4305|4306|4307|4308|4309|4310|4311|4312|4313|4314|4315|4316|4317|4318|4319|4320|4321|4322|4323|4324|4325|4326|4327|4328|4329|4330|4331|4332|4333|4334|4335|4336|4337|4338|4339|4340|4341|4342|4343|4344|4345|4346|4347|4348|4349|4350|4351|4352 +4.250000000000,2048,128,34,4353|4354|4355|4356|4357|4358|4359|4360|4361|4362|4363|4364|4365|4366|4367|4368|4369|4370|4371|4372|4373|4374|4375|4376|4377|4378|4379|4380|4381|4382|4383|4384|4385|4386|4387|4388|4389|4390|4391|4392|4393|4394|4395|4396|4397|4398|4399|4400|4401|4402|4403|4404|4405|4406|4407|4408|4409|4410|4411|4412|4413|4414|4415|4416|4417|4418|4419|4420|4421|4422|4423|4424|4425|4426|4427|4428|4429|4430|4431|4432|4433|4434|4435|4436|4437|4438|4439|4440|4441|4442|4443|4444|4445|4446|4447|4448|4449|4450|4451|4452|4453|4454|4455|4456|4457|4458|4459|4460|4461|4462|4463|4464|4465|4466|4467|4468|4469|4470|4471|4472|4473|4474|4475|4476|4477|4478|4479|4480 +4.375000000000,2048,128,35,4481|4482|4483|4484|4485|4486|4487|4488|4489|4490|4491|4492|4493|4494|4495|4496|4497|4498|4499|4500|4501|4502|4503|4504|4505|4506|4507|4508|4509|4510|4511|4512|4513|4514|4515|4516|4517|4518|4519|4520|4521|4522|4523|4524|4525|4526|4527|4528|4529|4530|4531|4532|4533|4534|4535|4536|4537|4538|4539|4540|4541|4542|4543|4544|4545|4546|4547|4548|4549|4550|4551|4552|4553|4554|4555|4556|4557|4558|4559|4560|4561|4562|4563|4564|4565|4566|4567|4568|4569|4570|4571|4572|4573|4574|4575|4576|4577|4578|4579|4580|4581|4582|4583|4584|4585|4586|4587|4588|4589|4590|4591|4592|4593|4594|4595|4596|4597|4598|4599|4600|4601|4602|4603|4604|4605|4606|4607|4608 +4.500000000000,2048,128,36,4609|4610|4611|4612|4613|4614|4615|4616|4617|4618|4619|4620|4621|4622|4623|4624|4625|4626|4627|4628|4629|4630|4631|4632|4633|4634|4635|4636|4637|4638|4639|4640|4641|4642|4643|4644|4645|4646|4647|4648|4649|4650|4651|4652|4653|4654|4655|4656|4657|4658|4659|4660|4661|4662|4663|4664|4665|4666|4667|4668|4669|4670|4671|4672|4673|4674|4675|4676|4677|4678|4679|4680|4681|4682|4683|4684|4685|4686|4687|4688|4689|4690|4691|4692|4693|4694|4695|4696|4697|4698|4699|4700|4701|4702|4703|4704|4705|4706|4707|4708|4709|4710|4711|4712|4713|4714|4715|4716|4717|4718|4719|4720|4721|4722|4723|4724|4725|4726|4727|4728|4729|4730|4731|4732|4733|4734|4735|4736 +4.625000000000,2048,128,37,4737|4738|4739|4740|4741|4742|4743|4744|4745|4746|4747|4748|4749|4750|4751|4752|4753|4754|4755|4756|4757|4758|4759|4760|4761|4762|4763|4764|4765|4766|4767|4768|4769|4770|4771|4772|4773|4774|4775|4776|4777|4778|4779|4780|4781|4782|4783|4784|4785|4786|4787|4788|4789|4790|4791|4792|4793|4794|4795|4796|4797|4798|4799|4800|4801|4802|4803|4804|4805|4806|4807|4808|4809|4810|4811|4812|4813|4814|4815|4816|4817|4818|4819|4820|4821|4822|4823|4824|4825|4826|4827|4828|4829|4830|4831|4832|4833|4834|4835|4836|4837|4838|4839|4840|4841|4842|4843|4844|4845|4846|4847|4848|4849|4850|4851|4852|4853|4854|4855|4856|4857|4858|4859|4860|4861|4862|4863|4864 +4.750000000000,2048,128,38,4865|4866|4867|4868|4869|4870|4871|4872|4873|4874|4875|4876|4877|4878|4879|4880|4881|4882|4883|4884|4885|4886|4887|4888|4889|4890|4891|4892|4893|4894|4895|4896|4897|4898|4899|4900|4901|4902|4903|4904|4905|4906|4907|4908|4909|4910|4911|4912|4913|4914|4915|4916|4917|4918|4919|4920|4921|4922|4923|4924|4925|4926|4927|4928|4929|4930|4931|4932|4933|4934|4935|4936|4937|4938|4939|4940|4941|4942|4943|4944|4945|4946|4947|4948|4949|4950|4951|4952|4953|4954|4955|4956|4957|4958|4959|4960|4961|4962|4963|4964|4965|4966|4967|4968|4969|4970|4971|4972|4973|4974|4975|4976|4977|4978|4979|4980|4981|4982|4983|4984|4985|4986|4987|4988|4989|4990|4991|4992 +4.875000000000,2048,128,39,4993|4994|4995|4996|4997|4998|4999|5000|5001|5002|5003|5004|5005|5006|5007|5008|5009|5010|5011|5012|5013|5014|5015|5016|5017|5018|5019|5020|5021|5022|5023|5024|5025|5026|5027|5028|5029|5030|5031|5032|5033|5034|5035|5036|5037|5038|5039|5040|5041|5042|5043|5044|5045|5046|5047|5048|5049|5050|5051|5052|5053|5054|5055|5056|5057|5058|5059|5060|5061|5062|5063|5064|5065|5066|5067|5068|5069|5070|5071|5072|5073|5074|5075|5076|5077|5078|5079|5080|5081|5082|5083|5084|5085|5086|5087|5088|5089|5090|5091|5092|5093|5094|5095|5096|5097|5098|5099|5100|5101|5102|5103|5104|5105|5106|5107|5108|5109|5110|5111|5112|5113|5114|5115|5116|5117|5118|5119|5120 +5.000000000000,2048,128,40,5121|5122|5123|5124|5125|5126|5127|5128|5129|5130|5131|5132|5133|5134|5135|5136|5137|5138|5139|5140|5141|5142|5143|5144|5145|5146|5147|5148|5149|5150|5151|5152|5153|5154|5155|5156|5157|5158|5159|5160|5161|5162|5163|5164|5165|5166|5167|5168|5169|5170|5171|5172|5173|5174|5175|5176|5177|5178|5179|5180|5181|5182|5183|5184|5185|5186|5187|5188|5189|5190|5191|5192|5193|5194|5195|5196|5197|5198|5199|5200|5201|5202|5203|5204|5205|5206|5207|5208|5209|5210|5211|5212|5213|5214|5215|5216|5217|5218|5219|5220|5221|5222|5223|5224|5225|5226|5227|5228|5229|5230|5231|5232|5233|5234|5235|5236|5237|5238|5239|5240|5241|5242|5243|5244|5245|5246|5247|5248 +5.125000000000,2048,128,41,5249|5250|5251|5252|5253|5254|5255|5256|5257|5258|5259|5260|5261|5262|5263|5264|5265|5266|5267|5268|5269|5270|5271|5272|5273|5274|5275|5276|5277|5278|5279|5280|5281|5282|5283|5284|5285|5286|5287|5288|5289|5290|5291|5292|5293|5294|5295|5296|5297|5298|5299|5300|5301|5302|5303|5304|5305|5306|5307|5308|5309|5310|5311|5312|5313|5314|5315|5316|5317|5318|5319|5320|5321|5322|5323|5324|5325|5326|5327|5328|5329|5330|5331|5332|5333|5334|5335|5336|5337|5338|5339|5340|5341|5342|5343|5344|5345|5346|5347|5348|5349|5350|5351|5352|5353|5354|5355|5356|5357|5358|5359|5360|5361|5362|5363|5364|5365|5366|5367|5368|5369|5370|5371|5372|5373|5374|5375|5376 +5.250000000000,2048,128,42,5377|5378|5379|5380|5381|5382|5383|5384|5385|5386|5387|5388|5389|5390|5391|5392|5393|5394|5395|5396|5397|5398|5399|5400|5401|5402|5403|5404|5405|5406|5407|5408|5409|5410|5411|5412|5413|5414|5415|5416|5417|5418|5419|5420|5421|5422|5423|5424|5425|5426|5427|5428|5429|5430|5431|5432|5433|5434|5435|5436|5437|5438|5439|5440|5441|5442|5443|5444|5445|5446|5447|5448|5449|5450|5451|5452|5453|5454|5455|5456|5457|5458|5459|5460|5461|5462|5463|5464|5465|5466|5467|5468|5469|5470|5471|5472|5473|5474|5475|5476|5477|5478|5479|5480|5481|5482|5483|5484|5485|5486|5487|5488|5489|5490|5491|5492|5493|5494|5495|5496|5497|5498|5499|5500|5501|5502|5503|5504 +5.375000000000,2048,128,43,5505|5506|5507|5508|5509|5510|5511|5512|5513|5514|5515|5516|5517|5518|5519|5520|5521|5522|5523|5524|5525|5526|5527|5528|5529|5530|5531|5532|5533|5534|5535|5536|5537|5538|5539|5540|5541|5542|5543|5544|5545|5546|5547|5548|5549|5550|5551|5552|5553|5554|5555|5556|5557|5558|5559|5560|5561|5562|5563|5564|5565|5566|5567|5568|5569|5570|5571|5572|5573|5574|5575|5576|5577|5578|5579|5580|5581|5582|5583|5584|5585|5586|5587|5588|5589|5590|5591|5592|5593|5594|5595|5596|5597|5598|5599|5600|5601|5602|5603|5604|5605|5606|5607|5608|5609|5610|5611|5612|5613|5614|5615|5616|5617|5618|5619|5620|5621|5622|5623|5624|5625|5626|5627|5628|5629|5630|5631|5632 +5.500000000000,2048,128,44,5633|5634|5635|5636|5637|5638|5639|5640|5641|5642|5643|5644|5645|5646|5647|5648|5649|5650|5651|5652|5653|5654|5655|5656|5657|5658|5659|5660|5661|5662|5663|5664|5665|5666|5667|5668|5669|5670|5671|5672|5673|5674|5675|5676|5677|5678|5679|5680|5681|5682|5683|5684|5685|5686|5687|5688|5689|5690|5691|5692|5693|5694|5695|5696|5697|5698|5699|5700|5701|5702|5703|5704|5705|5706|5707|5708|5709|5710|5711|5712|5713|5714|5715|5716|5717|5718|5719|5720|5721|5722|5723|5724|5725|5726|5727|5728|5729|5730|5731|5732|5733|5734|5735|5736|5737|5738|5739|5740|5741|5742|5743|5744|5745|5746|5747|5748|5749|5750|5751|5752|5753|5754|5755|5756|5757|5758|5759|5760 +5.625000000000,2048,128,45,5761|5762|5763|5764|5765|5766|5767|5768|5769|5770|5771|5772|5773|5774|5775|5776|5777|5778|5779|5780|5781|5782|5783|5784|5785|5786|5787|5788|5789|5790|5791|5792|5793|5794|5795|5796|5797|5798|5799|5800|5801|5802|5803|5804|5805|5806|5807|5808|5809|5810|5811|5812|5813|5814|5815|5816|5817|5818|5819|5820|5821|5822|5823|5824|5825|5826|5827|5828|5829|5830|5831|5832|5833|5834|5835|5836|5837|5838|5839|5840|5841|5842|5843|5844|5845|5846|5847|5848|5849|5850|5851|5852|5853|5854|5855|5856|5857|5858|5859|5860|5861|5862|5863|5864|5865|5866|5867|5868|5869|5870|5871|5872|5873|5874|5875|5876|5877|5878|5879|5880|5881|5882|5883|5884|5885|5886|5887|5888 +5.750000000000,2048,128,46,5889|5890|5891|5892|5893|5894|5895|5896|5897|5898|5899|5900|5901|5902|5903|5904|5905|5906|5907|5908|5909|5910|5911|5912|5913|5914|5915|5916|5917|5918|5919|5920|5921|5922|5923|5924|5925|5926|5927|5928|5929|5930|5931|5932|5933|5934|5935|5936|5937|5938|5939|5940|5941|5942|5943|5944|5945|5946|5947|5948|5949|5950|5951|5952|5953|5954|5955|5956|5957|5958|5959|5960|5961|5962|5963|5964|5965|5966|5967|5968|5969|5970|5971|5972|5973|5974|5975|5976|5977|5978|5979|5980|5981|5982|5983|5984|5985|5986|5987|5988|5989|5990|5991|5992|5993|5994|5995|5996|5997|5998|5999|6000|6001|6002|6003|6004|6005|6006|6007|6008|6009|6010|6011|6012|6013|6014|6015|6016 +5.875000000000,2048,128,47,6017|6018|6019|6020|6021|6022|6023|6024|6025|6026|6027|6028|6029|6030|6031|6032|6033|6034|6035|6036|6037|6038|6039|6040|6041|6042|6043|6044|6045|6046|6047|6048|6049|6050|6051|6052|6053|6054|6055|6056|6057|6058|6059|6060|6061|6062|6063|6064|6065|6066|6067|6068|6069|6070|6071|6072|6073|6074|6075|6076|6077|6078|6079|6080|6081|6082|6083|6084|6085|6086|6087|6088|6089|6090|6091|6092|6093|6094|6095|6096|6097|6098|6099|6100|6101|6102|6103|6104|6105|6106|6107|6108|6109|6110|6111|6112|6113|6114|6115|6116|6117|6118|6119|6120|6121|6122|6123|6124|6125|6126|6127|6128|6129|6130|6131|6132|6133|6134|6135|6136|6137|6138|6139|6140|6141|6142|6143|6144 +6.000000000000,2048,128,48,6145|6146|6147|6148|6149|6150|6151|6152|6153|6154|6155|6156|6157|6158|6159|6160|6161|6162|6163|6164|6165|6166|6167|6168|6169|6170|6171|6172|6173|6174|6175|6176|6177|6178|6179|6180|6181|6182|6183|6184|6185|6186|6187|6188|6189|6190|6191|6192|6193|6194|6195|6196|6197|6198|6199|6200|6201|6202|6203|6204|6205|6206|6207|6208|6209|6210|6211|6212|6213|6214|6215|6216|6217|6218|6219|6220|6221|6222|6223|6224|6225|6226|6227|6228|6229|6230|6231|6232|6233|6234|6235|6236|6237|6238|6239|6240|6241|6242|6243|6244|6245|6246|6247|6248|6249|6250|6251|6252|6253|6254|6255|6256|6257|6258|6259|6260|6261|6262|6263|6264|6265|6266|6267|6268|6269|6270|6271|6272 +6.125000000000,2048,128,49,6273|6274|6275|6276|6277|6278|6279|6280|6281|6282|6283|6284|6285|6286|6287|6288|6289|6290|6291|6292|6293|6294|6295|6296|6297|6298|6299|6300|6301|6302|6303|6304|6305|6306|6307|6308|6309|6310|6311|6312|6313|6314|6315|6316|6317|6318|6319|6320|6321|6322|6323|6324|6325|6326|6327|6328|6329|6330|6331|6332|6333|6334|6335|6336|6337|6338|6339|6340|6341|6342|6343|6344|6345|6346|6347|6348|6349|6350|6351|6352|6353|6354|6355|6356|6357|6358|6359|6360|6361|6362|6363|6364|6365|6366|6367|6368|6369|6370|6371|6372|6373|6374|6375|6376|6377|6378|6379|6380|6381|6382|6383|6384|6385|6386|6387|6388|6389|6390|6391|6392|6393|6394|6395|6396|6397|6398|6399|6400 +6.250000000000,2048,128,50,6401|6402|6403|6404|6405|6406|6407|6408|6409|6410|6411|6412|6413|6414|6415|6416|6417|6418|6419|6420|6421|6422|6423|6424|6425|6426|6427|6428|6429|6430|6431|6432|6433|6434|6435|6436|6437|6438|6439|6440|6441|6442|6443|6444|6445|6446|6447|6448|6449|6450|6451|6452|6453|6454|6455|6456|6457|6458|6459|6460|6461|6462|6463|6464|6465|6466|6467|6468|6469|6470|6471|6472|6473|6474|6475|6476|6477|6478|6479|6480|6481|6482|6483|6484|6485|6486|6487|6488|6489|6490|6491|6492|6493|6494|6495|6496|6497|6498|6499|6500|6501|6502|6503|6504|6505|6506|6507|6508|6509|6510|6511|6512|6513|6514|6515|6516|6517|6518|6519|6520|6521|6522|6523|6524|6525|6526|6527|6528 +6.375000000000,2048,128,51,6529|6530|6531|6532|6533|6534|6535|6536|6537|6538|6539|6540|6541|6542|6543|6544|6545|6546|6547|6548|6549|6550|6551|6552|6553|6554|6555|6556|6557|6558|6559|6560|6561|6562|6563|6564|6565|6566|6567|6568|6569|6570|6571|6572|6573|6574|6575|6576|6577|6578|6579|6580|6581|6582|6583|6584|6585|6586|6587|6588|6589|6590|6591|6592|6593|6594|6595|6596|6597|6598|6599|6600|6601|6602|6603|6604|6605|6606|6607|6608|6609|6610|6611|6612|6613|6614|6615|6616|6617|6618|6619|6620|6621|6622|6623|6624|6625|6626|6627|6628|6629|6630|6631|6632|6633|6634|6635|6636|6637|6638|6639|6640|6641|6642|6643|6644|6645|6646|6647|6648|6649|6650|6651|6652|6653|6654|6655|6656 +6.500000000000,2048,128,52,6657|6658|6659|6660|6661|6662|6663|6664|6665|6666|6667|6668|6669|6670|6671|6672|6673|6674|6675|6676|6677|6678|6679|6680|6681|6682|6683|6684|6685|6686|6687|6688|6689|6690|6691|6692|6693|6694|6695|6696|6697|6698|6699|6700|6701|6702|6703|6704|6705|6706|6707|6708|6709|6710|6711|6712|6713|6714|6715|6716|6717|6718|6719|6720|6721|6722|6723|6724|6725|6726|6727|6728|6729|6730|6731|6732|6733|6734|6735|6736|6737|6738|6739|6740|6741|6742|6743|6744|6745|6746|6747|6748|6749|6750|6751|6752|6753|6754|6755|6756|6757|6758|6759|6760|6761|6762|6763|6764|6765|6766|6767|6768|6769|6770|6771|6772|6773|6774|6775|6776|6777|6778|6779|6780|6781|6782|6783|6784 +6.625000000000,2048,128,53,6785|6786|6787|6788|6789|6790|6791|6792|6793|6794|6795|6796|6797|6798|6799|6800|6801|6802|6803|6804|6805|6806|6807|6808|6809|6810|6811|6812|6813|6814|6815|6816|6817|6818|6819|6820|6821|6822|6823|6824|6825|6826|6827|6828|6829|6830|6831|6832|6833|6834|6835|6836|6837|6838|6839|6840|6841|6842|6843|6844|6845|6846|6847|6848|6849|6850|6851|6852|6853|6854|6855|6856|6857|6858|6859|6860|6861|6862|6863|6864|6865|6866|6867|6868|6869|6870|6871|6872|6873|6874|6875|6876|6877|6878|6879|6880|6881|6882|6883|6884|6885|6886|6887|6888|6889|6890|6891|6892|6893|6894|6895|6896|6897|6898|6899|6900|6901|6902|6903|6904|6905|6906|6907|6908|6909|6910|6911|6912 +6.750000000000,2048,128,54,6913|6914|6915|6916|6917|6918|6919|6920|6921|6922|6923|6924|6925|6926|6927|6928|6929|6930|6931|6932|6933|6934|6935|6936|6937|6938|6939|6940|6941|6942|6943|6944|6945|6946|6947|6948|6949|6950|6951|6952|6953|6954|6955|6956|6957|6958|6959|6960|6961|6962|6963|6964|6965|6966|6967|6968|6969|6970|6971|6972|6973|6974|6975|6976|6977|6978|6979|6980|6981|6982|6983|6984|6985|6986|6987|6988|6989|6990|6991|6992|6993|6994|6995|6996|6997|6998|6999|7000|7001|7002|7003|7004|7005|7006|7007|7008|7009|7010|7011|7012|7013|7014|7015|7016|7017|7018|7019|7020|7021|7022|7023|7024|7025|7026|7027|7028|7029|7030|7031|7032|7033|7034|7035|7036|7037|7038|7039|7040 +6.875000000000,2048,128,55,7041|7042|7043|7044|7045|7046|7047|7048|7049|7050|7051|7052|7053|7054|7055|7056|7057|7058|7059|7060|7061|7062|7063|7064|7065|7066|7067|7068|7069|7070|7071|7072|7073|7074|7075|7076|7077|7078|7079|7080|7081|7082|7083|7084|7085|7086|7087|7088|7089|7090|7091|7092|7093|7094|7095|7096|7097|7098|7099|7100|7101|7102|7103|7104|7105|7106|7107|7108|7109|7110|7111|7112|7113|7114|7115|7116|7117|7118|7119|7120|7121|7122|7123|7124|7125|7126|7127|7128|7129|7130|7131|7132|7133|7134|7135|7136|7137|7138|7139|7140|7141|7142|7143|7144|7145|7146|7147|7148|7149|7150|7151|7152|7153|7154|7155|7156|7157|7158|7159|7160|7161|7162|7163|7164|7165|7166|7167|7168 +7.000000000000,2048,128,56,7169|7170|7171|7172|7173|7174|7175|7176|7177|7178|7179|7180|7181|7182|7183|7184|7185|7186|7187|7188|7189|7190|7191|7192|7193|7194|7195|7196|7197|7198|7199|7200|7201|7202|7203|7204|7205|7206|7207|7208|7209|7210|7211|7212|7213|7214|7215|7216|7217|7218|7219|7220|7221|7222|7223|7224|7225|7226|7227|7228|7229|7230|7231|7232|7233|7234|7235|7236|7237|7238|7239|7240|7241|7242|7243|7244|7245|7246|7247|7248|7249|7250|7251|7252|7253|7254|7255|7256|7257|7258|7259|7260|7261|7262|7263|7264|7265|7266|7267|7268|7269|7270|7271|7272|7273|7274|7275|7276|7277|7278|7279|7280|7281|7282|7283|7284|7285|7286|7287|7288|7289|7290|7291|7292|7293|7294|7295|7296 +7.125000000000,2048,128,57,7297|7298|7299|7300|7301|7302|7303|7304|7305|7306|7307|7308|7309|7310|7311|7312|7313|7314|7315|7316|7317|7318|7319|7320|7321|7322|7323|7324|7325|7326|7327|7328|7329|7330|7331|7332|7333|7334|7335|7336|7337|7338|7339|7340|7341|7342|7343|7344|7345|7346|7347|7348|7349|7350|7351|7352|7353|7354|7355|7356|7357|7358|7359|7360|7361|7362|7363|7364|7365|7366|7367|7368|7369|7370|7371|7372|7373|7374|7375|7376|7377|7378|7379|7380|7381|7382|7383|7384|7385|7386|7387|7388|7389|7390|7391|7392|7393|7394|7395|7396|7397|7398|7399|7400|7401|7402|7403|7404|7405|7406|7407|7408|7409|7410|7411|7412|7413|7414|7415|7416|7417|7418|7419|7420|7421|7422|7423|7424 +7.250000000000,2048,128,58,7425|7426|7427|7428|7429|7430|7431|7432|7433|7434|7435|7436|7437|7438|7439|7440|7441|7442|7443|7444|7445|7446|7447|7448|7449|7450|7451|7452|7453|7454|7455|7456|7457|7458|7459|7460|7461|7462|7463|7464|7465|7466|7467|7468|7469|7470|7471|7472|7473|7474|7475|7476|7477|7478|7479|7480|7481|7482|7483|7484|7485|7486|7487|7488|7489|7490|7491|7492|7493|7494|7495|7496|7497|7498|7499|7500|7501|7502|7503|7504|7505|7506|7507|7508|7509|7510|7511|7512|7513|7514|7515|7516|7517|7518|7519|7520|7521|7522|7523|7524|7525|7526|7527|7528|7529|7530|7531|7532|7533|7534|7535|7536|7537|7538|7539|7540|7541|7542|7543|7544|7545|7546|7547|7548|7549|7550|7551|7552 +7.375000000000,2048,128,59,7553|7554|7555|7556|7557|7558|7559|7560|7561|7562|7563|7564|7565|7566|7567|7568|7569|7570|7571|7572|7573|7574|7575|7576|7577|7578|7579|7580|7581|7582|7583|7584|7585|7586|7587|7588|7589|7590|7591|7592|7593|7594|7595|7596|7597|7598|7599|7600|7601|7602|7603|7604|7605|7606|7607|7608|7609|7610|7611|7612|7613|7614|7615|7616|7617|7618|7619|7620|7621|7622|7623|7624|7625|7626|7627|7628|7629|7630|7631|7632|7633|7634|7635|7636|7637|7638|7639|7640|7641|7642|7643|7644|7645|7646|7647|7648|7649|7650|7651|7652|7653|7654|7655|7656|7657|7658|7659|7660|7661|7662|7663|7664|7665|7666|7667|7668|7669|7670|7671|7672|7673|7674|7675|7676|7677|7678|7679|7680 +7.500000000000,2048,128,60,7681|7682|7683|7684|7685|7686|7687|7688|7689|7690|7691|7692|7693|7694|7695|7696|7697|7698|7699|7700|7701|7702|7703|7704|7705|7706|7707|7708|7709|7710|7711|7712|7713|7714|7715|7716|7717|7718|7719|7720|7721|7722|7723|7724|7725|7726|7727|7728|7729|7730|7731|7732|7733|7734|7735|7736|7737|7738|7739|7740|7741|7742|7743|7744|7745|7746|7747|7748|7749|7750|7751|7752|7753|7754|7755|7756|7757|7758|7759|7760|7761|7762|7763|7764|7765|7766|7767|7768|7769|7770|7771|7772|7773|7774|7775|7776|7777|7778|7779|7780|7781|7782|7783|7784|7785|7786|7787|7788|7789|7790|7791|7792|7793|7794|7795|7796|7797|7798|7799|7800|7801|7802|7803|7804|7805|7806|7807|7808 +7.625000000000,2048,128,61,7809|7810|7811|7812|7813|7814|7815|7816|7817|7818|7819|7820|7821|7822|7823|7824|7825|7826|7827|7828|7829|7830|7831|7832|7833|7834|7835|7836|7837|7838|7839|7840|7841|7842|7843|7844|7845|7846|7847|7848|7849|7850|7851|7852|7853|7854|7855|7856|7857|7858|7859|7860|7861|7862|7863|7864|7865|7866|7867|7868|7869|7870|7871|7872|7873|7874|7875|7876|7877|7878|7879|7880|7881|7882|7883|7884|7885|7886|7887|7888|7889|7890|7891|7892|7893|7894|7895|7896|7897|7898|7899|7900|7901|7902|7903|7904|7905|7906|7907|7908|7909|7910|7911|7912|7913|7914|7915|7916|7917|7918|7919|7920|7921|7922|7923|7924|7925|7926|7927|7928|7929|7930|7931|7932|7933|7934|7935|7936 +7.750000000000,2048,128,62,7937|7938|7939|7940|7941|7942|7943|7944|7945|7946|7947|7948|7949|7950|7951|7952|7953|7954|7955|7956|7957|7958|7959|7960|7961|7962|7963|7964|7965|7966|7967|7968|7969|7970|7971|7972|7973|7974|7975|7976|7977|7978|7979|7980|7981|7982|7983|7984|7985|7986|7987|7988|7989|7990|7991|7992|7993|7994|7995|7996|7997|7998|7999|8000|8001|8002|8003|8004|8005|8006|8007|8008|8009|8010|8011|8012|8013|8014|8015|8016|8017|8018|8019|8020|8021|8022|8023|8024|8025|8026|8027|8028|8029|8030|8031|8032|8033|8034|8035|8036|8037|8038|8039|8040|8041|8042|8043|8044|8045|8046|8047|8048|8049|8050|8051|8052|8053|8054|8055|8056|8057|8058|8059|8060|8061|8062|8063|8064 +7.875000000000,2048,128,63,8065|8066|8067|8068|8069|8070|8071|8072|8073|8074|8075|8076|8077|8078|8079|8080|8081|8082|8083|8084|8085|8086|8087|8088|8089|8090|8091|8092|8093|8094|8095|8096|8097|8098|8099|8100|8101|8102|8103|8104|8105|8106|8107|8108|8109|8110|8111|8112|8113|8114|8115|8116|8117|8118|8119|8120|8121|8122|8123|8124|8125|8126|8127|8128|8129|8130|8131|8132|8133|8134|8135|8136|8137|8138|8139|8140|8141|8142|8143|8144|8145|8146|8147|8148|8149|8150|8151|8152|8153|8154|8155|8156|8157|8158|8159|8160|8161|8162|8163|8164|8165|8166|8167|8168|8169|8170|8171|8172|8173|8174|8175|8176|8177|8178|8179|8180|8181|8182|8183|8184|8185|8186|8187|8188|8189|8190|8191|8192 diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/assemble_profiles.py b/runs/frontier-multicase-sufficiency-v0/best_effort/assemble_profiles.py new file mode 100644 index 0000000..a060526 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/assemble_profiles.py @@ -0,0 +1,378 @@ +#!/usr/bin/env python3 +"""Assemble and validate the Qwen3-235B Frontier best-effort profiles.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +from datetime import datetime, timezone +from pathlib import Path + +import pandas as pd + + +MODEL = "Qwen3-235B-A22B-FP8" +HARDWARE = "h20" +TOKENS = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384] +QUANT_SIGNATURE = "method=fp8|act=dynamic|serialized=True|block=128x128" +MEASUREMENT_TYPE = "CUDA_EVENT" +PROFILING_PRECISION = "BF16" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--linear-tp4", type=Path, required=True) + parser.add_argument("--linear-tp8", type=Path, required=True) + parser.add_argument("--attention-standard", type=Path, required=True) + parser.add_argument("--attention-mixed", type=Path, required=True) + parser.add_argument("--moe-tp4-ep1", type=Path, required=True) + parser.add_argument("--moe-tp1-ep8", type=Path, required=True) + parser.add_argument("--moe-standalone-tp4-ep1", type=Path, required=True) + parser.add_argument("--moe-standalone-tp1-ep8", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument( + "--frontier-commit", + default="d9cfeb6d8791fbf2f295dd9744c56a666171776e", + ) + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def read_csv(path: Path, label: str) -> pd.DataFrame: + if not path.is_file(): + raise FileNotFoundError(f"{label}: missing input CSV: {path}") + data = pd.read_csv(path) + if data.empty: + raise ValueError(f"{label}: input CSV is empty: {path}") + return data + + +def require_columns(data: pd.DataFrame, columns: list[str], label: str) -> None: + missing = [column for column in columns if column not in data.columns] + if missing: + raise ValueError(f"{label}: missing columns: {missing}") + + +def require_exact_values( + data: pd.DataFrame, column: str, expected: set[object], label: str +) -> None: + actual = set(data[column].dropna().unique().tolist()) + if actual != expected: + raise ValueError( + f"{label}: {column} values mismatch: expected={sorted(expected)!r}, " + f"actual={sorted(actual)!r}" + ) + + +def validate_common(data: pd.DataFrame, label: str) -> None: + columns = ["quant_signature", "profiling_precision", "measurement_type"] + require_columns(data, columns, label) + require_exact_values(data, "quant_signature", {QUANT_SIGNATURE}, label) + require_exact_values(data, "profiling_precision", {PROFILING_PRECISION}, label) + require_exact_values(data, "measurement_type", {MEASUREMENT_TYPE}, label) + + +def validate_time_columns( + data: pd.DataFrame, columns: list[str], label: str +) -> None: + require_columns(data, columns, label) + if data[columns].isna().any(axis=None): + counts = data[columns].isna().sum() + raise ValueError(f"{label}: NaN timing values: {counts[counts > 0].to_dict()}") + if (data[columns] < 0).any(axis=None): + raise ValueError(f"{label}: negative timing value") + + +def validate_token_grid( + data: pd.DataFrame, + group_columns: list[str], + expected_rows_per_point: int, + label: str, +) -> None: + for group, rows in data.groupby(group_columns, dropna=False): + counts = rows["num_tokens"].value_counts().to_dict() + expected = {token: expected_rows_per_point for token in TOKENS} + if counts != expected: + raise ValueError( + f"{label}: token grid mismatch for group={group}: " + f"expected={expected}, actual={counts}" + ) + + +def assemble_linear(tp4: pd.DataFrame, tp8: pd.DataFrame) -> pd.DataFrame: + label = "linear" + for source_label, data, expected_tp in ( + ("linear-tp4", tp4, {1, 4}), + ("linear-tp8", tp8, {1, 8}), + ): + validate_common(data, source_label) + require_columns(data, ["num_tensor_parallel_workers", "num_tokens"], source_label) + require_exact_values(data, "num_tensor_parallel_workers", expected_tp, source_label) + validate_token_grid(data, ["num_tensor_parallel_workers"], 1, source_label) + + # The TP=1 rows are replicated operators. Keep the rows from the TP4 run and + # add only the TP=8 sharded operators from the second run. + combined = pd.concat( + [tp4, tp8[tp8["num_tensor_parallel_workers"] == 8]], ignore_index=True + ) + require_exact_values(combined, "num_tensor_parallel_workers", {1, 4, 8}, label) + validate_token_grid(combined, ["num_tensor_parallel_workers"], 1, label) + + replicated = combined[combined["num_tensor_parallel_workers"] == 1] + sharded = combined[combined["num_tensor_parallel_workers"] > 1] + validate_time_columns( + replicated, + [ + "time_stats.emb.median", + "time_stats.input_layernorm.median", + "time_stats.post_attention_layernorm.median", + ], + "linear replicated operators", + ) + validate_time_columns( + sharded, + [ + "time_stats.attn_pre_proj.median", + "time_stats.attn_rope.median", + "time_stats.attn_post_proj.median", + ], + "linear sharded attention operators", + ) + return combined.sort_values( + ["num_tensor_parallel_workers", "num_tokens"], kind="stable" + ).reset_index(drop=True) + + +def assemble_attention(standard: pd.DataFrame, mixed: pd.DataFrame) -> pd.DataFrame: + timing_columns = [ + "time_stats.attn_input_reshape.median", + "time_stats.attn_kv_cache_save.median", + "time_stats.attn_prefill.median", + "time_stats.attn_decode.median", + "time_stats.attn_output_reshape.median", + ] + for label, data, expected_mixed, expected_rows in ( + ("attention-standard", standard, {False}, 390), + ("attention-mixed", mixed, {True}, 336), + ): + validate_common(data, label) + require_columns( + data, + ["num_tensor_parallel_workers", "is_prefill", "is_mixed_batch"], + label, + ) + require_exact_values(data, "num_tensor_parallel_workers", {4, 8}, label) + require_exact_values(data, "is_prefill", {True}, label) + require_exact_values(data, "is_mixed_batch", expected_mixed, label) + validate_time_columns(data, timing_columns, label) + if len(data) != expected_rows: + raise ValueError( + f"{label}: row count mismatch: expected={expected_rows}, actual={len(data)}" + ) + + combined = pd.concat([standard, mixed], ignore_index=True) + if len(combined) != 726: + raise ValueError(f"attention: expected 726 rows, got {len(combined)}") + sort_columns = [ + "num_tensor_parallel_workers", + "is_mixed_batch", + "total_tokens", + "total_prefill_tokens", + "kv_cache_size", + "batch_size", + ] + return combined.sort_values(sort_columns, kind="stable").reset_index(drop=True) + + +def assemble_moe( + tp4_ep1: pd.DataFrame, + tp1_ep8: pd.DataFrame, + standalone_tp4_ep1: pd.DataFrame, + standalone_tp1_ep8: pd.DataFrame, +) -> pd.DataFrame: + timing_columns = [ + "time_stats.moe_gating_linear.median", + "time_stats.moe_gating_routing_topk.median", + "time_stats.moe_shuffling.median", + "time_stats.moe_grouped_gemm.median", + ] + cases = ( + ("moe-tp4-ep1", tp4_ep1, 4, 1, 128, "prefill_hot"), + ("moe-tp1-ep8", tp1_ep8, 1, 8, 16, "prefill_hot"), + ( + "moe-standalone-tp4-ep1", + standalone_tp4_ep1, + 4, + 1, + 128, + "standalone_legacy", + ), + ( + "moe-standalone-tp1-ep8", + standalone_tp1_ep8, + 1, + 8, + 16, + "standalone_legacy", + ), + ) + for ( + label, + data, + expected_tp, + expected_ep, + expected_local_experts, + expected_context, + ) in cases: + validate_common(data, label) + require_columns( + data, + [ + "num_tensor_parallel_workers", + "expert_parallel_size", + "num_experts_per_device", + "num_tokens", + "load_distribution", + "seed", + "routing_runtime_path", + "gating_runtime_context", + ], + label, + ) + require_exact_values(data, "num_tensor_parallel_workers", {expected_tp}, label) + require_exact_values(data, "expert_parallel_size", {expected_ep}, label) + require_exact_values(data, "num_experts_per_device", {expected_local_experts}, label) + require_exact_values( + data, + "load_distribution", + {"uniform", "skewed", "extremely_skewed"}, + label, + ) + require_exact_values(data, "seed", {0, 1}, label) + require_exact_values(data, "routing_runtime_path", {"standard_fused_topk"}, label) + require_exact_values( + data, "gating_runtime_context", {expected_context}, label + ) + validate_token_grid( + data, + ["num_tensor_parallel_workers", "expert_parallel_size"], + 6, + label, + ) + validate_time_columns( + data, + timing_columns if expected_context == "prefill_hot" else timing_columns[:2], + label, + ) + if len(data) != 90: + raise ValueError(f"{label}: expected 90 rows, got {len(data)}") + + # Standalone rows are needed only for Frontier's pure-decode gating models. + # Do not duplicate the shuffling/grouped-GEMM observations merely because + # the profiler collected them while measuring a second gating context. + standalone = pd.concat( + [standalone_tp4_ep1, standalone_tp1_ep8], ignore_index=True + ) + standalone[timing_columns[2:]] = float("nan") + combined = pd.concat([tp4_ep1, tp1_ep8, standalone], ignore_index=True) + return combined.sort_values( + [ + "num_tensor_parallel_workers", + "expert_parallel_size", + "gating_runtime_context", + "num_tokens", + "load_distribution", + "seed", + ], + kind="stable", + ).reset_index(drop=True) + + +def main() -> None: + args = parse_args() + input_paths = { + "linear_tp4": args.linear_tp4, + "linear_tp8": args.linear_tp8, + "attention_standard": args.attention_standard, + "attention_mixed": args.attention_mixed, + "moe_tp4_ep1": args.moe_tp4_ep1, + "moe_tp1_ep8": args.moe_tp1_ep8, + "moe_standalone_tp4_ep1": args.moe_standalone_tp4_ep1, + "moe_standalone_tp1_ep8": args.moe_standalone_tp1_ep8, + } + inputs = {name: read_csv(path, name) for name, path in input_paths.items()} + + outputs = { + "linear_op.csv": assemble_linear(inputs["linear_tp4"], inputs["linear_tp8"]), + "attention.csv": assemble_attention( + inputs["attention_standard"], inputs["attention_mixed"] + ), + "moe.csv": assemble_moe( + inputs["moe_tp4_ep1"], + inputs["moe_tp1_ep8"], + inputs["moe_standalone_tp4_ep1"], + inputs["moe_standalone_tp1_ep8"], + ), + } + + output_dir = args.output_root / "compute" / HARDWARE / MODEL + output_dir.mkdir(parents=True, exist_ok=True) + output_paths: dict[str, Path] = {} + for name, data in outputs.items(): + path = output_dir / name + data.to_csv(path, index=False) + output_paths[name] = path + + manifest = { + "generated_at_utc": datetime.now(timezone.utc).isoformat(), + "frontier_commit": args.frontier_commit, + "model": MODEL, + "hardware": HARDWARE, + "contract": { + "quant_signature": QUANT_SIGNATURE, + "profiling_precision": PROFILING_PRECISION, + "profiling_precision_semantics": "BF16 kernel output/accumulation; weights and activations are block FP8 W8A8", + "measurement_type": MEASUREMENT_TYPE, + "linear_tensor_parallel_sizes": [1, 4, 8], + "attention_tensor_parallel_sizes": [4, 8], + "moe_layouts": [ + {"tensor_parallel_size": 4, "expert_parallel_size": 1}, + {"tensor_parallel_size": 1, "expert_parallel_size": 8}, + ], + "moe_gating_runtime_contexts": [ + "prefill_hot", + "standalone_legacy", + ], + }, + "inputs": { + name: { + "path": str(path.resolve()), + "sha256": sha256(path), + "rows": len(inputs[name]), + } + for name, path in input_paths.items() + }, + "outputs": { + name: { + "path": str(path.resolve()), + "sha256": sha256(path), + "rows": len(outputs[name]), + } + for name, path in output_paths.items() + }, + } + manifest_path = args.output_root / "profile_manifest.json" + manifest_path.write_text(json.dumps(manifest, indent=2) + "\n", encoding="utf-8") + print(json.dumps(manifest, indent=2)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/community_prefill_grid.py b/runs/frontier-multicase-sufficiency-v0/best_effort/community_prefill_grid.py new file mode 100644 index 0000000..74eb149 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/community_prefill_grid.py @@ -0,0 +1,455 @@ +#!/usr/bin/env python3 +"""Run the community-vLLM side of the frozen Qwen235B prefill grid.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import os +import random +import subprocess +import time +from dataclasses import asdict, dataclass +from pathlib import Path +from typing import Any + + +SCHEMA = "community-vllm-qwen235b-prefill-grid-v2" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8" +SERVED_MODEL = "qwen3-235b-community-prefill" +TRACE = ( + "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/" + "thinking_w20260327_1000.jsonl" +) +WINDOWS = "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/windows.json" +TRACE_SHA256 = "f878e9af18f94dcfaced94a8e1e6b20a2f7d97d64aa862448025660dbbd965b2" +ORDER_SEED = 20260715 +CLIENT_MAX_CONCURRENCY = 256 + + +@dataclass(frozen=True) +class GridConfig: + tp: int + mns: int + mbt: int + expert_parallel: bool + num_gpu_blocks: int + + @property + def name(self) -> str: + return f"tp{self.tp}_mns{self.mns}_mbt{self.mbt}" + + +GRID = tuple( + GridConfig( + tp=tp, + mns=mns, + mbt=mbt, + expert_parallel=tp == 8, + num_gpu_blocks=26101 if tp == 4 else 62351, + ) + for tp in (4, 8) + for mns in (64, 128) + for mbt in (8192, 16384) +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def search_spec() -> dict[str, Any]: + return { + "low": 0.0, + "high": 0.125, + "tolerance": 0.001, + "max_probes": 6, + "sample_seed": 20260325, + "inherit_incumbent_floor": False, + "auto_high": { + "enabled": False, + "max_sampling_u": 1.0, + "require_human_confirmation_beyond_trace": True, + }, + } + + +def study_payload( + *, + tp: int, + repo: Path, + python: Path, + vllm: Path, + model: Path, + trace: Path, + windows: Path, + port: int, +) -> dict[str, Any]: + base_flags: dict[str, Any] = { + "host": "127.0.0.1", + "port": port, + "served-model-name": SERVED_MODEL, + "tensor-parallel-size": tp, + "disable-custom-all-reduce": True, + "quantization": "fp8", + "gpu-memory-utilization": 0.80, + "num-gpu-blocks-override": 26101 if tp == 4 else 62351, + "kv-cache-dtype": "auto", + "max-model-len": 40960, + "max-num-batched-tokens": 8192, + "max-num-seqs": 64, + "enable-prefix-caching": False, + "enable-chunked-prefill": True, + "enforce-eager": True, + "disable-log-requests": True, + } + if tp == 8: + base_flags["enable-expert-parallel"] = True + return { + "study_id": f"community-qwen235b-prefill-tp{tp}-v2", + "hardware": { + "gpu_count": tp, + "gpu_model": "NVIDIA H20", + "host_candidates": ["dash0"], + }, + "model": {"model_id": str(model), "served_model_name": SERVED_MODEL}, + "engine": { + "engine_name": "vllm", + "engine_version": "0.10.2-community", + "exec_path": str(vllm), + "cwd": str(repo), + "host": "127.0.0.1", + "port": port, + "ready_timeout_s": 1800, + "request_timeout_s": 1800, + "healthcheck_path": "/v1/models", + "launch_args": ["serve", str(model)], + "base_envs": { + "CUDA_VISIBLE_DEVICES": ",".join(str(index) for index in range(tp)), + }, + "base_flags": base_flags, + "tunable_envs": [], + "tunable_flags": ["max-num-seqs", "max-num-batched-tokens"], + "python_executable": str(python), + }, + "trace": { + "windows_path": str(windows), + "window_id": "thinking_w20260327_1000", + "trace_file_override": str(trace), + "request_mode": "raw_completion", + "completion_tokens_override": 1, + "u_field": "sampling_u", + "timestamp_field": "timestamp", + # Keep the request generator above the largest server-side MNS so + # MNS=128 is exercised by vLLM rather than clipped by the client. + # The extra headroom also leaves an explicit engine waiting queue + # under overload instead of moving that queue into the replay loop. + "max_concurrency": CLIENT_MAX_CONCURRENCY, + "input_length_filter": { + "min_input_tokens": 0, + "max_input_tokens": 32768, + }, + "replay_time_scale": 1.0, + "early_stop_max_lag_s": 180.0, + "early_stop_max_elapsed_s": 1200.0, + "restart_engine_after_early_stop": False, + "adaptive_stop": {"enabled": False}, + }, + "slo": { + "target_pass_rate": 0.95, + "ttft_rule": { + "kind": "step_ms", + "buckets": [ + {"max_input_tokens": 8191, "threshold_ms": 1000}, + {"max_input_tokens": 32767, "threshold_ms": 2000}, + {"threshold_ms": 2000}, + ], + }, + }, + "search": search_spec(), + "llm": {"use_harness": False}, + } + + +def git_fingerprint(repo: Path) -> dict[str, Any]: + def run(*args: str) -> str: + return subprocess.run( + ["git", *args], + cwd=repo, + check=True, + stdout=subprocess.PIPE, + text=True, + ).stdout.strip() + + return { + "commit": run("rev-parse", "HEAD"), + "status_porcelain": run("status", "--porcelain=v1").splitlines(), + } + + +def prepare(args: argparse.Namespace) -> None: + repo = args.repo.resolve() + python = args.python.resolve() + vllm = args.vllm.resolve() + model = args.model.resolve() + trace = args.trace.resolve() + windows = args.windows.resolve() + for path in (repo, python, vllm, model, trace, windows): + if not path.exists(): + raise FileNotFoundError(path) + actual_trace_sha = sha256(trace) + if actual_trace_sha != args.expected_trace_sha256: + raise ValueError( + f"trace SHA256 mismatch: expected={args.expected_trace_sha256}, " + f"actual={actual_trace_sha}" + ) + output = args.output_root.resolve() + output.mkdir(parents=True, exist_ok=True) + studies: dict[int, dict[str, str]] = {} + for tp, port in ((4, args.port), (8, args.port)): + study_dir = output / "studies" / f"tp{tp}" + study_path = study_dir / "study.json" + pointer_path = study_dir / "study_spec.source" + write_json( + study_path, + study_payload( + tp=tp, + repo=repo, + python=python, + vllm=vllm, + model=model, + trace=trace, + windows=windows, + port=port, + ), + ) + pointer_path.write_text(str(study_path) + "\n") + studies[tp] = { + "study_path": str(study_path), + "study_sha256": sha256(study_path), + "pointer_path": str(pointer_path), + } + + order = list(GRID) + random.Random(args.order_seed).shuffle(order) + trials = [] + for index, config in enumerate(order, start=1): + trial_dir = output / "trials" / config.name + trial_path = trial_dir / "trial_spec.json" + trial = { + "study_id": f"community-qwen235b-prefill-tp{config.tp}-v2", + "trial_id": config.name, + "config_patch": { + "env_patch": {}, + "flag_patch": { + "max-num-seqs": config.mns, + "max-num-batched-tokens": config.mbt, + }, + }, + "search": search_spec(), + "study_spec_path": studies[config.tp]["pointer_path"], + "artifact_dir": str(trial_dir), + "probe_log_path": str(trial_dir / "probe_history.json"), + "engine_log_path": str(trial_dir / "engine.log"), + "result_path": str(trial_dir / "result.json"), + "search_evidence": { + "enabled": False, + "original_high": 0.125, + "effective_high": 0.125, + "trace_max_sampling_u": None, + "max_sampling_u": 1.0, + "require_human_confirmation_beyond_trace": True, + "reason": "auto_high_disabled", + }, + } + write_json(trial_path, trial) + trials.append( + { + "execution_index": index, + "config": asdict(config) | {"name": config.name}, + "trial_spec_path": str(trial_path), + "trial_spec_sha256": sha256(trial_path), + } + ) + manifest = { + "schema": SCHEMA, + "created_unix_s": time.time(), + "repository": git_fingerprint(repo), + "runtime": {"python": str(python), "vllm": str(vllm)}, + "model": { + "path": str(model), + "config_sha256": sha256(model / "config.json"), + }, + "trace": {"path": str(trace), "sha256": actual_trace_sha}, + "contract": { + "metric": "maximum SLO-feasible request_rate_per_gpu", + "request_mode": "raw_completion", + "completion_tokens_override": 1, + "target_pass_rate": 0.95, + "search": search_spec(), + "prefix_caching": False, + "chunked_prefill": True, + "cuda_graphs": False, + "custom_all_reduce": False, + "client_max_concurrency": CLIENT_MAX_CONCURRENCY, + "max_server_mns": max(config.mns for config in GRID), + "kv_blocks_source": "community_vllm_measured_and_fixed_per_topology", + }, + "order": {"method": "python_random_shuffle", "seed": args.order_seed}, + "studies": studies, + "trials": trials, + } + write_json(output / "run_manifest.json", manifest) + print(output / "run_manifest.json") + + +def run(args: argparse.Namespace) -> None: + manifest = json.loads(args.manifest.read_text()) + if manifest.get("schema") != SCHEMA: + raise ValueError(f"unexpected manifest schema: {manifest.get('schema')}") + repo = args.repo.resolve() + env = os.environ.copy() + env["PYTHONPATH"] = str(repo / "src") + for trial in manifest["trials"]: + trial_path = Path(trial["trial_spec_path"]) + trial_dir = trial_path.parent + result_path = trial_dir / "result.json" + if result_path.is_file(): + result = json.loads(result_path.read_text()) + if result.get("status") == "completed": + print(json.dumps({"config": trial["config"]["name"], "skipped": True})) + continue + command = [ + str(args.python.resolve()), + "-m", + "aituner.cli", + "worker", + "run-trial", + "--trial-spec", + str(trial_path), + ] + write_json(trial_dir / "worker_command.json", command) + started = time.time() + with (trial_dir / "worker.log").open("w", encoding="utf-8") as output: + completed = subprocess.run( + command, + cwd=repo, + env=env, + stdout=output, + stderr=subprocess.STDOUT, + check=False, + ) + record = { + "config": trial["config"]["name"], + "returncode": completed.returncode, + "elapsed_seconds": time.time() - started, + } + print(json.dumps(record), flush=True) + if completed.returncode != 0: + raise RuntimeError(f"community trial failed: {record}") + assemble(args.manifest) + + +def capacity_interval(probes: list[dict[str, Any]]) -> list[float]: + low, high = 0.0, 0.125 + for probe in probes: + midpoint = float(probe["threshold"]) + if probe["feasible"]: + low = midpoint + else: + high = midpoint + return [low, high] + + +def assemble(manifest_path: Path) -> Path: + manifest = json.loads(manifest_path.read_text()) + records = [] + for trial in manifest["trials"]: + result_path = Path(trial["trial_spec_path"]).parent / "result.json" + if not result_path.is_file(): + raise FileNotFoundError(result_path) + result = json.loads(result_path.read_text()) + if result.get("status") != "completed": + raise ValueError(f"trial did not complete: {result_path}") + tp = int(trial["config"]["tp"]) + request_rate = result.get("best_request_rate") + records.append( + { + "config": trial["config"], + "result_path": str(result_path), + "result_sha256": sha256(result_path), + "capacity_interval_sampling_u": capacity_interval(result["probes"]), + "best_sampling_u": result.get("best_sampling_u"), + "best_request_rate": request_rate, + "best_request_rate_per_gpu": ( + float(request_rate) / tp if request_rate is not None else None + ), + "best_pass_rate": result.get("best_pass_rate"), + } + ) + records.sort( + key=lambda item: ( + -(item["best_request_rate_per_gpu"] or -1.0), + item["config"]["name"], + ) + ) + freeze = { + "schema": SCHEMA, + "run_manifest_sha256": sha256(manifest_path), + "ranking": [dict(record, rank=index + 1) for index, record in enumerate(records)], + } + path = manifest_path.parent / "community_ranking_frozen.json" + write_json(path, freeze) + print(path) + return path + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + subparsers = parser.add_subparsers(dest="command", required=True) + prepare_parser = subparsers.add_parser("prepare") + prepare_parser.add_argument("--repo", type=Path, required=True) + prepare_parser.add_argument("--python", type=Path, required=True) + prepare_parser.add_argument("--vllm", type=Path, required=True) + prepare_parser.add_argument("--model", type=Path, default=Path(MODEL)) + prepare_parser.add_argument("--trace", type=Path, default=Path(TRACE)) + prepare_parser.add_argument("--windows", type=Path, default=Path(WINDOWS)) + prepare_parser.add_argument("--expected-trace-sha256", default=TRACE_SHA256) + prepare_parser.add_argument("--output-root", type=Path, required=True) + prepare_parser.add_argument("--order-seed", type=int, default=ORDER_SEED) + prepare_parser.add_argument("--port", type=int, default=18918) + run_parser = subparsers.add_parser("run") + run_parser.add_argument("--repo", type=Path, required=True) + run_parser.add_argument("--python", type=Path, required=True) + run_parser.add_argument("--manifest", type=Path, required=True) + assemble_parser = subparsers.add_parser("assemble") + assemble_parser.add_argument("--manifest", type=Path, required=True) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + if args.command == "prepare": + prepare(args) + elif args.command == "run": + run(args) + elif args.command == "assemble": + assemble(args.manifest) + else: + raise AssertionError(args.command) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/refined_protocol_manifest.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/refined_protocol_manifest.json new file mode 100644 index 0000000..8671510 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/refined_protocol_manifest.json @@ -0,0 +1,790 @@ +{ + "cohort": [ + { + "input_bin": 0, + "input_tokens": 865, + "prompt_sha256": "c9e22fa0559b409b259bdb15ed6eec1bb36ecb8611ae367b2e122bc6303f56ae", + "sampling_u": 0.46149221772182225, + "source_arrival_s": 8.92599999999993, + "source_request_id": "250", + "source_row_index": 250 + }, + { + "input_bin": 4, + "input_tokens": 9154, + "prompt_sha256": "53b0a0af6cb1c0278183c61b8b6c026977b6cc46e07fca825b5bc8c5802b0e17", + "sampling_u": 0.8156596361943489, + "source_arrival_s": 12.708999999999833, + "source_request_id": "367", + "source_row_index": 367 + }, + { + "input_bin": 0, + "input_tokens": 718, + "prompt_sha256": "87d2846513ce3def0c4a7f9d3707f2eda02069f74535ba69dba6fb761bbfadf5", + "sampling_u": 0.9701010298706338, + "source_arrival_s": 25.190000000000055, + "source_request_id": "682", + "source_row_index": 682 + }, + { + "input_bin": 1, + "input_tokens": 1519, + "prompt_sha256": "7dbffc44084f1dfd9af43c2c40b5e669728a2c01b0a1e29f65ea17bd37e5bffc", + "sampling_u": 0.48115960917218603, + "source_arrival_s": 25.639999999999873, + "source_request_id": "690", + "source_row_index": 690 + }, + { + "input_bin": 1, + "input_tokens": 1968, + "prompt_sha256": "65e1015fbc28a8a2af5ffced6947478c4d68b39cd07a79c5e2fb937ecf9d64a7", + "sampling_u": 0.7394708415743882, + "source_arrival_s": 36.82400000000007, + "source_request_id": "1022", + "source_row_index": 1022 + }, + { + "input_bin": 0, + "input_tokens": 62, + "prompt_sha256": "6ba251e2a1653df7c13c70ed9aa3317af23f42d0ec14b14e46b1376b86f548b7", + "sampling_u": 0.5901445515787429, + "source_arrival_s": 39.96399999999994, + "source_request_id": "1098", + "source_row_index": 1098 + }, + { + "input_bin": 0, + "input_tokens": 702, + "prompt_sha256": "ec3dbd6ac6674268fb420b0e006f7edbbad38906f178faf4c41d5ac3aa7a7968", + "sampling_u": 0.4229843993649856, + "source_arrival_s": 60.55600000000004, + "source_request_id": "1571", + "source_row_index": 1571 + }, + { + "input_bin": 1, + "input_tokens": 1472, + "prompt_sha256": "1fc79d848cc4d3f92ee5a9ad6ecc28a3173737f77ec0b2eaca1b38026225c701", + "sampling_u": 0.38189150492188706, + "source_arrival_s": 69.46300000000019, + "source_request_id": "1792", + "source_row_index": 1792 + }, + { + "input_bin": 2, + "input_tokens": 2701, + "prompt_sha256": "084b29ad36178a06d93a1fdfbed02bc0718b42f43187d184bc06f70f03df4dfc", + "sampling_u": 0.8169083216699623, + "source_arrival_s": 69.79100000000017, + "source_request_id": "1799", + "source_row_index": 1799 + }, + { + "input_bin": 3, + "input_tokens": 7825, + "prompt_sha256": "044f618323ac49afa864bf6b62b1e06a9f3c351647a953cbe38b3173542caf51", + "sampling_u": 0.3751377474987089, + "source_arrival_s": 70.71099999999979, + "source_request_id": "1818", + "source_row_index": 1818 + }, + { + "input_bin": 0, + "input_tokens": 715, + "prompt_sha256": "e5f9aba41593f81170000f1299c71d0f09f52afc58addc2968fb862f8b740359", + "sampling_u": 0.7098663141194472, + "source_arrival_s": 75.24699999999984, + "source_request_id": "1935", + "source_row_index": 1935 + }, + { + "input_bin": 1, + "input_tokens": 1573, + "prompt_sha256": "d69852bb4dbc6b7653ed3a200e891190af5d052d20e57c2ff3abbccc6bf254aa", + "sampling_u": 0.6074460889792176, + "source_arrival_s": 84.5329999999999, + "source_request_id": "2176", + "source_row_index": 2176 + }, + { + "input_bin": 5, + "input_tokens": 29385, + "prompt_sha256": "86cb5ce843bad59ada0e7b9e1a60c1162534b037f5c1b5a7fadca5be26254047", + "sampling_u": 0.17186958800332738, + "source_arrival_s": 93.04399999999987, + "source_request_id": "2396", + "source_row_index": 2396 + }, + { + "input_bin": 1, + "input_tokens": 1939, + "prompt_sha256": "06bad3b6fc6db1c9227a8bb2d4ce349b31802e51e056b541858773debb0a388d", + "sampling_u": 0.29421530368234206, + "source_arrival_s": 98.4050000000002, + "source_request_id": "2505", + "source_row_index": 2505 + }, + { + "input_bin": 1, + "input_tokens": 1532, + "prompt_sha256": "9117e580cf5b8dcedb3c6c02318610c4780fbbf1c4413367db6f7eb60b73459b", + "sampling_u": 0.41616945652066173, + "source_arrival_s": 116.05200000000013, + "source_request_id": "2913", + "source_row_index": 2913 + }, + { + "input_bin": 0, + "input_tokens": 513, + "prompt_sha256": "6cf0a32f0a62d4a16f6abddf0bed8af587ba8f0aff335fd54c72f5772da4464a", + "sampling_u": 0.3613591846324354, + "source_arrival_s": 118.6239999999998, + "source_request_id": "2979", + "source_row_index": 2979 + }, + { + "input_bin": 1, + "input_tokens": 1508, + "prompt_sha256": "ad49c5bb6cf0ced603cc052215c18159d5c25927048ddb7efd6bf315bd84c937", + "sampling_u": 0.1650938738573754, + "source_arrival_s": 121.26600000000008, + "source_request_id": "3056", + "source_row_index": 3056 + }, + { + "input_bin": 0, + "input_tokens": 705, + "prompt_sha256": "c7b6f66c2ade7054abed59f4459bbe4ab4659a3d53654197d27ce26f650450e1", + "sampling_u": 0.8484961251249794, + "source_arrival_s": 130.1909999999998, + "source_request_id": "3301", + "source_row_index": 3301 + }, + { + "input_bin": 4, + "input_tokens": 8287, + "prompt_sha256": "e597785700703dc05da72b1045e734f326ed491276f2d9f45f663f03f1ccd509", + "sampling_u": 0.7289458530141959, + "source_arrival_s": 133.10300000000007, + "source_request_id": "3379", + "source_row_index": 3379 + }, + { + "input_bin": 1, + "input_tokens": 1756, + "prompt_sha256": "b9631ff1dfec71fbd9e3a084bccf30e40a058b74743ae9878e3dd4be77efdf83", + "sampling_u": 0.4082222424943361, + "source_arrival_s": 152.64800000000014, + "source_request_id": "3844", + "source_row_index": 3844 + }, + { + "input_bin": 0, + "input_tokens": 703, + "prompt_sha256": "a943002270d455f4e65022747cf40807044f47fb574089d95deadffed56fcf2e", + "sampling_u": 0.49508596140129646, + "source_arrival_s": 166.43499999999995, + "source_request_id": "4163", + "source_row_index": 4163 + }, + { + "input_bin": 1, + "input_tokens": 1478, + "prompt_sha256": "4b84065a6336bfb9ddd174099740689adfeb5ce294fe670885334934cf583d8b", + "sampling_u": 0.2476209159147146, + "source_arrival_s": 172.2800000000002, + "source_request_id": "4323", + "source_row_index": 4323 + }, + { + "input_bin": 0, + "input_tokens": 686, + "prompt_sha256": "b2154c41a891d72e417d3bb3d67920590349860655d93cb3ff478ad3da74e2f8", + "sampling_u": 0.7190432102580353, + "source_arrival_s": 193.67599999999993, + "source_request_id": "4838", + "source_row_index": 4838 + }, + { + "input_bin": 1, + "input_tokens": 1600, + "prompt_sha256": "9f4cafecff466decfe98d4a9544d05c6246853f24cfeb6e77c9e4d1a5cb6cb02", + "sampling_u": 0.71711230795038, + "source_arrival_s": 194.66400000000021, + "source_request_id": "4870", + "source_row_index": 4870 + }, + { + "input_bin": 2, + "input_tokens": 4000, + "prompt_sha256": "30c385ba8463097d52f838bbfa69f18a32c91b38878e15c11e6b69d6bf81abc7", + "sampling_u": 0.12696475188087025, + "source_arrival_s": 216.49299999999994, + "source_request_id": "5469", + "source_row_index": 5469 + }, + { + "input_bin": 0, + "input_tokens": 710, + "prompt_sha256": "807c8b29c95a0623c585ee7d15d82e72be5858f3683d3657ef3295e71ef6aa10", + "sampling_u": 0.32798737479046675, + "source_arrival_s": 223.39199999999983, + "source_request_id": "5648", + "source_row_index": 5648 + }, + { + "input_bin": 0, + "input_tokens": 727, + "prompt_sha256": "b872bc3174db8e4b17e182608dbb48f99196220dfd55096c2490484a6cadd595", + "sampling_u": 0.3644506007131907, + "source_arrival_s": 239.0999999999999, + "source_request_id": "6031", + "source_row_index": 6031 + }, + { + "input_bin": 4, + "input_tokens": 8786, + "prompt_sha256": "5757f2f9e6be7d4a29a79ad82408e51069044828f55b8c21b7216abbf73b0757", + "sampling_u": 0.12629800224853702, + "source_arrival_s": 260.4920000000002, + "source_request_id": "6604", + "source_row_index": 6604 + }, + { + "input_bin": 4, + "input_tokens": 9518, + "prompt_sha256": "730aee2b4ec388efbf4ca55cbfcbe31c8aa9636c8876b624b37b499d5dcb00f5", + "sampling_u": 0.29338926296423934, + "source_arrival_s": 260.91499999999996, + "source_request_id": "6611", + "source_row_index": 6611 + }, + { + "input_bin": 0, + "input_tokens": 796, + "prompt_sha256": "e0f7312dcc97af6236edac6e47a828f9e18e3e118023537d82f6cb540025191e", + "sampling_u": 0.47864598488768767, + "source_arrival_s": 274.52, + "source_request_id": "6905", + "source_row_index": 6905 + }, + { + "input_bin": 0, + "input_tokens": 228, + "prompt_sha256": "b85c14bcfaec63a99ebc382a7e20d32390a65999062e84c303441344832c3279", + "sampling_u": 0.8157098758872824, + "source_arrival_s": 274.6399999999999, + "source_request_id": "6911", + "source_row_index": 6911 + }, + { + "input_bin": 5, + "input_tokens": 18439, + "prompt_sha256": "8f7259429d8b3b6ab062c885d990fa2136e8ac061ce5b59169e5dbe7cdd5c055", + "sampling_u": 0.44219574639241527, + "source_arrival_s": 287.1959999999999, + "source_request_id": "7271", + "source_row_index": 7271 + }, + { + "input_bin": 0, + "input_tokens": 736, + "prompt_sha256": "1c56453602e159f1dfaae09b5b40785d0dbc176d9b27844362e342e3731c2032", + "sampling_u": 0.6889130017060773, + "source_arrival_s": 291.97400000000016, + "source_request_id": "7403", + "source_row_index": 7403 + }, + { + "input_bin": 4, + "input_tokens": 8557, + "prompt_sha256": "14346a1cc9d1382dd67eaf700f5b52798b77fc8a08bc1df71768d38d613e9e43", + "sampling_u": 0.3112632330750767, + "source_arrival_s": 292.692, + "source_request_id": "7422", + "source_row_index": 7422 + }, + { + "input_bin": 1, + "input_tokens": 1586, + "prompt_sha256": "4d24bf64da9142747d8ee8d87daf50f81309928c3d4792f7c345ecd9dd0c24ac", + "sampling_u": 0.23031083293934718, + "source_arrival_s": 304.69900000000007, + "source_request_id": "7741", + "source_row_index": 7741 + }, + { + "input_bin": 1, + "input_tokens": 1871, + "prompt_sha256": "5a4917df4ebb8a2e4e3dcb0f15c231cd46830af3dc63452ebf421b4433756159", + "sampling_u": 0.10577279384258167, + "source_arrival_s": 312.683, + "source_request_id": "7981", + "source_row_index": 7981 + }, + { + "input_bin": 0, + "input_tokens": 698, + "prompt_sha256": "610618e93c2f4d7bd0ca26b2611b61ffdbbbe092fe0ff3a5f7d7c91f52de7c74", + "sampling_u": 0.43535389643319383, + "source_arrival_s": 319.1579999999999, + "source_request_id": "8179", + "source_row_index": 8179 + }, + { + "input_bin": 0, + "input_tokens": 710, + "prompt_sha256": "3b397d03d7a7001ab8f3915b1b88ed8f8b859ad51769ac752ff35f567ccb1183", + "sampling_u": 0.6383291776591391, + "source_arrival_s": 321.4380000000001, + "source_request_id": "8224", + "source_row_index": 8224 + }, + { + "input_bin": 1, + "input_tokens": 1452, + "prompt_sha256": "671c93bcdd7a292d1a4b4dcce44c679369955fdb4d250d451170c63391216436", + "sampling_u": 0.12477813551209259, + "source_arrival_s": 337.8319999999999, + "source_request_id": "8612", + "source_row_index": 8612 + }, + { + "input_bin": 5, + "input_tokens": 24271, + "prompt_sha256": "a84a31f74c755d02154a7eaeec8bd71450c16813d8b7e713999ff52419f84ed6", + "sampling_u": 0.5538066537263959, + "source_arrival_s": 348.692, + "source_request_id": "8891", + "source_row_index": 8891 + }, + { + "input_bin": 3, + "input_tokens": 4607, + "prompt_sha256": "720cf238e668f4ae567ab909b2ec0185d222a14eb474ae2ef481141a885b3aad", + "sampling_u": 0.03539900309623126, + "source_arrival_s": 382.55499999999984, + "source_request_id": "9768", + "source_row_index": 9768 + }, + { + "input_bin": 0, + "input_tokens": 111, + "prompt_sha256": "8a12215d12ddcf36d1e9013ec0b4a433e4f27c5c2a774ead0f66c93bb31a8063", + "sampling_u": 0.9241501254905812, + "source_arrival_s": 402.8620000000001, + "source_request_id": "10305", + "source_row_index": 10305 + }, + { + "input_bin": 0, + "input_tokens": 714, + "prompt_sha256": "e1dd5099fb8565866691241945b15ee40a5ea8ad9045d10e1dd7774b668d9fa4", + "sampling_u": 0.4533435244095546, + "source_arrival_s": 405.0459999999998, + "source_request_id": "10383", + "source_row_index": 10383 + }, + { + "input_bin": 1, + "input_tokens": 1792, + "prompt_sha256": "9d94bba8c056a775a5d00ecfd6567e83030c99dc20f71ce53d787683ca2450e9", + "sampling_u": 0.8983379345538031, + "source_arrival_s": 405.08100000000013, + "source_request_id": "10385", + "source_row_index": 10385 + }, + { + "input_bin": 0, + "input_tokens": 697, + "prompt_sha256": "7a66c3cec2e1727fa78ed1a7dbe1d39d2c877cc2835c0b10479b396204749984", + "sampling_u": 0.43645445582603326, + "source_arrival_s": 413.16499999999996, + "source_request_id": "10618", + "source_row_index": 10618 + }, + { + "input_bin": 0, + "input_tokens": 709, + "prompt_sha256": "a71447eebb513ece438e0a81a0a9788fbbf34a71d9a66987e6f476cf76f0b7d5", + "sampling_u": 0.4730229506348141, + "source_arrival_s": 424.991, + "source_request_id": "10911", + "source_row_index": 10911 + }, + { + "input_bin": 0, + "input_tokens": 699, + "prompt_sha256": "e04aaa160224e82157f9b0c89180dc341e2c8eebbd518ea8d3327eb73ebfac1c", + "sampling_u": 0.5350250700640652, + "source_arrival_s": 430.51800000000003, + "source_request_id": "11058", + "source_row_index": 11058 + }, + { + "input_bin": 2, + "input_tokens": 3130, + "prompt_sha256": "ba93e2babcf7a0491ba97930f5b3bfb3cc2170dc3cd7b3ca3c3f9fe2905dc72e", + "sampling_u": 0.6597844956490088, + "source_arrival_s": 434.3449999999998, + "source_request_id": "11166", + "source_row_index": 11166 + }, + { + "input_bin": 0, + "input_tokens": 711, + "prompt_sha256": "80c05739d748df64d97f92742f28e384365b26f4f5f1e999666ea2f7f187fd16", + "sampling_u": 0.7240395487416784, + "source_arrival_s": 441.3699999999999, + "source_request_id": "11373", + "source_row_index": 11373 + }, + { + "input_bin": 1, + "input_tokens": 1536, + "prompt_sha256": "ee1209838aca67d2392bb9156b74d69de22d166d97c88b56501ca9cfcc70578e", + "sampling_u": 0.9806533325226275, + "source_arrival_s": 463.8670000000002, + "source_request_id": "11911", + "source_row_index": 11911 + }, + { + "input_bin": 1, + "input_tokens": 1580, + "prompt_sha256": "a4e5fdb5d6119e9207384fd3d32ac027469b3ecca278738c7078026551999786", + "sampling_u": 0.8508070546796023, + "source_arrival_s": 471.08899999999994, + "source_request_id": "12100", + "source_row_index": 12100 + }, + { + "input_bin": 1, + "input_tokens": 1466, + "prompt_sha256": "09a6942491467ee3597eef41c4f36de46bd88b0533af8c1084bb9e15cb150832", + "sampling_u": 0.03536418626599094, + "source_arrival_s": 488.5859999999998, + "source_request_id": "12587", + "source_row_index": 12587 + }, + { + "input_bin": 0, + "input_tokens": 750, + "prompt_sha256": "6a53081dfa161f2b1e50e03885b0d2868c7390f91130d0a1cd95b4b75cdb1126", + "sampling_u": 0.4344415561103455, + "source_arrival_s": 488.6289999999999, + "source_request_id": "12588", + "source_row_index": 12588 + }, + { + "input_bin": 0, + "input_tokens": 736, + "prompt_sha256": "3514d193ff83b6d6589d52f24072b0159fa00d7b44ca8c6ec269515db4cfba89", + "sampling_u": 0.7257534252552246, + "source_arrival_s": 497.2449999999999, + "source_request_id": "12824", + "source_row_index": 12824 + }, + { + "input_bin": 0, + "input_tokens": 90, + "prompt_sha256": "915b3aee8d1288262c7f363f9035af19236e45540b748a9ced17c2d23f428f34", + "sampling_u": 0.7669962160527245, + "source_arrival_s": 520.1750000000002, + "source_request_id": "13350", + "source_row_index": 13350 + }, + { + "input_bin": 0, + "input_tokens": 697, + "prompt_sha256": "af93c55d3eb4f5161eb69410e73039e1d500a92190bff884e1b4040e491c8ed5", + "sampling_u": 0.594626944330893, + "source_arrival_s": 525.469, + "source_request_id": "13491", + "source_row_index": 13491 + }, + { + "input_bin": 1, + "input_tokens": 1470, + "prompt_sha256": "dd86d891c322f43310d4207c02f27a7d5e76885c420793164c76020f7ff166c4", + "sampling_u": 0.9612563305576178, + "source_arrival_s": 526.116, + "source_request_id": "13510", + "source_row_index": 13510 + }, + { + "input_bin": 1, + "input_tokens": 1592, + "prompt_sha256": "2ab08103a5ec0e01aea2f74f9215c52222ef8722a2fb9327bd9187a644df1c4b", + "sampling_u": 0.7977149220573079, + "source_arrival_s": 529.1809999999996, + "source_request_id": "13612", + "source_row_index": 13612 + }, + { + "input_bin": 5, + "input_tokens": 26776, + "prompt_sha256": "d09d69f9110f28c69d0767bdd501a8bdba167775ae1c6e6c53d15cf0fdcf5f6e", + "sampling_u": 0.9888861232028356, + "source_arrival_s": 546.9939999999997, + "source_request_id": "14026", + "source_row_index": 14026 + }, + { + "input_bin": 1, + "input_tokens": 1589, + "prompt_sha256": "973120a2be961f8f44523988a44f59b2071e82081a292cf37f10fb72d2874535", + "sampling_u": 0.8325378218001509, + "source_arrival_s": 549.6970000000001, + "source_request_id": "14094", + "source_row_index": 14094 + }, + { + "input_bin": 1, + "input_tokens": 1550, + "prompt_sha256": "8d899d80a6b9c47f414f6d65ed1638b5fc3bc782eaa19b2592878f35ff5484e7", + "sampling_u": 0.7285184471546545, + "source_arrival_s": 556.8310000000001, + "source_request_id": "14316", + "source_row_index": 14316 + }, + { + "input_bin": 2, + "input_tokens": 2501, + "prompt_sha256": "4a88c92c72a1fd9a653a7dbbd169eff30774f1b95cb5482f60c45532923d9f47", + "sampling_u": 0.38599257949920635, + "source_arrival_s": 558.3209999999999, + "source_request_id": "14359", + "source_row_index": 14359 + }, + { + "input_bin": 0, + "input_tokens": 741, + "prompt_sha256": "9c2341411206e2d28d6479ef20d5ce9e90daac574e72c4066130e2601731624c", + "sampling_u": 0.5930037066322071, + "source_arrival_s": 583.5829999999996, + "source_request_id": "15069", + "source_row_index": 15069 + }, + { + "input_bin": 3, + "input_tokens": 6044, + "prompt_sha256": "614a0d743b4cec35549b490110bc71d8fc3a22cd07c68de7281fc98270f81ee8", + "sampling_u": 0.4726761733329724, + "source_arrival_s": 594.8950000000004, + "source_request_id": "15377", + "source_row_index": 15377 + } + ], + "created_unix_s": 1784139549.1404486, + "load_contract": { + "binary_search": false, + "completion_tokens_override": 1, + "monotonicity_assumed": false, + "offered_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "only_mutated_variable": "uniform_arrival_timeline_scale", + "request_count_per_point": 64, + "target_pass_rate": 0.95 + }, + "primary_slo": "linear_8k_primary", + "rates": { + "r_0p150": { + "duration_s": 426.6666666666667, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.15, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p150.csv", + "request_count": 64, + "request_rate_recomputed": 0.15, + "sha256": "f35ce4424c17b2ee16144bb59011e467a07d3f57895716165fcdc73ba19839d9", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p200": { + "duration_s": 320.0, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.2, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p200.csv", + "request_count": 64, + "request_rate_recomputed": 0.2, + "sha256": "b81a48706a5c41222477aef8da357ae14c7b0bc19d35ea5ecc0c12b86d32590c", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p250": { + "duration_s": 256.0, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.25, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p250.csv", + "request_count": 64, + "request_rate_recomputed": 0.25, + "sha256": "721d622abb5d0a9080d6577063af910906b4e9928b402f48804ee55b25f108fc", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p300": { + "duration_s": 213.33333333333334, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.3, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p300.csv", + "request_count": 64, + "request_rate_recomputed": 0.3, + "sha256": "a5c8a7b93f314bf4365763fc0157597c582229993bb79b41b697ef3247495e49", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p350": { + "duration_s": 182.85714285714286, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.35, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p350.csv", + "request_count": 64, + "request_rate_recomputed": 0.35, + "sha256": "ec9debce28b76981f4f5363f0c77faa963c550a5d3ecbed327a304dc0d495242", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p400": { + "duration_s": 160.0, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.4, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p400.csv", + "request_count": 64, + "request_rate_recomputed": 0.4, + "sha256": "aea1cae38b09ddb0ad1a7d6da03bd6ad76afdaee3b2879b7cb65eba9b3ef5bab", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p450": { + "duration_s": 142.22222222222223, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.45, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p450.csv", + "request_count": 64, + "request_rate_recomputed": 0.44999999999999996, + "sha256": "672197deca834d9b9c5aea963448805a03e86495721d5eb270ccec2416faada2", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p500": { + "duration_s": 128.0, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.5, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p500.csv", + "request_count": 64, + "request_rate_recomputed": 0.5, + "sha256": "7cd2d16904e0fafb26a67718c4f2d7b6752934864219ab96813b8f62b47019d4", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_0p750": { + "duration_s": 85.33333333333333, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 0.75, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_0p750.csv", + "request_count": 64, + "request_rate_recomputed": 0.75, + "sha256": "e8b00f14b6c85ac8bd1c8fe3b83797f5c61903d70d8bbf6acfe9a1bd384db443", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_1p000": { + "duration_s": 64.0, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 1.0, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_1p000.csv", + "request_count": 64, + "request_rate_recomputed": 1.0, + "sha256": "7071e0247414af9214a1bf06f05a4865965568ba812bbae6713648e994a40a34", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + }, + "r_1p500": { + "duration_s": 42.666666666666664, + "input_length_order_sha256": "d796fc955a9e7278512b56ce0544afd465d689e0060920c18839e85632360645", + "offered_request_rate": 1.5, + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/traces/r_1p500.csv", + "request_count": 64, + "request_rate_recomputed": 1.5, + "sha256": "86b03ef1fe87ad3c589f1585a279cd3136975009316e36e0910fe8340b972eee", + "source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5" + } + }, + "schema": "qwen235b-prefill-fixed-cohort-protocol-v1", + "selection": { + "cohort_bin_quotas": [ + 27, + 21, + 4, + 3, + 5, + 4 + ], + "cohort_size": 64, + "cohort_source_row_order_sha256": "d3945dc4a79840a0a157a73f10f2836006a3f0bbc5ecdbd657886865497affa5", + "input_bin_edges": [ + 0, + 1024, + 2048, + 4096, + 8192, + 16384, + 32769 + ], + "method": "length_stratified_smallest_sha256", + "seed": 2026071501, + "source_arrival_span_s": 585.9690000000005, + "source_bin_counts": [ + 6549, + 5111, + 916, + 743, + 1194, + 896 + ], + "stable_order": [ + "source_arrival_s", + "source_row_index" + ] + }, + "slo_variants": { + "legacy_step_1s_2s": { + "buckets": [ + { + "max_input_tokens": 8191, + "threshold_ms": 1000.0 + }, + { + "threshold_ms": 2000.0 + } + ], + "description": "Original strict 1 s/2 s step SLO (null-capacity sensitivity)", + "kind": "step_ms" + }, + "linear_10k": { + "description": "TTFT <= 1000 ms + input_tokens / 10000 tokens/s", + "intercept_ms": 1000.0, + "kind": "linear_ms", + "per_token_ms": 0.1 + }, + "linear_6k": { + "description": "TTFT <= 1000 ms + input_tokens / 6000 tokens/s", + "intercept_ms": 1000.0, + "kind": "linear_ms", + "per_token_ms": 0.16666666666666666 + }, + "linear_8k_primary": { + "description": "TTFT <= 1000 ms + input_tokens / 8000 tokens/s", + "intercept_ms": 1000.0, + "kind": "linear_ms", + "per_token_ms": 0.125 + } + }, + "source": { + "eligible_request_count": 15409, + "path": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/thinking_w20260327_1000.jsonl", + "sha256": "f878e9af18f94dcfaced94a8e1e6b20a2f7d97d64aa862448025660dbbd965b2" + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/report.md b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/report.md new file mode 100644 index 0000000..dfe9855 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/report.md @@ -0,0 +1,204 @@ +# Qwen3-235B FP8：Frontier 与 community vLLM config rank 对比 + +## 结论 + +在本实验限定的 Qwen3-235B-A22B-FP8、H20、prefill-only workload 上，经过 best-effort 补齐 FP8/MoE serving path 后，Frontier **能够选对满足 SLO 的 per-GPU throughput 最优配置**：`TP=4, MBT=16384`，`MNS=64/128` 并列。细化网格上的 Spearman rank correlation 为 `0.9487`，20 个可比较非 tie config pair 全部同序，选择 regret 为 `0`。 + +但这不等于 Frontier 已精确模拟系统。原始粗网格给出 56/56 load labels 正确、Spearman=1 的表象；加入 decision-boundary rates 后，Frontier 在 33 个被测试边界点中产生 6 个 false-infeasible，均为两个 MNS 配置上可复现的稳定误差。它系统性地把三个 action family 的最大经验 SLO throughput 低估一个 `0.05 req/s` 档位： + +| TP | MBT | Frontier capacity | real capacity | real v2→v3 transition | 结论 | +|---:|---:|---:|---:|---|---| +| 4 | 8192 | 0.15 | 0.20 | `0.20: 61→61`, `0.25: 60→60` | 稳定低估 0.05 | +| 4 | 16384 | 0.25 | 0.30 | `0.30: 61→61`, `0.35: 59→59` | 稳定低估 0.05 | +| 8 | 8192 | 0.40 | 0.45 | `0.45: 61→61`, `0.50: 60→60` | 稳定低估 0.05 | +| 8 | 16384 | 0.45 | 0.45 | v2: `0.45: 61`, `0.50: 60` | 边界正确 | + +表中 pass count 分母均为 64，经验可行门槛为 61。每一行的两个 MNS 结果一致；v3 对前三行的两个 MNS 都做了 fresh-server 复测。 + +因此，对这个 case 最准确的判断是： + +- Frontier 有较好的 **decision/ordinal fidelity**，不能再用“它找不到最优 config”作为该 case 的研究 premise。 +- Frontier 没有达到 **capacity calibration fidelity**，也没有完全学对 action effect；它适合筛选，但不能在 SLO knee 附近代替真机 capacity certification。 +- 粗网格上的“完全匹配”并不足以证明 simulator solve 了问题；它会同时掩盖绝对容量误差和错误的 action differential。 + +## 固定实验口径 + +| 项目 | 固定值 | +|---|---| +| model | `models/Qwen/Qwen3-235B-A22B-FP8` | +| real runtime | community vLLM 0.10.2 | +| hardware | dash0，8×NVIDIA H20 | +| workload | prefill-only，raw completion，output tokens=1，prefix cache disabled | +| config surface | `TP∈{4,8} × MNS∈{64,128} × MBT∈{8192,16384}` | +| cohort | 64 个固定 prompt；所有 config/rate 的 request ID、顺序、输入长度完全相同 | +| input length | p50=1478，p95=18439，max=29385，mean=3511.55 tokens | +| rates | 0.15, 0.20, 0.25, 0.30, 0.35, 0.40, 0.45, 0.50, 0.75, 1.00, 1.50 req/s | +| primary SLO | `TTFT ≤ 1000 ms + input_tokens / 8000 tokens/s`，至少 95% requests 通过 | +| capacity | 所有 64 requests 完整结束后，最大被测试可行 offered rate；不做 early stop/二分单调假设 | +| primary objective | capacity / TP，即 per-GPU request throughput | + +cohort 只通过 uniform arrival scaling 改变 load，不随 rate 重新抽样。该设计消除了 workload composition confound，但 64-request 固定 cohort 的结果仍是经验容量，不代表总体 workload distribution 的置信界。 + +## Frontier best-effort 公平性 + +这里比较的不是未经修改的 stock Frontier。为了避免“故意把 baseline 调差”的质疑，我们在真机 ranking 冻结前完成并冻结了以下补齐： + +- 使用 community vLLM 0.10.2 在 H20 上得到的真实 FP8 operator profiles；quantization contract 为 FP8 W8A8 dynamic activation、BF16 output/accumulation，而不是用 BF16 compute profile 冒充 FP8。 +- 补齐 Qwen MoE community serving path、MoE plan 与 serving entrypoint。 +- 按真实 expert route 构造 EP local lanes，并以最慢 lane 作为同步 critical path。 +- 修正 TP/EP-aware profile cache key，避免不同 topology 错用缓存结果。 +- 使用 batched lane prediction；与逐 lane exact path 做 byte-identical equivalence 后才用于完整 sweep，模拟运行加速约 2.57×。 +- KV block capacity 使用真机测量值：TP4=26101，TP8=62351。 +- Frontier refined 8×11=88 个结果在真机运行前冻结；旧 rates 的 56 个结果与先前 freeze byte-identical。没有用真机 v2/v3 结果回调 simulator 参数。 + +相关实现与审计入口包括： + +- `fixed_cohort_rank.py` +- `frontier_fp8_profile_contract.patch` +- `frontier_moe_serving_path.patch` +- `frontier_moe_ep_prefill_exact_cache.patch` +- `frontier_moe_ep_prefill_use_batched_lanes.patch` +- `frontier_moe_ep_prefill_batched_lane_predict.patch` + +这说明结果反映的是“给 Frontier 当前抽象最大的合理帮助以后，它能做到什么”，不是 stock baseline 的缺功能测试。 + +## 三阶段实验结果 + +### v1：blind 全表面、粗网格 + +Frontier 先冻结，再运行 community vLLM 的 8 configs × 7 rates： + +- 56/56 config-load primary feasibility labels 正确。 +- raw capacity 与 per-GPU efficiency 的 top set 均完全匹配。 +- Spearman=1，20/20 非 tie pair 同序，top-1 regret=0。 +- 但 3584 个 request 的 TTFT MAE 为 124.20 ms，MAPE 为 20.98%，平均 `predicted-real=+93.20 ms`。 +- request-level SLO label accuracy 为 96.82%；Frontier/real failure-set Jaccard 仅 0.7569（103 false fail、11 false pass）。 + +因此，aggregate capacity 完全匹配时，具体失败请求和 request latency mechanism 已明显不匹配。 + +### v2：反向 config/rate 顺序，加入细化边界 + +v2 只重跑预注册 SLO transition 及中间 refinement rates，共 8 configs、33 loads、2112 requests： + +- 0 server/request failure，0 early stop，0 monotonicity violation；所有 trace hash 匹配 refined protocol。 +- per-GPU top set 仍精确匹配：`TP4, MBT16384, MNS64/128`。 +- Frontier/real per-GPU capacity Spearman=0.9487，20/20 非 tie pair 同序，regret=0。 +- 33 个边界点中 27 个 feasibility label 正确；6 个错误全部是 false-infeasible，没有 false-feasible。 +- pass-rate MAE=1.278 percentage points，平均 `Frontier-real=-1.278 pp`。 +- 2112 个 request 的 TTFT MAE=83.15 ms，MAPE=16.34%，平均 `predicted-real=+53.02 ms`。 +- request SLO label accuracy=98.34%，failure-set Jaccard=0.7535。 + +v1 与 v2 的 19 个共同 config-load 点包含 1216 个相同 requests： + +- 真机 trial-to-trial TTFT MAE=14.05 ms，absolute-error p95=57.94 ms，平均 v2-v1=+1.75 ms。 +- primary request label 只有 1/1216 flip；19/19 load labels 和 8/8 原始网格 capacity 全部稳定。 +- Frontier 有 23 个 request label 在两次真机 trial 中都稳定预测错误。 + +这排除了 config order 和普通 runtime noise 是主要 residual 来源。 + +### v3：只复测 v2 的全部争议 transition + +v3 的选择规则是 v2 中所有 Frontier/real primary capacity 不同的 config；它是 data-dependent replication，不作为新的 blind primary result。共 6 configs、12 loads、768 requests： + +- 0 failure/early-stop/monotonicity violation，trace hash 全部匹配。 +- v2 与 v3 的 768/768 primary request labels 完全一致，failure-set Jaccard=1.0。 +- 12/12 config-load labels、6/6 capacity boundaries 完全一致。 +- 真机 trial-to-trial TTFT MAE=14.98 ms，absolute-error p95=62.52 ms，平均 v3-v2=+1.55 ms。 +- Frontier 在该争议集合上仍有 6/12 false-infeasible；18 个 request labels 在 v2/v3 都稳定预测错误。 + +这证明细化边界误差不是一次随机跨阈值。 + +## 系统层面的理解 + +### 1. 绝对误差与 action differential 必须分开 + +如果只是一个全局 latency scale 偏差，所有 topology 下 MBT 的相对收益应基本保持。数据并非如此： + +- TP4：Frontier 预测 MBT `8192→16384` 的 capacity 增益为 `0.15→0.25`，即 +0.10 req/s；真机为 `0.20→0.30`，同样 +0.10。 +- TP8：Frontier 预测 `0.40→0.45`，即 +0.05；真机为 `0.45→0.45`，实际增益为 0。 + +也就是说，Frontier 在 TP4 学对了 MBT differential、错了 absolute offset；在 TP8 则连 differential 都错了。该误差依赖 TP/EP topology，不能靠一个全局 calibration factor 修复。这正是 execution configuration 与 scheduler state/action effect 相互耦合的证据。 + +### 2. 决策正确不代表 simulator 对系统有正确解释 + +per-GPU objective 下，TP4/MBT16384 与其他配置有足够大的 margin，所以即便 capacity 每类偏差 0.05，最优 config 仍不变。raw single-replica objective 下，Frontier 只把 TP8/MBT16384 列为 top set，而真机四个 TP8 config 全部在 0.45 并列；虽然从 Frontier top set 中任选仍是 zero regret，但它虚构了 TP8 上 MBT 的收益。 + +因此需要同时报告： + +1. selected-config regret； +2. rank/order fidelity； +3. capacity error； +4. action differential error； +5. request/failure-set fidelity。 + +只报其中任意一层都会得到偏颇结论。 + +### 3. residual 是状态/长度相关尾部误差，不是 profile 全局比例错误 + +v2 中 Frontier 的平均 signed TTFT error 随输入长度显著变化： + +| input tokens | mean `predicted-real` | MAE | +|---|---:|---:| +| [0,1024) | -3.61 ms | 40.74 ms | +| [1024,2048) | +44.45 ms | 57.63 ms | +| [2048,4096) | +91.21 ms | 109.73 ms | +| [4096,8192) | +138.80 ms | 140.80 ms | +| [8192,16384) | +237.37 ms | 237.37 ms | +| [16384,32769) | +147.36 ms | 240.89 ms | + +短请求甚至略偏乐观，中长请求明显偏保守；与此同时 aggregate load label 只由 1–2 个请求是否跨 SLO 阈值决定。这更像 batch composition、critical-lane/EP synchronization 和 queueing state 共同造成的 residual，而不是 operator profile 全局慢了固定百分比。 + +### 4. MNS 在该 workload 上是非活跃维度，但不能外推 + +在所有共同测试点中，MNS=64/128 的 pass counts 基本逐点相同，Frontier 也预测 tie。对这个 prefill-only、固定 cohort、当前 concurrency 范围，MBT/token budget 和 topology 是活跃因素，MNS 没有形成独立瓶颈。但 decode/mixed workload 的 sequence residency、KV pressure 和 iteration cadence 不同，不能据此宣称 MNS 普遍无效。 + +## 对研究方向的判断 + +这个 case 不支持以下论文 premise： + +> Frontier 类 simulator 无法为 Qwen3-235B FP8 选择最优配置，所以必须由一个新 tuner 替代。 + +在公平补齐 FP8/MoE path 后，它已经能以 zero regret 选出该 workload 的 per-GPU 最优配置。继续用这一 case 证明“simulator 排序失败”会被 reviewer 用本组数据直接反驳。 + +本 case 真正揭示、且更有 systems 内容的问题是： + +> 一个 simulator 可以在 coarse grid 上拥有完美 rank，并在最终 objective 上 zero regret，同时仍错误表达 topology-dependent action differential 和 SLO knee；我们如何判断一个模拟决策何时足够可信，何时必须回到真机？ + +可继续发展的研究问题包括： + +- decision-aware uncertainty：不追求每条 TTFT 精确，而是给 config ordering/capacity boundary 提供可校验置信区间。 +- counterexample-guided measurement：只在模拟 action margin 小或 differential 不稳定的 boundary/state 上调用真机，而不是全表面 sweep。 +- topology-aware residual model：显式建模 TP/EP、batch composition、critical lane 与输入长度共同造成的尾部偏差，避免全局比例校准。 +- distinguish safe screening from certification：Frontier 当前在该 case 是保守筛选器（0 false-feasible），但不是精确 capacity certifier。 + +下一步要判断这是否足以构成更一般的 research premise，应转向已计划的 Qwen3-235B decode-only 和 mixed prefill/decode case。它们会激活 KV state、replica residency、iteration cadence 和 communication overlap；如果只有这些 case 出现稳定排序/regret 错误，研究问题就应围绕 simulator 在 stateful execution/scheduling coupling 下的决策失效，而不是泛化地宣称 simulator 不工作。 + +## 证据与复现 + +本目录保存紧凑 comparison、trial-stability 与 manifest；完整 request-level freeze 保留在 dash0： + +`/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1` + +关键 artifact SHA256： + +| artifact | SHA256 | +|---|---| +| refined protocol manifest | `695c0500b9a5d51cdd4e39b722e3c9db15f8bed48795454f4d0d6442d24929cb` | +| Frontier refined freeze | `318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336` | +| v1 blind comparison | `fc87c54179b891e2148a738dbdfeed328595fa32198f2248c892c890625c9fe2` | +| v2 run manifest | `cf1a722043f42ff8f435e3521d0b2b2892a8ea0922e1e4870ec10fba26c0f998` | +| v2 community freeze | `5a9c59fc4082db47bec7b59432e9ebe9f907f70baa95a9f0d3480e85f8992ff2` | +| v2 comparison | `d21f2dd846c4d8361668f312f0c063f2e8f949ec7a9d3b19100c813b3da3b586` | +| v1-v2 stability | `a522189618ad26df68f18d2d9b67c38b1fe59c5c8f13af633eced0b092f8e06d` | +| v3 run manifest | `0f8908671b8903e516f0dc8e68f556483a15ff7ec51f7cfbc18a9c74b3a5ba61` | +| v3 community freeze | `4edd25edde3d7508d8feb9b024b5bca4ba6a5f25335ac84008f09c524ed7bc34` | +| v3 comparison | `e04f40ead995e007d64b2f6f0f2731999204907770f05b4c2d037d224e49c573` | +| v2-v3 stability | `7be2ff27dd0393c99f0512bd6a0b7b35eb404ee0bfd22bd3ec33a4b8ce14748b` | + +## 限制 + +- 只测试了一个模型、H20、community vLLM 0.10.2 和 prefill-only workload。 +- 固定 64-request cohort 适合 simulator-vs-real paired comparison，但不足以给总体 workload distribution 的 95% SLO capacity 做统计认证。 +- v3 是在观察 v2 后选择全部争议点的 replication，不是 blind discovery dataset。 +- operator profiles、KV capacity 和 MoE route contract 本身使用了真机信息；本结果评估的是 best-effort simulator 的决策能力与剩余误差，不是 zero-measurement transfer。 +- 没有测试 decode、mixed phase、prefix caching、multi-replica placement、异构 GPU 或跨节点通信。 diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_blind_comparison.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_blind_comparison.json new file mode 100644 index 0000000..001eafa --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_blind_comparison.json @@ -0,0 +1,6384 @@ +{ + "community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/community_ranking_frozen.json", + "sha256": "055fe2cf1e0c01a0503145a272097b8ceda257f1b86bdae39abf17277d9a18d9" + }, + "created_unix_s": 1784139490.8308787, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/frontier-v3-batched-par4/frontier_ranking_frozen.json", + "sha256": "9a971f25cc97975845968cc776241c7d44a49dee2ec2f015f227bcff82e66971" + }, + "pass_rate_surface_residuals": { + "legacy_step_1s_2s": { + "false_feasible_points": 0, + "false_infeasible_points": 0, + "feasibility_label_accuracy": 1.0, + "mae": 0.041294642857142856, + "matched_config_load_points": 56, + "max_absolute_error": 0.15625, + "signed_error_mean": -0.03962053571428571 + }, + "linear_10k": { + "false_feasible_points": 2, + "false_infeasible_points": 0, + "feasibility_label_accuracy": 0.9642857142857143, + "mae": 0.033482142857142856, + "matched_config_load_points": 56, + "max_absolute_error": 0.140625, + "signed_error_mean": -0.029575892857142856 + }, + "linear_6k": { + "false_feasible_points": 0, + "false_infeasible_points": 1, + "feasibility_label_accuracy": 0.9821428571428571, + "mae": 0.022600446428571428, + "matched_config_load_points": 56, + "max_absolute_error": 0.109375, + "signed_error_mean": -0.020926339285714284 + }, + "linear_8k_primary": { + "false_feasible_points": 0, + "false_infeasible_points": 0, + "feasibility_label_accuracy": 1.0, + "mae": 0.026227678571428572, + "matched_config_load_points": 56, + "max_absolute_error": 0.125, + "signed_error_mean": -0.025669642857142856 + } + }, + "primary_slo": "linear_8k_primary", + "request_level_ttft_residuals": { + "by_input_bin": { + "[0,1024)": { + "count": 1512, + "mae_ms": 80.67140471814413, + "signed_error_ms_mean": 40.081426915862 + }, + "[1024,2048)": { + "count": 1176, + "mae_ms": 113.2322821016931, + "signed_error_ms_mean": 95.35554067785822 + }, + "[16384,32769)": { + "count": 224, + "mae_ms": 248.44144323616518, + "signed_error_ms_mean": 151.67347724381298 + }, + "[2048,4096)": { + "count": 224, + "mae_ms": 146.4061790530529, + "signed_error_ms_mean": 116.84675686840379 + }, + "[4096,8192)": { + "count": 168, + "mae_ms": 150.66978579178092, + "signed_error_ms_mean": 148.27281627513304 + }, + "[8192,16384)": { + "count": 280, + "mae_ms": 272.270894330094, + "signed_error_ms_mean": 272.270894330094 + } + }, + "mae_ms": 124.19937765168956, + "mape": 0.20982454623810715, + "matched_successful_requests": 3584, + "signed_error_ms_mean": 93.2018552795005 + }, + "request_slo_classification": { + "legacy_step_1s_2s": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "1098" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "6611", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "1799", + "3379", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "1935", + "4163", + "4323", + "8612" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "6611", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "1799", + "3379", + "6611", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "1935", + "4163", + "4323", + "8612" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "2979", + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "1935", + "8179" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "15377", + "1799", + "3379", + "3844", + "6611", + "6905", + "6911", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.84375, + "matched_requests": 64, + "matching_labels": 54, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "2979", + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "1935", + "8179" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1799", + "1818", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "15377", + "1799", + "3379", + "3844", + "6611", + "6905", + "6911", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.84375, + "matched_requests": 64, + "matching_labels": 54, + "offered_request_rate": 1.5 + } + ], + "failure_set_intersection": 581, + "failure_set_jaccard": 0.7926330150068213, + "failure_set_union": 733, + "false_slo_fail_requests": 147, + "false_slo_pass_requests": 5, + "frontier_failure_requests": 728, + "label_accuracy": 0.9575892857142857, + "matched_requests": 3584, + "matching_labels": 3432, + "real_failure_requests": 586 + }, + "linear_10k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "1098", + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7741" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "6611", + "7981" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611", + "6905", + "6911", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "6604", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "6604", + "7981" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "6604", + "6905", + "6911", + "8612" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.90625, + "matched_requests": 64, + "matching_labels": 58, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7741" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "6611", + "7981" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611", + "6905", + "6911", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "6604", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "6604", + "7981" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "6604", + "6905", + "8612" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "2979", + "7271", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "3379", + "7271", + "7981", + "8179" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "2396", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "15069", + "2396", + "3379", + "6611", + "7271", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.859375, + "matched_requests": 64, + "matching_labels": 55, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "2979", + "7271", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "3379", + "7271", + "7981", + "8179" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "2396", + "3056", + "6611", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "15069", + "2396", + "3379", + "6611", + "7271", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.859375, + "matched_requests": 64, + "matching_labels": 55, + "offered_request_rate": 1.5 + } + ], + "failure_set_intersection": 415, + "failure_set_jaccard": 0.7167530224525043, + "failure_set_union": 579, + "false_slo_fail_requests": 135, + "false_slo_pass_requests": 29, + "frontier_failure_requests": 550, + "label_accuracy": 0.9542410714285714, + "matched_requests": 3584, + "matching_labels": 3420, + "real_failure_requests": 444 + }, + "linear_6k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "1098" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6905", + "6911", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.90625, + "matched_requests": 64, + "matching_labels": 58, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15377", + "1935", + "4163", + "6611", + "6905", + "6911", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6905", + "6911", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.90625, + "matched_requests": 64, + "matching_labels": 58, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15377", + "1935", + "4163", + "6611", + "6905", + "6911", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "2979", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7422", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14094" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "2505" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "2913", + "2979", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7422", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14094" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "2505" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + } + ], + "failure_set_intersection": 306, + "failure_set_jaccard": 0.7906976744186046, + "failure_set_union": 387, + "false_slo_fail_requests": 78, + "false_slo_pass_requests": 3, + "frontier_failure_requests": 384, + "label_accuracy": 0.9773995535714286, + "matched_requests": 3584, + "matching_labels": 3503, + "real_failure_requests": 309 + }, + "linear_8k_primary": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "1098" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "14026", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7741" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "7981" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611", + "6905", + "6911" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "2396" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7271", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "6611", + "7271", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "6611", + "6905", + "6911", + "7271", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "14026", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7741" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "3379", + "7981" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1935", + "4163", + "4323", + "6611", + "6905", + "6911" + ], + "false_slo_pass_request_ids": [ + "14026" + ], + "label_accuracy": 0.890625, + "matched_requests": 64, + "matching_labels": 57, + "offered_request_rate": 1.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "2396" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7271", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.75 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "6611", + "7271", + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 1.0 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15377", + "1935", + "4163", + "6611", + "6905", + "6911", + "7271", + "9768" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.875, + "matched_requests": 64, + "matching_labels": 56, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "14094" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "2979", + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "14094" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "2979", + "7422", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7981" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 1.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "3056", + "7741" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 1.0 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "15069", + "3379", + "7981", + "8179", + "8224" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 1.5 + } + ], + "failure_set_intersection": 355, + "failure_set_jaccard": 0.7569296375266524, + "failure_set_union": 469, + "false_slo_fail_requests": 103, + "false_slo_pass_requests": 11, + "frontier_failure_requests": 458, + "label_accuracy": 0.9681919642857143, + "matched_requests": 3584, + "matching_labels": 3470, + "real_failure_requests": 366 + } + }, + "schema": "frontier-community-qwen235b-rank-comparison-v1", + "variants": { + "legacy_step_1s_2s": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.84375, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.84375, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.875, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.875, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.875, + 0.84375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.890625, + 0.859375, + 0.84375, + 0.78125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.875, + 0.84375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.890625, + 0.859375, + 0.84375, + 0.78125, + 0.75, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.921875, + 0.875, + 0.859375, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.921875, + 0.875, + 0.859375, + 0.8125, + 0.734375 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.921875, + 0.875, + 0.859375, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.921875, + 0.875, + 0.859375, + 0.8125, + 0.734375 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "frontier_capacity": { + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + } + } + }, + "linear_10k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.859375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.859375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0375, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.0375, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.921875, + 0.890625, + 0.828125 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.90625, + 0.875, + 0.828125 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.921875, + 0.890625, + 0.828125 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.90625, + 0.875, + 0.828125 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.765625, + 0.671875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.75, + 0.625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.875, + 0.84375, + 0.8125, + 0.765625, + 0.671875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.75, + 0.609375 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + } + } + }, + "linear_6k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.9375, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.921875, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.9375, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.921875, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0625, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0625, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 12, + "correct_pairs": 12, + "frontier_tied_pairs": 12, + "real_tied_pairs": 13 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0625, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0625, + "tp8_mns128_mbt16384": 0.0625, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192", + "tp8_mns128_mbt16384" + ], + "spearman_rank_correlation": 0.7745966692414834, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "top1_set_match": false + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.828125, + 0.734375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.828125, + 0.734375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.875, + 0.828125, + 0.734375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.828125, + 0.734375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.5, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.84375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.890625, + 0.828125 + ], + "rank": 6, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.921875, + 0.84375 + ], + "rank": 6, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.890625, + 0.828125 + ], + "rank": 6, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.25, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 16, + "correct_pairs": 16, + "frontier_tied_pairs": 12, + "real_tied_pairs": 9 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.25, + "tp8_mns128_mbt16384": 0.5, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384" + ], + "spearman_rank_correlation": 0.9561828874675149, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.30000000000000004, + "top1_set_intersection": [ + "tp8_mns128_mbt16384" + ], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.25 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.25 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "top1_set_match": true + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.5, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.30000000000000004, + "top1_set_intersection": [ + "tp8_mns128_mbt16384" + ], + "top1_set_match": false + } + } + }, + "linear_8k_primary": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.890625, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.890625, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 8, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.953125, + 0.921875, + 0.90625, + 0.84375, + 0.78125, + 0.6875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.921875, + 0.90625, + 0.84375, + 0.78125, + 0.6875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.90625, + 0.828125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.921875, + 0.875, + 0.828125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.96875, + 0.9375, + 0.921875, + 0.90625, + 0.828125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.921875, + 0.875, + 0.828125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.90625, + 0.875, + 0.84375, + 0.796875, + 0.6875 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.90625, + 0.875, + 0.84375, + 0.796875, + 0.703125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.25, + 0.35, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 8, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 4, + "correct_pairs": 4, + "frontier_tied_pairs": 2, + "real_tied_pairs": 2 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + } + } + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_v2_trial_stability.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_v2_trial_stability.json new file mode 100644 index 0000000..b64a7d5 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v1_v2_trial_stability.json @@ -0,0 +1,1038 @@ +{ + "created_unix_s": 1784150483.6645162, + "inputs": { + "first_community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/community_ranking_frozen.json", + "sha256": "055fe2cf1e0c01a0503145a272097b8ceda257f1b86bdae39abf17277d9a18d9" + }, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "repeat_community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/community_ranking_frozen.json", + "sha256": "5a9c59fc4082db47bec7b59432e9ebe9f907f70baa95a9f0d3480e85f8992ff2" + } + }, + "primary_original_grid_capacity_stability": { + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 8, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "records": [ + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns128_mbt16384", + "first_capacity": 0.25, + "first_next_tested_rate": 0.35, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns128_mbt8192", + "first_capacity": 0.15, + "first_next_tested_rate": 0.25, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns64_mbt16384", + "first_capacity": 0.25, + "first_next_tested_rate": 0.35, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns64_mbt8192", + "first_capacity": 0.15, + "first_next_tested_rate": 0.25, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns128_mbt16384", + "first_capacity": 0.35, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns128_mbt8192", + "first_capacity": 0.35, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns64_mbt16384", + "first_capacity": 0.35, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns64_mbt8192", + "first_capacity": 0.35, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 8, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "stable_configs": 8, + "total_configs": 8 + }, + "real_trial_ttft_ms": { + "absolute_error_p95": 57.944677013438195, + "mae": 14.052099088554602, + "paired_successful_requests": 1216, + "repeat_minus_first_mean": 1.7543344055363712 + }, + "schema": "community-qwen235b-trial-stability-v1", + "shared_config_load_points": 19, + "shared_requests": 1216, + "variants": { + "legacy_step_1s_2s": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.15, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 55, + "frontier_feasible": false, + "frontier_passed_request_count": 52, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 55 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.15, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 55, + "frontier_feasible": false, + "frontier_passed_request_count": 52, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 55 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 55, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.75, + "repeat_feasible": false, + "repeat_passed_request_count": 55 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + } + ], + "frontier_correct_first": 1177, + "frontier_correct_repeat": 1178, + "frontier_load_label_accuracy_first": 1.0, + "frontier_load_label_accuracy_repeat": 1.0, + "frontier_request_label_accuracy_first": 0.9679276315789473, + "frontier_request_label_accuracy_repeat": 0.96875, + "frontier_wrong_both_stable_real_label": 38, + "matching_real_trial_labels": 1215, + "real_trial_failure_set_jaccard": 0.9925373134328358, + "real_trial_label_flips": 1, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 0.9991776315789473, + "shared_config_load_points": 19, + "shared_requests": 1216 + }, + "linear_10k": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.15, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.15, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.75, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 1173, + "frontier_correct_repeat": 1174, + "frontier_load_label_accuracy_first": 1.0, + "frontier_load_label_accuracy_repeat": 1.0, + "frontier_request_label_accuracy_first": 0.9646381578947368, + "frontier_request_label_accuracy_repeat": 0.9654605263157895, + "frontier_wrong_both_stable_real_label": 42, + "matching_real_trial_labels": 1215, + "real_trial_failure_set_jaccard": 0.9891304347826086, + "real_trial_label_flips": 1, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 0.9991776315789473, + "shared_config_load_points": 19, + "shared_requests": 1216 + }, + "linear_6k": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 63, + "frontier_feasible": true, + "frontier_passed_request_count": 63, + "offered_request_rate": 0.15, + "repeat_feasible": true, + "repeat_passed_request_count": 63 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 63, + "frontier_feasible": true, + "frontier_passed_request_count": 63, + "offered_request_rate": 0.15, + "repeat_feasible": true, + "repeat_passed_request_count": 63 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.75, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 1206, + "frontier_correct_repeat": 1208, + "frontier_load_label_accuracy_first": 0.9473684210526315, + "frontier_load_label_accuracy_repeat": 1.0, + "frontier_request_label_accuracy_first": 0.9917763157894737, + "frontier_request_label_accuracy_repeat": 0.993421052631579, + "frontier_wrong_both_stable_real_label": 8, + "matching_real_trial_labels": 1214, + "real_trial_failure_set_jaccard": 0.9655172413793104, + "real_trial_label_flips": 2, + "real_trial_load_label_accuracy": 0.9473684210526315, + "real_trial_request_label_accuracy": 0.9983552631578947, + "shared_config_load_points": 19, + "shared_requests": 1216 + }, + "linear_8k_primary": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.15, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.15, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.75, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.35, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 1192, + "frontier_correct_repeat": 1193, + "frontier_load_label_accuracy_first": 1.0, + "frontier_load_label_accuracy_repeat": 1.0, + "frontier_request_label_accuracy_first": 0.9802631578947368, + "frontier_request_label_accuracy_repeat": 0.9810855263157895, + "frontier_wrong_both_stable_real_label": 23, + "matching_real_trial_labels": 1215, + "real_trial_failure_set_jaccard": 0.9852941176470589, + "real_trial_label_flips": 1, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 0.9991776315789473, + "shared_config_load_points": 19, + "shared_requests": 1216 + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json new file mode 100644 index 0000000..9574186 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json @@ -0,0 +1,5248 @@ +{ + "community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/community_ranking_frozen.json", + "sha256": "5a9c59fc4082db47bec7b59432e9ebe9f907f70baa95a9f0d3480e85f8992ff2" + }, + "created_unix_s": 1784150358.538842, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "pass_rate_surface_residuals": { + "legacy_step_1s_2s": { + "false_feasible_points": 0, + "false_infeasible_points": 0, + "feasibility_label_accuracy": 1.0, + "mae": 0.028409090909090908, + "matched_config_load_points": 33, + "max_absolute_error": 0.046875, + "signed_error_mean": -0.026515151515151516 + }, + "linear_10k": { + "false_feasible_points": 0, + "false_infeasible_points": 8, + "feasibility_label_accuracy": 0.7575757575757576, + "mae": 0.022727272727272728, + "matched_config_load_points": 33, + "max_absolute_error": 0.046875, + "signed_error_mean": -0.020833333333333332 + }, + "linear_6k": { + "false_feasible_points": 0, + "false_infeasible_points": 2, + "feasibility_label_accuracy": 0.9393939393939394, + "mae": 0.006628787878787879, + "matched_config_load_points": 33, + "max_absolute_error": 0.015625, + "signed_error_mean": -0.004734848484848485 + }, + "linear_8k_primary": { + "false_feasible_points": 0, + "false_infeasible_points": 6, + "feasibility_label_accuracy": 0.8181818181818182, + "mae": 0.01278409090909091, + "matched_config_load_points": 33, + "max_absolute_error": 0.046875, + "signed_error_mean": -0.01278409090909091 + } + }, + "primary_slo": "linear_8k_primary", + "request_level_ttft_residuals": { + "by_input_bin": { + "[0,1024)": { + "count": 891, + "mae_ms": 40.73985496630227, + "signed_error_ms_mean": -3.610411376010397 + }, + "[1024,2048)": { + "count": 693, + "mae_ms": 57.62684905459171, + "signed_error_ms_mean": 44.446875608364664 + }, + "[16384,32769)": { + "count": 132, + "mae_ms": 240.89426466203582, + "signed_error_ms_mean": 147.35987190047473 + }, + "[2048,4096)": { + "count": 132, + "mae_ms": 109.72593289069287, + "signed_error_ms_mean": 91.2075693704322 + }, + "[4096,8192)": { + "count": 99, + "mae_ms": 140.80238805214807, + "signed_error_ms_mean": 138.7995225585974 + }, + "[8192,16384)": { + "count": 165, + "mae_ms": 237.3733609970516, + "signed_error_ms_mean": 237.3733609970516 + } + }, + "mae_ms": 83.15460427483131, + "mape": 0.1633638442645967, + "matched_successful_requests": 2112, + "signed_error_ms_mean": 53.022475287000866 + }, + "request_slo_classification": { + "legacy_step_1s_2s": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "2505" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "1818", + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1799", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "1818", + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 225, + "failure_set_jaccard": 0.7785467128027682, + "failure_set_union": 289, + "false_slo_fail_requests": 60, + "false_slo_pass_requests": 4, + "frontier_failure_requests": 285, + "label_accuracy": 0.9696969696969697, + "matched_requests": 2112, + "matching_labels": 2048, + "real_failure_requests": 229 + }, + "linear_10k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7271", + "7403", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7271", + "7403", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "6604", + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14316", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14316", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 129, + "failure_set_jaccard": 0.6172248803827751, + "failure_set_union": 209, + "false_slo_fail_requests": 62, + "false_slo_pass_requests": 18, + "frontier_failure_requests": 191, + "label_accuracy": 0.9621212121212122, + "matched_requests": 2112, + "matching_labels": 2032, + "real_failure_requests": 147 + }, + "linear_6k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 91, + "failure_set_jaccard": 0.8666666666666667, + "failure_set_union": 105, + "false_slo_fail_requests": 12, + "false_slo_pass_requests": 2, + "frontier_failure_requests": 103, + "label_accuracy": 0.9933712121212122, + "matched_requests": 2112, + "matching_labels": 2098, + "real_failure_requests": 93 + }, + "linear_8k_primary": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "2396" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "2396" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.15 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7271" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.75 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.35 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.4 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 107, + "failure_set_jaccard": 0.7535211267605634, + "failure_set_union": 142, + "false_slo_fail_requests": 31, + "false_slo_pass_requests": 4, + "frontier_failure_requests": 138, + "label_accuracy": 0.9834280303030303, + "matched_requests": 2112, + "matching_labels": 2077, + "real_failure_requests": 111 + } + }, + "schema": "frontier-community-qwen235b-rank-comparison-v1", + "variants": { + "legacy_step_1s_2s": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.9375, + 0.921875, + 0.921875, + 0.890625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.9375, + 0.921875, + 0.921875, + 0.890625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.90625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.90625, + 0.890625, + 0.890625, + 0.859375 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.90625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.90625, + 0.890625, + 0.890625, + 0.859375 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.875, + 0.875, + 0.859375 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5, + 0.75 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.875, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "frontier_capacity": { + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + } + } + }, + "linear_10k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.90625, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.90625, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.9375, + 0.875, + 0.859375, + 0.84375, + 0.84375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.9375, + 0.875, + 0.859375, + 0.84375, + 0.84375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.828125, + 0.828125, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.828125, + 0.828125, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0375, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.0375, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5, + 0.75 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.90625, + 0.890625, + 0.875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.921875, + 0.90625, + 0.890625, + 0.875 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "top1_set_match": true + } + } + }, + "linear_6k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.90625, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.90625, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 7, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 7, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.075, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.075, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 16, + "correct_pairs": 16, + "frontier_tied_pairs": 8, + "real_tied_pairs": 12 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.075, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.075, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.953125, + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5, + 0.75 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 16, + "correct_pairs": 16, + "frontier_tied_pairs": 8, + "real_tied_pairs": 12 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9428090415820635, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "top1_set_match": true + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 2, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "top1_set_match": false + } + } + }, + "linear_8k_primary": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.921875, + 0.890625, + 0.890625, + 0.875, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.921875, + 0.890625, + 0.890625, + 0.875, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.9375, + 0.90625, + 0.859375, + 0.859375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.9375, + 0.90625, + 0.859375, + 0.859375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 4, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.05, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.05, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 0.9486832980505138, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.921875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.921875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5, + 0.75 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.953125, + 0.9375 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.96875, + 0.953125, + 0.9375 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.2, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.9375, + 0.921875, + 0.90625 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.2, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125, + 0.9375, + 0.921875, + 0.90625 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 20, + "correct_pairs": 20, + "frontier_tied_pairs": 4, + "real_tied_pairs": 8 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9486832980505138, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 4, + "correct_pairs": 4, + "frontier_tied_pairs": 2, + "real_tied_pairs": 2 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 2, + "real_tied_pairs": 6 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "top1_set_match": false + } + } + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_run_manifest.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_run_manifest.json new file mode 100644 index 0000000..189ada2 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_run_manifest.json @@ -0,0 +1,583 @@ +{ + "configs": [ + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 128, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp8_mns128_mbt16384/engine.log", + "execution_index": 1, + "rate_order": [ + 0.75, + 0.5, + 0.35, + 0.45, + 0.4 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp8_mns128_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": true, + "mbt": 16384, + "mns": 64, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp8_mns64_mbt16384/engine.log", + "execution_index": 2, + "rate_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp8_mns64_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp4_mns64_mbt16384/engine.log", + "execution_index": 3, + "rate_order": [ + 0.35, + 0.25, + 0.3 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp4_mns64_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp8_mns64_mbt8192/engine.log", + "execution_index": 4, + "rate_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp8_mns64_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp8_mns128_mbt8192/engine.log", + "execution_index": 5, + "rate_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp8_mns128_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp4_mns128_mbt8192/engine.log", + "execution_index": 6, + "rate_order": [ + 0.15, + 0.35, + 0.25, + 0.3, + 0.2 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp4_mns128_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp4_mns64_mbt8192/engine.log", + "execution_index": 7, + "rate_order": [ + 0.25, + 0.35, + 0.15, + 0.3, + 0.2 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp4_mns64_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/runs/tp4_mns128_mbt16384/engine.log", + "execution_index": 8, + "rate_order": [ + 0.35, + 0.25, + 0.3 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/results/tp4_mns128_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + } + ], + "created_unix_s": 1784142580.9845052, + "execution_contract": { + "config_order": "reverse_of_first_pass", + "estimated_replay_wall_seconds_excluding_model_loads": 6639.746031746032, + "fallback_when_no_transition": "top_two_if_all_feasible_else_bottom_two", + "one_fresh_server_launch_per_config": true, + "rate_order": "reverse_first_pass_boundary_order_then_descending_refinement_rates", + "schema": "community-qwen235b-boundary-repeat-selection-v4", + "selection": "all_adjacent_pre_registered_slo_feasibility_transitions_plus_between_rate_refinement", + "selections": [ + { + "added_refinement_rates": [ + 0.4, + 0.45 + ], + "boundary_rates": [ + 0.35, + 0.5, + 0.75 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [ + 0.35, + 0.5 + ], + "linear_6k": [ + 0.5, + 0.75 + ], + "linear_8k_primary": [ + 0.35, + 0.5 + ] + }, + "config": "tp8_mns128_mbt16384", + "excluded_unrankable_slos": [ + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.75, + 0.5, + 0.35, + 0.45, + 0.4 + ], + "selected_rates": [ + 0.35, + 0.4, + 0.45, + 0.5, + 0.75 + ], + "source_relative_order": [ + 0.35, + 0.5, + 0.75 + ] + }, + { + "added_refinement_rates": [ + 0.4, + 0.45 + ], + "boundary_rates": [ + 0.35, + 0.5 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [ + 0.35, + 0.5 + ], + "linear_6k": [ + 0.35, + 0.5 + ], + "linear_8k_primary": [ + 0.35, + 0.5 + ] + }, + "config": "tp8_mns64_mbt16384", + "excluded_unrankable_slos": [ + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "selected_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "source_relative_order": [ + 0.35, + 0.5 + ] + }, + { + "added_refinement_rates": [ + 0.3 + ], + "boundary_rates": [ + 0.25, + 0.35 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [], + "linear_6k": [ + 0.25, + 0.35 + ], + "linear_8k_primary": [ + 0.25, + 0.35 + ] + }, + "config": "tp4_mns64_mbt16384", + "excluded_unrankable_slos": [ + "linear_10k", + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.35, + 0.25, + 0.3 + ], + "selected_rates": [ + 0.25, + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.25, + 0.35 + ] + }, + { + "added_refinement_rates": [ + 0.4, + 0.45 + ], + "boundary_rates": [ + 0.35, + 0.5 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [ + 0.35, + 0.5 + ], + "linear_6k": [ + 0.35, + 0.5 + ], + "linear_8k_primary": [ + 0.35, + 0.5 + ] + }, + "config": "tp8_mns64_mbt8192", + "excluded_unrankable_slos": [ + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "selected_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "source_relative_order": [ + 0.35, + 0.5 + ] + }, + { + "added_refinement_rates": [ + 0.4, + 0.45 + ], + "boundary_rates": [ + 0.35, + 0.5 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [ + 0.35, + 0.5 + ], + "linear_6k": [ + 0.35, + 0.5 + ], + "linear_8k_primary": [ + 0.35, + 0.5 + ] + }, + "config": "tp8_mns128_mbt8192", + "excluded_unrankable_slos": [ + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.5, + 0.35, + 0.45, + 0.4 + ], + "selected_rates": [ + 0.35, + 0.4, + 0.45, + 0.5 + ], + "source_relative_order": [ + 0.35, + 0.5 + ] + }, + { + "added_refinement_rates": [ + 0.2, + 0.3 + ], + "boundary_rates": [ + 0.15, + 0.25, + 0.35 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [], + "linear_6k": [ + 0.25, + 0.35 + ], + "linear_8k_primary": [ + 0.15, + 0.25 + ] + }, + "config": "tp4_mns128_mbt8192", + "excluded_unrankable_slos": [ + "linear_10k", + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.15, + 0.35, + 0.25, + 0.3, + 0.2 + ], + "selected_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.25, + 0.35, + 0.15 + ] + }, + { + "added_refinement_rates": [ + 0.2, + 0.3 + ], + "boundary_rates": [ + 0.15, + 0.25, + 0.35 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [], + "linear_6k": [ + 0.25, + 0.35 + ], + "linear_8k_primary": [ + 0.15, + 0.25 + ] + }, + "config": "tp4_mns64_mbt8192", + "excluded_unrankable_slos": [ + "linear_10k", + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.25, + 0.35, + 0.15, + 0.3, + 0.2 + ], + "selected_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.15, + 0.35, + 0.25 + ] + }, + { + "added_refinement_rates": [ + 0.3 + ], + "boundary_rates": [ + 0.25, + 0.35 + ], + "boundary_rates_by_slo": { + "legacy_step_1s_2s": [], + "linear_10k": [], + "linear_6k": [ + 0.25, + 0.35 + ], + "linear_8k_primary": [ + 0.25, + 0.35 + ] + }, + "config": "tp4_mns128_mbt16384", + "excluded_unrankable_slos": [ + "linear_10k", + "legacy_step_1s_2s" + ], + "repeat_order": [ + 0.35, + 0.25, + 0.3 + ], + "selected_rates": [ + 0.25, + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.25, + 0.35 + ] + } + ], + "unrankable_slo_policy": "exclude_per_config_surfaces_with_no_feasible_load", + "warmup": { + "latencies_discarded": true, + "method": "one_disjoint_request_per_input_length_bin", + "request_count": 6, + "seed": 2026071504 + } + }, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "model": { + "config_sha256": "702c46d431bb984db9035a1225186bbfdb52c0d19c82104df4a37cd005e0369e", + "path": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8" + }, + "protocol_manifest": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/protocol_manifest.json", + "sha256": "695c0500b9a5d51cdd4e39b722e3c9db15f8bed48795454f4d0d6442d24929cb" + }, + "repeat_of": { + "first_pass_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/community_ranking_frozen.json", + "sha256": "055fe2cf1e0c01a0503145a272097b8ceda257f1b86bdae39abf17277d9a18d9" + }, + "source_manifest": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/run_manifest.json", + "sha256": "6da81a34140d801cb6055e089bed7a12adb14c09dc87ca6dfa7f79df4ac34920" + } + }, + "repository": { + "commit": "501ceb817122747a23dc6d9fdc9d71dcb3aa8e72", + "status_porcelain": [] + }, + "runtime": { + "python": "/usr/bin/python3.12", + "vllm": "/tmp/wjh-frontier-vllm0102-smoke/.venv/bin/vllm" + }, + "schema": "community-qwen235b-prefill-fixed-cohort-v1", + "studies": { + "4": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json", + "sha256": "ea0d7d31e4a5102cf530e900754debb1d8b18f860748ead88ebf7de57d92faa3" + }, + "8": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json", + "sha256": "48c4d917b5531f742a05bab3bcd5c880f2eaddf9992b7629e9f48d8e00fd1906" + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_v3_trial_stability.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_v3_trial_stability.json new file mode 100644 index 0000000..94ea5de --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_v3_trial_stability.json @@ -0,0 +1,732 @@ +{ + "created_unix_s": 1784153851.3812342, + "inputs": { + "first_community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/community_ranking_frozen.json", + "sha256": "5a9c59fc4082db47bec7b59432e9ebe9f907f70baa95a9f0d3480e85f8992ff2" + }, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "repeat_community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/community_ranking_frozen.json", + "sha256": "4edd25edde3d7508d8feb9b024b5bca4ba6a5f25335ac84008f09c524ed7bc34" + } + }, + "primary_original_grid_capacity_stability": { + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.05, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.05, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 12, + "correct_pairs": 12, + "frontier_tied_pairs": 3, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.05, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.05, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "records": [ + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns128_mbt16384", + "first_capacity": 0.3, + "first_next_tested_rate": 0.35, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns128_mbt8192", + "first_capacity": 0.2, + "first_next_tested_rate": 0.25, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns64_mbt16384", + "first_capacity": 0.3, + "first_next_tested_rate": 0.35, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp4_mns64_mbt8192", + "first_capacity": 0.2, + "first_next_tested_rate": 0.25, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns128_mbt8192", + "first_capacity": 0.45, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + }, + { + "boundary_repeated": true, + "capacity_still_feasible": true, + "config": "tp8_mns64_mbt8192", + "first_capacity": 0.45, + "first_next_tested_rate": 0.5, + "next_rate_still_infeasible": true, + "stable_original_grid_capacity": true + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 12, + "correct_pairs": 12, + "frontier_tied_pairs": 3, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.9933992677987828, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "top1_set_match": false + }, + "stable_configs": 6, + "total_configs": 6 + }, + "real_trial_ttft_ms": { + "absolute_error_p95": 62.524845998268574, + "mae": 14.980625974885697, + "paired_successful_requests": 768, + "repeat_minus_first_mean": 1.5463623839574818 + }, + "schema": "community-qwen235b-trial-stability-v1", + "shared_config_load_points": 12, + "shared_requests": 768, + "variants": { + "legacy_step_1s_2s": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.3, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.2, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.3, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.2, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.45, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 57, + "frontier_feasible": false, + "frontier_passed_request_count": 54, + "offered_request_rate": 0.45, + "repeat_feasible": false, + "repeat_passed_request_count": 57 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 53, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + } + ], + "frontier_correct_first": 742, + "frontier_correct_repeat": 742, + "frontier_load_label_accuracy_first": 1.0, + "frontier_load_label_accuracy_repeat": 1.0, + "frontier_request_label_accuracy_first": 0.9661458333333334, + "frontier_request_label_accuracy_repeat": 0.9661458333333334, + "frontier_wrong_both_stable_real_label": 26, + "matching_real_trial_labels": 768, + "real_trial_failure_set_jaccard": 1.0, + "real_trial_label_flips": 0, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 1.0, + "shared_config_load_points": 12, + "shared_requests": 768 + }, + "linear_10k": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.3, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.2, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 56, + "offered_request_rate": 0.3, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 56, + "frontier_feasible": false, + "frontier_passed_request_count": 55, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 56 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.2, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 58, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 58 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 58, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 726, + "frontier_correct_repeat": 726, + "frontier_load_label_accuracy_first": 0.8333333333333334, + "frontier_load_label_accuracy_repeat": 0.8333333333333334, + "frontier_request_label_accuracy_first": 0.9453125, + "frontier_request_label_accuracy_repeat": 0.9453125, + "frontier_wrong_both_stable_real_label": 42, + "matching_real_trial_labels": 768, + "real_trial_failure_set_jaccard": 1.0, + "real_trial_label_flips": 0, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 1.0, + "shared_config_load_points": 12, + "shared_requests": 768 + }, + "linear_6k": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.3, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.2, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 61, + "offered_request_rate": 0.3, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 62, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.2, + "repeat_feasible": true, + "repeat_passed_request_count": 62 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": true, + "frontier_passed_request_count": 62, + "offered_request_rate": 0.25, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 760, + "frontier_correct_repeat": 760, + "frontier_load_label_accuracy_first": 0.8333333333333334, + "frontier_load_label_accuracy_repeat": 0.8333333333333334, + "frontier_request_label_accuracy_first": 0.9895833333333334, + "frontier_request_label_accuracy_repeat": 0.9895833333333334, + "frontier_wrong_both_stable_real_label": 8, + "matching_real_trial_labels": 768, + "real_trial_failure_set_jaccard": 1.0, + "real_trial_label_flips": 0, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 1.0, + "shared_config_load_points": 12, + "shared_requests": 768 + }, + "linear_8k_primary": { + "config_load_records": [ + { + "config": "tp4_mns128_mbt16384", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.3, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns128_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.2, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.3, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns64_mbt16384", + "first_feasible": false, + "first_passed_request_count": 59, + "frontier_feasible": false, + "frontier_passed_request_count": 57, + "offered_request_rate": 0.35, + "repeat_feasible": false, + "repeat_passed_request_count": 59 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.2, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp4_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.25, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns128_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": true, + "first_passed_request_count": 61, + "frontier_feasible": false, + "frontier_passed_request_count": 60, + "offered_request_rate": 0.45, + "repeat_feasible": true, + "repeat_passed_request_count": 61 + }, + { + "config": "tp8_mns64_mbt8192", + "first_feasible": false, + "first_passed_request_count": 60, + "frontier_feasible": false, + "frontier_passed_request_count": 59, + "offered_request_rate": 0.5, + "repeat_feasible": false, + "repeat_passed_request_count": 60 + } + ], + "frontier_correct_first": 750, + "frontier_correct_repeat": 750, + "frontier_load_label_accuracy_first": 0.5, + "frontier_load_label_accuracy_repeat": 0.5, + "frontier_request_label_accuracy_first": 0.9765625, + "frontier_request_label_accuracy_repeat": 0.9765625, + "frontier_wrong_both_stable_real_label": 18, + "matching_real_trial_labels": 768, + "real_trial_failure_set_jaccard": 1.0, + "real_trial_label_flips": 0, + "real_trial_load_label_accuracy": 1.0, + "real_trial_request_label_accuracy": 1.0, + "shared_config_load_points": 12, + "shared_requests": 768 + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_run_manifest.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_run_manifest.json new file mode 100644 index 0000000..99aab5f --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_run_manifest.json @@ -0,0 +1,345 @@ +{ + "configs": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp4_mns128_mbt16384/engine.log", + "execution_index": 1, + "rate_order": [ + 0.3, + 0.35 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp4_mns128_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp4_mns64_mbt8192/engine.log", + "execution_index": 2, + "rate_order": [ + 0.2, + 0.25 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp4_mns64_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp4_mns128_mbt8192/engine.log", + "execution_index": 3, + "rate_order": [ + 0.2, + 0.25 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp4_mns128_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp8_mns128_mbt8192/engine.log", + "execution_index": 4, + "rate_order": [ + 0.45, + 0.5 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp8_mns128_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp8_mns64_mbt8192/engine.log", + "execution_index": 5, + "rate_order": [ + 0.45, + 0.5 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp8_mns64_mbt8192.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json" + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "engine_log_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/runs/tp4_mns64_mbt16384/engine.log", + "execution_index": 6, + "rate_order": [ + 0.3, + 0.35 + ], + "result_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/results/tp4_mns64_mbt16384.json", + "study_path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json" + } + ], + "created_unix_s": 1784150504.0091515, + "execution_contract": { + "config_order": "reverse_of_first_pass", + "estimated_replay_wall_seconds_excluding_model_loads": 2484.8253968253966, + "fallback_when_no_transition": "top_two_if_all_feasible_else_bottom_two", + "one_fresh_server_launch_per_config": true, + "rate_order": "reverse_first_pass_boundary_order_then_descending_refinement_rates", + "schema": "community-qwen235b-boundary-repeat-selection-v4", + "selected_configs": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "selected_slos": [ + "linear_8k_primary" + ], + "selection": "all_adjacent_pre_registered_slo_feasibility_transitions_plus_between_rate_refinement", + "selections": [ + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.3, + 0.35 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.3, + 0.35 + ] + }, + "config": "tp4_mns128_mbt16384", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.3, + 0.35 + ], + "selected_rates": [ + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.35, + 0.3 + ] + }, + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.2, + 0.25 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.2, + 0.25 + ] + }, + "config": "tp4_mns64_mbt8192", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.2, + 0.25 + ], + "selected_rates": [ + 0.2, + 0.25 + ], + "source_relative_order": [ + 0.25, + 0.2 + ] + }, + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.2, + 0.25 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.2, + 0.25 + ] + }, + "config": "tp4_mns128_mbt8192", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.2, + 0.25 + ], + "selected_rates": [ + 0.2, + 0.25 + ], + "source_relative_order": [ + 0.25, + 0.2 + ] + }, + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.45, + 0.5 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.45, + 0.5 + ] + }, + "config": "tp8_mns128_mbt8192", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.45, + 0.5 + ], + "selected_rates": [ + 0.45, + 0.5 + ], + "source_relative_order": [ + 0.5, + 0.45 + ] + }, + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.45, + 0.5 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.45, + 0.5 + ] + }, + "config": "tp8_mns64_mbt8192", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.45, + 0.5 + ], + "selected_rates": [ + 0.45, + 0.5 + ], + "source_relative_order": [ + 0.5, + 0.45 + ] + }, + { + "added_refinement_rates": [], + "boundary_rates": [ + 0.3, + 0.35 + ], + "boundary_rates_by_slo": { + "linear_8k_primary": [ + 0.3, + 0.35 + ] + }, + "config": "tp4_mns64_mbt16384", + "excluded_unrankable_slos": [], + "repeat_order": [ + 0.3, + 0.35 + ], + "selected_rates": [ + 0.3, + 0.35 + ], + "source_relative_order": [ + 0.35, + 0.3 + ] + } + ], + "unrankable_slo_policy": "exclude_per_config_surfaces_with_no_feasible_load", + "warmup": { + "latencies_discarded": true, + "method": "one_disjoint_request_per_input_length_bin", + "request_count": 6, + "seed": 2026071504 + } + }, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "model": { + "config_sha256": "702c46d431bb984db9035a1225186bbfdb52c0d19c82104df4a37cd005e0369e", + "path": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-235B-A22B-FP8" + }, + "protocol_manifest": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/protocol/protocol_manifest.json", + "sha256": "695c0500b9a5d51cdd4e39b722e3c9db15f8bed48795454f4d0d6442d24929cb" + }, + "repeat_of": { + "first_pass_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/community_ranking_frozen.json", + "sha256": "5a9c59fc4082db47bec7b59432e9ebe9f907f70baa95a9f0d3480e85f8992ff2" + }, + "source_manifest": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/run_manifest.json", + "sha256": "cf1a722043f42ff8f435e3521d0b2b2892a8ea0922e1e4870ec10fba26c0f998" + } + }, + "repository": { + "commit": "344af3a4285ccfdc6e10e540767808907ad81ddf", + "status_porcelain": [] + }, + "runtime": { + "python": "/usr/bin/python3.12", + "vllm": "/tmp/wjh-frontier-vllm0102-smoke/.venv/bin/vllm" + }, + "schema": "community-qwen235b-prefill-fixed-cohort-v1", + "studies": { + "4": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp4/study.json", + "sha256": "ea0d7d31e4a5102cf530e900754debb1d8b18f860748ead88ebf7de57d92faa3" + }, + "8": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/studies/tp8/study.json", + "sha256": "48c4d917b5531f742a05bab3bcd5c880f2eaddf9992b7629e9f48d8e00fd1906" + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_targeted_comparison.json b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_targeted_comparison.json new file mode 100644 index 0000000..485ca5e --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v3_targeted_comparison.json @@ -0,0 +1,3830 @@ +{ + "community_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/community_ranking_frozen.json", + "sha256": "4edd25edde3d7508d8feb9b024b5bca4ba6a5f25335ac84008f09c524ed7bc34" + }, + "created_unix_s": 1784153851.1556726, + "frontier_freeze": { + "path": "/home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/refinement-v1/frontier-v4-refined-rates-batched-par4/frontier_ranking_frozen.json", + "sha256": "318b3c7742c8371191e2dbd429f0e0e63c1a42f4806a599e018b0548cb9d6336" + }, + "pass_rate_surface_residuals": { + "legacy_step_1s_2s": { + "false_feasible_points": 0, + "false_infeasible_points": 0, + "feasibility_label_accuracy": 1.0, + "mae": 0.028645833333333332, + "matched_config_load_points": 12, + "max_absolute_error": 0.046875, + "signed_error_mean": -0.028645833333333332 + }, + "linear_10k": { + "false_feasible_points": 0, + "false_infeasible_points": 2, + "feasibility_label_accuracy": 0.8333333333333334, + "mae": 0.028645833333333332, + "matched_config_load_points": 12, + "max_absolute_error": 0.046875, + "signed_error_mean": -0.028645833333333332 + }, + "linear_6k": { + "false_feasible_points": 0, + "false_infeasible_points": 2, + "feasibility_label_accuracy": 0.8333333333333334, + "mae": 0.010416666666666666, + "matched_config_load_points": 12, + "max_absolute_error": 0.015625, + "signed_error_mean": -0.005208333333333333 + }, + "linear_8k_primary": { + "false_feasible_points": 0, + "false_infeasible_points": 6, + "feasibility_label_accuracy": 0.5, + "mae": 0.018229166666666668, + "matched_config_load_points": 12, + "max_absolute_error": 0.03125, + "signed_error_mean": -0.018229166666666668 + } + }, + "primary_slo": "linear_8k_primary", + "request_level_ttft_residuals": { + "by_input_bin": { + "[0,1024)": { + "count": 324, + "mae_ms": 50.51891975082174, + "signed_error_ms_mean": -9.882795155125912 + }, + "[1024,2048)": { + "count": 252, + "mae_ms": 59.564509081632316, + "signed_error_ms_mean": 41.897250452193944 + }, + "[16384,32769)": { + "count": 48, + "mae_ms": 280.18396195154355, + "signed_error_ms_mean": 186.5034642541033 + }, + "[2048,4096)": { + "count": 48, + "mae_ms": 94.21679883383375, + "signed_error_ms_mean": 67.31957197829648 + }, + "[4096,8192)": { + "count": 36, + "mae_ms": 110.9597465520623, + "signed_error_ms_mean": 108.56515721844958 + }, + "[8192,16384)": { + "count": 60, + "mae_ms": 241.64142363249, + "signed_error_ms_mean": 241.64142363249 + } + }, + "mae_ms": 88.33679570229081, + "mape": 0.16159539870054299, + "matched_successful_requests": 768, + "signed_error_ms_mean": 49.40939882898548 + }, + "request_slo_classification": { + "legacy_step_1s_2s": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6611" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316", + "14359", + "1818" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359", + "1818", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 82, + "failure_set_jaccard": 0.7592592592592593, + "failure_set_union": 108, + "false_slo_fail_requests": 24, + "false_slo_pass_requests": 2, + "frontier_failure_requests": 106, + "label_accuracy": 0.9661458333333334, + "matched_requests": 768, + "matching_labels": 742, + "real_failure_requests": 84 + }, + "linear_10k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7271", + "7403", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7271", + "7403", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.921875, + "matched_requests": 64, + "matching_labels": 59, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "6611", + "7422" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "8891" + ], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "6604", + "6611" + ], + "false_slo_pass_request_ids": [ + "7403", + "8891" + ], + "label_accuracy": 0.9375, + "matched_requests": 64, + "matching_labels": 60, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14316", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14316", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026", + "14359", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.953125, + "matched_requests": 64, + "matching_labels": 61, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 52, + "failure_set_jaccard": 0.5531914893617021, + "failure_set_union": 94, + "false_slo_fail_requests": 32, + "false_slo_pass_requests": 10, + "frontier_failure_requests": 84, + "label_accuracy": 0.9453125, + "matched_requests": 768, + "matching_labels": 726, + "real_failure_requests": 62 + }, + "linear_6k": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [], + "label_accuracy": 1.0, + "matched_requests": 64, + "matching_labels": 64, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 34, + "failure_set_jaccard": 0.8095238095238095, + "failure_set_union": 42, + "false_slo_fail_requests": 6, + "false_slo_pass_requests": 2, + "frontier_failure_requests": 40, + "label_accuracy": 0.9895833333333334, + "matched_requests": 768, + "matching_labels": 760, + "real_failure_requests": 36 + }, + "linear_8k_primary": { + "by_config_load": [ + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns128_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7403" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.3 + }, + { + "config": "tp4_mns64_mbt16384", + "false_slo_fail_request_ids": [ + "7271", + "7422" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.35 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.2 + }, + { + "config": "tp4_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14026" + ], + "false_slo_pass_request_ids": [ + "7403" + ], + "label_accuracy": 0.96875, + "matched_requests": 64, + "matching_labels": 62, + "offered_request_rate": 0.25 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns128_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14316" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.45 + }, + { + "config": "tp8_mns64_mbt8192", + "false_slo_fail_request_ids": [ + "14359" + ], + "false_slo_pass_request_ids": [], + "label_accuracy": 0.984375, + "matched_requests": 64, + "matching_labels": 63, + "offered_request_rate": 0.5 + } + ], + "failure_set_intersection": 42, + "failure_set_jaccard": 0.7, + "failure_set_union": 60, + "false_slo_fail_requests": 16, + "false_slo_pass_requests": 2, + "frontier_failure_requests": 58, + "label_accuracy": 0.9765625, + "matched_requests": 768, + "matching_labels": 750, + "real_failure_requests": 44 + } + }, + "schema": "frontier-community-qwen235b-rank-comparison-v1", + "variants": { + "legacy_step_1s_2s": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.6875, + 0.515625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.890625, + 0.890625, + 0.84375, + 0.8125, + 0.8125, + 0.8125, + 0.8125, + 0.734375, + 0.671875, + 0.5 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.9375, + 0.921875, + 0.921875, + 0.890625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.9375, + 0.921875, + 0.921875, + 0.890625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.765625, + 0.6875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.921875, + 0.90625, + 0.90625, + 0.875, + 0.859375, + 0.84375, + 0.828125, + 0.828125, + 0.734375, + 0.578125 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.890625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.90625, + 0.890625 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.890625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.890625, + 0.875 + ], + "rank": 1, + "slo": "legacy_step_1s_2s", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "frontier_capacity": { + "tp8_mns128_mbt16384": null, + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt16384": null, + "tp8_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp8_mns128_mbt8192": null, + "tp8_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + } + } + }, + "linear_10k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.90625, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.9375, + 0.9375, + 0.921875, + 0.890625, + 0.84375, + 0.75 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.35, + "maximum_tested_feasible_request_rate_per_gpu": 0.04375, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.90625, + 0.890625, + 0.890625, + 0.796875, + 0.6875 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.9375, + 0.875, + 0.859375, + 0.84375, + 0.84375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375, + 0.9375, + 0.875, + 0.859375, + 0.84375, + 0.84375, + 0.84375, + 0.796875, + 0.734375, + 0.5625 + ], + "rank": 5, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.828125, + 0.828125, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.828125, + 0.828125, + 0.828125, + 0.828125, + 0.765625, + 0.703125, + 0.53125 + ], + "rank": 7, + "slo": "linear_10k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0375, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.0375, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.04375, + "tp8_mns128_mbt8192": 0.04375, + "tp8_mns64_mbt16384": 0.04375, + "tp8_mns64_mbt8192": 0.04375 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 1, + "real_tied_pairs": 1 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.917662935482247, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "top1_set_match": false + }, + "real_ranking": [ + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 1, + "slo": "linear_10k", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.875 + ], + "rank": 3, + "slo": "linear_10k", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625 + ], + "rank": 3, + "slo": "linear_10k", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.875 + ], + "rank": 3, + "slo": "linear_10k", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + false, + false + ], + "lower_censored": true, + "maximum_tested_feasible_request_rate": null, + "maximum_tested_feasible_request_rate_per_gpu": null, + "monotonicity_violations": [], + "pass_rates": [ + 0.921875, + 0.90625 + ], + "rank": 3, + "slo": "linear_10k", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 1, + "real_tied_pairs": 1 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.917662935482247, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.15, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": 0.15, + "tp4_mns64_mbt8192": null + }, + "rankable": false, + "real_capacity": { + "tp4_mns128_mbt16384": null, + "tp4_mns128_mbt8192": null, + "tp4_mns64_mbt16384": null, + "tp4_mns64_mbt8192": null + }, + "reason": "one_or_both_surfaces_have_no_feasible_config" + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.35, + "tp8_mns128_mbt8192": 0.35, + "tp8_mns64_mbt16384": 0.35, + "tp8_mns64_mbt8192": 0.35 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns128_mbt8192", + "tp8_mns64_mbt16384", + "tp8_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 1, + "real_tied_pairs": 1 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "top1_set_match": false + } + } + }, + "linear_6k": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.90625, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.9375, + 0.9375, + 0.90625, + 0.90625, + 0.859375, + 0.8125, + 0.640625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + true, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.921875, + 0.90625, + 0.90625, + 0.90625, + 0.84375, + 0.796875, + 0.625 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 1.0, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.875, + 0.8125 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 7, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.859375, + 0.75 + ], + "rank": 7, + "slo": "linear_6k", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.075, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.075, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 8, + "correct_pairs": 8, + "frontier_tied_pairs": 7, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.0625, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.0625, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 0.8660254037844387, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.16666666666666663, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": false + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375 + ], + "rank": 1, + "slo": "linear_6k", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125 + ], + "rank": 3, + "slo": "linear_6k", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": true + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.953125 + ], + "rank": 3, + "slo": "linear_6k", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": true + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_6k", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 8, + "correct_pairs": 8, + "frontier_tied_pairs": 7, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.25, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 0.8660254037844387, + "top1_regret_fraction_best_tie_break": null, + "top1_regret_fraction_worst_tie_break": null, + "top1_set_intersection": [], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.3, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.3 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns128_mbt8192", + "tp4_mns64_mbt16384", + "tp4_mns64_mbt8192" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 6, + "real_tied_pairs": 2 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.25, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.25 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.16666666666666663, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": false + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 1, + "real_tied_pairs": 1 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": null, + "top1_regret_fraction_worst_tie_break": null, + "top1_set_intersection": [], + "top1_set_match": false + } + } + }, + "linear_8k_primary": { + "frontier_ranking": [ + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.921875, + 0.890625, + 0.890625, + 0.875, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + true, + true, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.25, + "maximum_tested_feasible_request_rate_per_gpu": 0.0625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.96875, + 0.96875, + 0.921875, + 0.890625, + 0.890625, + 0.875, + 0.859375, + 0.828125, + 0.765625, + 0.609375 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 16384, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt16384", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + true, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.90625, + 0.859375, + 0.8125 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 8, + "moe_tp": 1, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + true, + true, + true, + true, + true, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.984375, + 0.984375, + 0.96875, + 0.96875, + 0.953125, + 0.953125, + 0.9375, + 0.921875, + 0.90625, + 0.84375, + 0.75 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 128, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.9375, + 0.90625, + 0.859375, + 0.859375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + }, + { + "config": { + "mbt": 8192, + "mns": 64, + "moe_ep": 1, + "moe_tp": 4, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false, + false, + false, + false, + false, + false, + false, + false, + false, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.15, + "maximum_tested_feasible_request_rate_per_gpu": 0.0375, + "monotonicity_violations": [], + "pass_rates": [ + 0.96875, + 0.9375, + 0.9375, + 0.90625, + 0.859375, + 0.859375, + 0.859375, + 0.859375, + 0.796875, + 0.734375, + 0.578125 + ], + "rank": 7, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.15, + 0.2, + 0.25, + 0.3, + 0.35, + 0.4, + 0.45, + 0.5, + 0.75, + 1.0, + 1.5 + ], + "upper_censored": false + } + ], + "per_gpu_efficiency": { + "capacity_field": "maximum_tested_feasible_request_rate_per_gpu", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.0625, + "tp4_mns128_mbt8192": 0.0375, + "tp4_mns64_mbt16384": 0.0625, + "tp4_mns64_mbt8192": 0.0375, + "tp8_mns128_mbt16384": 0.05625, + "tp8_mns128_mbt8192": 0.05, + "tp8_mns64_mbt16384": 0.05625, + "tp8_mns64_mbt8192": 0.05 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 12, + "correct_pairs": 12, + "frontier_tied_pairs": 3, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.075, + "tp4_mns128_mbt8192": 0.05, + "tp4_mns64_mbt16384": 0.075, + "tp4_mns64_mbt8192": 0.05, + "tp8_mns128_mbt8192": 0.05625, + "tp8_mns64_mbt8192": 0.05625 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 0.9819805060619656, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "real_ranking": [ + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 128, + "name": "tp4_mns128_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.921875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 16384, + "mns": 64, + "name": "tp4_mns64_mbt16384", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.921875 + ], + "rank": 1, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.3, + 0.35 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 128, + "name": "tp8_mns128_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": true, + "mbt": 8192, + "mns": 64, + "name": "tp8_mns64_mbt8192", + "num_gpu_blocks": 62351, + "tp": 8 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.45, + "maximum_tested_feasible_request_rate_per_gpu": 0.05625, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 3, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.45, + 0.5 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 128, + "name": "tp4_mns128_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.2, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + }, + { + "config": { + "expert_parallel": false, + "mbt": 8192, + "mns": 64, + "name": "tp4_mns64_mbt8192", + "num_gpu_blocks": 26101, + "tp": 4 + }, + "feasibility": [ + true, + false + ], + "lower_censored": false, + "maximum_tested_feasible_request_rate": 0.2, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + "monotonicity_violations": [], + "pass_rates": [ + 0.953125, + 0.9375 + ], + "rank": 5, + "slo": "linear_8k_primary", + "tested_rates": [ + 0.2, + 0.25 + ], + "upper_censored": false + } + ], + "single_replica_raw_capacity": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15, + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 12, + "correct_pairs": 12, + "frontier_tied_pairs": 3, + "real_tied_pairs": 3 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2, + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": null, + "top1_regret_fraction_worst_tie_break": null, + "top1_set_intersection": [], + "top1_set_match": false + }, + "within_topology_raw_capacity": { + "tp4": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp4_mns128_mbt16384": 0.25, + "tp4_mns128_mbt8192": 0.15, + "tp4_mns64_mbt16384": 0.25, + "tp4_mns64_mbt8192": 0.15 + }, + "frontier_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": 1.0, + "comparable_non_tied_pairs": 4, + "correct_pairs": 4, + "frontier_tied_pairs": 2, + "real_tied_pairs": 2 + }, + "rankable": true, + "real_capacity": { + "tp4_mns128_mbt16384": 0.3, + "tp4_mns128_mbt8192": 0.2, + "tp4_mns64_mbt16384": 0.3, + "tp4_mns64_mbt8192": 0.2 + }, + "real_top1_set": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "spearman_rank_correlation": 1.0, + "top1_regret_fraction_best_tie_break": 0.0, + "top1_regret_fraction_worst_tie_break": 0.0, + "top1_set_intersection": [ + "tp4_mns128_mbt16384", + "tp4_mns64_mbt16384" + ], + "top1_set_match": true + }, + "tp8": { + "capacity_field": "maximum_tested_feasible_request_rate", + "frontier_capacity": { + "tp8_mns128_mbt16384": 0.45, + "tp8_mns128_mbt8192": 0.4, + "tp8_mns64_mbt16384": 0.45, + "tp8_mns64_mbt8192": 0.4 + }, + "frontier_top1_set": [ + "tp8_mns128_mbt16384", + "tp8_mns64_mbt16384" + ], + "pairwise_ordering": { + "accuracy": null, + "comparable_non_tied_pairs": 0, + "correct_pairs": 0, + "frontier_tied_pairs": 1, + "real_tied_pairs": 1 + }, + "rankable": true, + "real_capacity": { + "tp8_mns128_mbt8192": 0.45, + "tp8_mns64_mbt8192": 0.45 + }, + "real_top1_set": [ + "tp8_mns128_mbt8192", + "tp8_mns64_mbt8192" + ], + "spearman_rank_correlation": null, + "top1_regret_fraction_best_tie_break": null, + "top1_regret_fraction_worst_tie_break": null, + "top1_set_intersection": [], + "top1_set_match": false + } + } + } + } +} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py new file mode 100644 index 0000000..0afc07c --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py @@ -0,0 +1,2208 @@ +#!/usr/bin/env python3 +"""Blind Frontier/community-vLLM rank comparison on one fixed prompt cohort. + +The earlier grid selected a different ``sampling_u`` subset at every load. That +confounds offered load with prompt mix and makes binary-search monotonicity an +untested assumption. This runner selects one length-stratified cohort, changes +load only by scaling its complete arrival timeline, evaluates every discrete +load, freezes Frontier first, and only then permits the community-vLLM run. +""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +import os +import random +import statistics +import subprocess +import sys +import time +from concurrent.futures import ThreadPoolExecutor, as_completed +from dataclasses import asdict, replace +from pathlib import Path +from typing import Any, Iterable + + +SCRIPT_DIR = Path(__file__).resolve().parent +REPO_ROOT = SCRIPT_DIR.parents[2] +for search_path in (SCRIPT_DIR, REPO_ROOT / "src"): + if str(search_path) not in sys.path: + sys.path.insert(0, str(search_path)) + +import community_prefill_grid as community_grid # noqa: E402 +import frontier_prefill_grid as frontier_grid # noqa: E402 +from aituner.engine import build_launch_recipe # noqa: E402 +from aituner.spec import ConfigPatch, load_study_spec # noqa: E402 +from aituner.trace import TraceRequest, load_trace_requests # noqa: E402 +from aituner.worker import ( # noqa: E402 + _ignore_sigterm_if_main, + _install_sigterm_as_keyboardinterrupt, + _replay_requests, + _restore_sigterm, + _terminate_process_tree, + _wait_for_server_or_exit, +) + + +PROTOCOL_SCHEMA = "qwen235b-prefill-fixed-cohort-protocol-v1" +FRONTIER_SCHEMA = "frontier-qwen235b-prefill-fixed-cohort-v1" +COMMUNITY_SCHEMA = "community-qwen235b-prefill-fixed-cohort-v1" +COMPARISON_SCHEMA = "frontier-community-qwen235b-rank-comparison-v1" +TRIAL_STABILITY_SCHEMA = "community-qwen235b-trial-stability-v1" +REPEAT_SELECTION_SCHEMA = "community-qwen235b-boundary-repeat-selection-v4" +COHORT_SIZE = 64 +COHORT_SEED = 2026071501 +CONFIG_ORDER_SEED = 2026071502 +RATE_ORDER_SEED = 2026071503 +WARMUP_SEED = 2026071504 +INPUT_BINS = (0, 1024, 2048, 4096, 8192, 16384, 32769) +OFFERED_RATES = (0.15, 0.25, 0.35, 0.50, 0.75, 1.00, 1.50) +TARGET_PASS_RATE = 0.95 +EXPECTED_QUANT_SIGNATURE = "method=fp8|act=dynamic|serialized=True|block=128x128" + +SLO_VARIANTS: dict[str, dict[str, Any]] = { + "linear_8k_primary": { + "description": "TTFT <= 1000 ms + input_tokens / 8000 tokens/s", + "kind": "linear_ms", + "intercept_ms": 1000.0, + "per_token_ms": 0.125, + }, + "linear_6k": { + "description": "TTFT <= 1000 ms + input_tokens / 6000 tokens/s", + "kind": "linear_ms", + "intercept_ms": 1000.0, + "per_token_ms": 1.0 / 6.0, + }, + "linear_10k": { + "description": "TTFT <= 1000 ms + input_tokens / 10000 tokens/s", + "kind": "linear_ms", + "intercept_ms": 1000.0, + "per_token_ms": 0.1, + }, + "legacy_step_1s_2s": { + "description": "Original strict 1 s/2 s step SLO (null-capacity sensitivity)", + "kind": "step_ms", + "buckets": [ + {"max_input_tokens": 8191, "threshold_ms": 1000.0}, + {"threshold_ms": 2000.0}, + ], + }, +} +PRIMARY_SLO = "linear_8k_primary" + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def sha256_text(value: str) -> str: + return hashlib.sha256(value.encode()).hexdigest() + + +def order_hash(values: Iterable[object]) -> str: + return sha256_text("\n".join(str(value) for value in values)) + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def rate_key(rate: float) -> str: + return f"r_{rate:.3f}".replace(".", "p") + + +def _percentile(values: list[float], fraction: float) -> float | None: + if not values: + return None + ordered = sorted(values) + index = min(len(ordered) - 1, max(0, math.ceil(fraction * len(ordered)) - 1)) + return float(ordered[index]) + + +def _latency_summary(values: list[float]) -> dict[str, Any]: + return { + "count": len(values), + "mean": statistics.fmean(values) if values else None, + "p50": _percentile(values, 0.50), + "p95": _percentile(values, 0.95), + "p99": _percentile(values, 0.99), + "max": max(values) if values else None, + } + + +def slo_threshold_ms(variant: dict[str, Any], input_tokens: int) -> float: + if variant["kind"] == "linear_ms": + return float(variant["intercept_ms"]) + float(variant["per_token_ms"]) * input_tokens + if variant["kind"] != "step_ms": + raise ValueError(f"unsupported SLO kind: {variant['kind']}") + for bucket in variant["buckets"]: + ceiling = bucket.get("max_input_tokens") + if ceiling is None or input_tokens <= int(ceiling): + return float(bucket["threshold_ms"]) + raise AssertionError("step SLO must have a terminal bucket") + + +def _input_bin(input_tokens: int) -> int: + for index, (low, high) in enumerate(zip(INPUT_BINS, INPUT_BINS[1:])): + if low <= input_tokens < high: + return index + raise ValueError(f"input length outside protocol bins: {input_tokens}") + + +def _allocate_quotas(counts: list[int], cohort_size: int) -> list[int]: + total = sum(counts) + if cohort_size <= 0 or cohort_size > total: + raise ValueError(f"invalid cohort size {cohort_size} for {total} rows") + exact = [cohort_size * count / total for count in counts] + quotas = [min(count, math.floor(value)) for count, value in zip(counts, exact)] + remaining = cohort_size - sum(quotas) + order = sorted( + range(len(counts)), + key=lambda index: (-(exact[index] - math.floor(exact[index])), index), + ) + while remaining: + advanced = False + for index in order: + if quotas[index] >= counts[index]: + continue + quotas[index] += 1 + remaining -= 1 + advanced = True + if not remaining: + break + if not advanced: + raise AssertionError("could not allocate complete cohort") + return quotas + + +def load_source_rows(trace: Path) -> list[dict[str, Any]]: + rows = [] + with trace.open(encoding="utf-8") as source: + for source_index, line in enumerate(source): + if not line.strip(): + continue + raw = json.loads(line) + input_tokens = int(raw["input_length"]) + if not 0 <= input_tokens <= 32768: + continue + prompt = raw.get("prompt") + if not isinstance(prompt, str) or not prompt: + raise ValueError(f"row {source_index} lacks raw prompt") + rows.append( + { + "source_row_index": source_index, + "source_request_id": str( + raw.get("request_id") or raw.get("id") or source_index + ), + "source_arrival_s": float(raw["timestamp"]), + "sampling_u": float(raw["sampling_u"]), + "input_tokens": input_tokens, + "prompt_sha256": sha256_text(prompt), + "input_bin": _input_bin(input_tokens), + } + ) + return rows + + +def select_cohort( + rows: list[dict[str, Any]], *, cohort_size: int, seed: int +) -> tuple[list[dict[str, Any]], list[int]]: + by_bin = [[] for _ in range(len(INPUT_BINS) - 1)] + for row in rows: + by_bin[int(row["input_bin"])].append(row) + quotas = _allocate_quotas([len(items) for items in by_bin], cohort_size) + selected = [] + for bin_index, (items, quota) in enumerate(zip(by_bin, quotas)): + ranked = sorted( + items, + key=lambda row: sha256_text( + f"{seed}|{bin_index}|{row['source_row_index']}|{row['prompt_sha256']}" + ), + ) + selected.extend(ranked[:quota]) + selected.sort(key=lambda row: (row["source_arrival_s"], row["source_row_index"])) + if len(selected) != cohort_size: + raise AssertionError("cohort selection returned wrong size") + return selected, quotas + + +def prepare_protocol(args: argparse.Namespace) -> Path: + trace = args.trace.resolve() + actual_sha = sha256(trace) + if args.expected_trace_sha256 and actual_sha != args.expected_trace_sha256: + raise ValueError( + f"trace SHA256 mismatch: expected={args.expected_trace_sha256}, actual={actual_sha}" + ) + rows = load_source_rows(trace) + cohort, quotas = select_cohort(rows, cohort_size=args.cohort_size, seed=args.seed) + first_arrival = float(cohort[0]["source_arrival_s"]) + source_span = float(cohort[-1]["source_arrival_s"]) - first_arrival + if source_span <= 0: + raise ValueError("selected cohort has no arrival span") + + output_root = args.output_root.resolve() + trace_dir = output_root / "traces" + trace_dir.mkdir(parents=True, exist_ok=True) + fields = [ + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "source_row_index", + "source_request_id", + "source_arrival_s", + "sampling_u", + "slo_ttft_ms", + ] + rates = {} + for rate in args.rate: + duration_s = len(cohort) / rate + target = trace_dir / f"{rate_key(rate)}.csv" + with target.open("w", encoding="utf-8", newline="") as output: + writer = csv.DictWriter(output, fieldnames=fields) + writer.writeheader() + for row in cohort: + arrival_s = ( + (float(row["source_arrival_s"]) - first_arrival) + / source_span + * duration_s + ) + writer.writerow( + { + "arrived_at": f"{arrival_s:.12f}", + "num_prefill_tokens": row["input_tokens"], + "num_decode_tokens": 1, + "source_row_index": row["source_row_index"], + "source_request_id": row["source_request_id"], + "source_arrival_s": f"{row['source_arrival_s']:.12f}", + "sampling_u": f"{row['sampling_u']:.12f}", + "slo_ttft_ms": f"{slo_threshold_ms(SLO_VARIANTS[PRIMARY_SLO], int(row['input_tokens'])):.6f}", + } + ) + rates[rate_key(rate)] = { + "offered_request_rate": rate, + "duration_s": duration_s, + "path": str(target.resolve()), + "sha256": sha256(target), + "request_count": len(cohort), + "request_rate_recomputed": len(cohort) / duration_s, + "source_row_order_sha256": order_hash( + row["source_row_index"] for row in cohort + ), + "input_length_order_sha256": order_hash( + row["input_tokens"] for row in cohort + ), + } + + bin_counts = [0] * (len(INPUT_BINS) - 1) + for row in rows: + bin_counts[int(row["input_bin"])] += 1 + manifest = { + "schema": PROTOCOL_SCHEMA, + "created_unix_s": time.time(), + "source": { + "path": str(trace), + "sha256": actual_sha, + "eligible_request_count": len(rows), + }, + "selection": { + "method": "length_stratified_smallest_sha256", + "seed": args.seed, + "cohort_size": len(cohort), + "input_bin_edges": list(INPUT_BINS), + "source_bin_counts": bin_counts, + "cohort_bin_quotas": quotas, + "stable_order": ["source_arrival_s", "source_row_index"], + "cohort_source_row_order_sha256": order_hash( + row["source_row_index"] for row in cohort + ), + "source_arrival_span_s": source_span, + }, + "load_contract": { + "only_mutated_variable": "uniform_arrival_timeline_scale", + "request_count_per_point": len(cohort), + "completion_tokens_override": 1, + "target_pass_rate": TARGET_PASS_RATE, + "binary_search": False, + "monotonicity_assumed": False, + "offered_rates": list(args.rate), + }, + "slo_variants": SLO_VARIANTS, + "primary_slo": PRIMARY_SLO, + "cohort": cohort, + "rates": rates, + } + manifest_path = output_root / "protocol_manifest.json" + write_json(manifest_path, manifest) + audit_protocol(manifest_path) + print(manifest_path) + return manifest_path + + +def audit_protocol(manifest_path: Path) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text()) + if manifest.get("schema") != PROTOCOL_SCHEMA: + raise ValueError(f"unexpected protocol schema: {manifest.get('schema')}") + expected_ids = [str(row["source_row_index"]) for row in manifest["cohort"]] + expected_lengths = [int(row["input_tokens"]) for row in manifest["cohort"]] + rate_checks = [] + for record in manifest["rates"].values(): + path = Path(record["path"]) + if sha256(path) != record["sha256"]: + raise ValueError(f"trace hash mismatch: {path}") + with path.open(newline="") as source: + rows = list(csv.DictReader(source)) + ids = [row["source_row_index"] for row in rows] + lengths = [int(row["num_prefill_tokens"]) for row in rows] + arrivals = [float(row["arrived_at"]) for row in rows] + if ids != expected_ids or lengths != expected_lengths: + raise ValueError(f"cohort/order drift in {path}") + if arrivals != sorted(arrivals) or abs(arrivals[0]) > 1e-9: + raise ValueError(f"invalid arrival sequence in {path}") + recomputed = len(rows) / (arrivals[-1] - arrivals[0]) + if not math.isclose( + recomputed, float(record["offered_request_rate"]), rel_tol=1e-9 + ): + raise ValueError(f"offered-rate mismatch in {path}: {recomputed}") + rate_checks.append( + {"rate": record["offered_request_rate"], "recomputed_rate": recomputed} + ) + result = { + "status": "passed", + "protocol_manifest_sha256": sha256(manifest_path), + "cohort_size": len(expected_ids), + "rate_checks": rate_checks, + } + write_json(manifest_path.parent / "protocol_audit.json", result) + return result + + +def _read_trace(path: Path) -> list[dict[str, Any]]: + with path.open(encoding="utf-8", newline="") as source: + return list(csv.DictReader(source)) + + +def score_requests(requests: list[dict[str, Any]]) -> dict[str, Any]: + scores = {} + for name, variant in SLO_VARIANTS.items(): + passed = 0 + for request in requests: + threshold = slo_threshold_ms(variant, int(request["input_tokens"])) + passed += int( + bool(request["success"]) + and request.get("ttft_ms") is not None + and float(request["ttft_ms"]) <= threshold + ) + count = len(requests) + pass_rate = passed / count if count else 0.0 + scores[name] = { + "request_count": count, + "passed_request_count": passed, + "slo_pass_rate": pass_rate, + "target_pass_rate": TARGET_PASS_RATE, + "feasible": pass_rate >= TARGET_PASS_RATE, + } + ttfts = [ + float(request["ttft_ms"]) + for request in requests + if request.get("ttft_ms") is not None + ] + return {"scores": scores, "ttft_ms": _latency_summary(ttfts)} + + +def _capacity_record( + results: list[dict[str, Any]], *, config: dict[str, Any], slo_name: str +) -> dict[str, Any]: + ordered = sorted(results, key=lambda item: float(item["offered_request_rate"])) + feasible = [ + float(item["offered_request_rate"]) + for item in ordered + if item["scores"][slo_name]["feasible"] + ] + vector = [bool(item["scores"][slo_name]["feasible"]) for item in ordered] + violations = [] + for lower_index in range(len(vector)): + for higher_index in range(lower_index + 1, len(vector)): + if not vector[lower_index] and vector[higher_index]: + violations.append( + [ + float(ordered[lower_index]["offered_request_rate"]), + float(ordered[higher_index]["offered_request_rate"]), + ] + ) + capacity = max(feasible) if feasible else None + tp = int(config["tp"]) + return { + "config": config, + "slo": slo_name, + "tested_rates": [float(item["offered_request_rate"]) for item in ordered], + "pass_rates": [float(item["scores"][slo_name]["slo_pass_rate"]) for item in ordered], + "feasibility": vector, + "maximum_tested_feasible_request_rate": capacity, + "maximum_tested_feasible_request_rate_per_gpu": ( + capacity / tp if capacity is not None else None + ), + "lower_censored": capacity is None, + "upper_censored": bool(feasible) and capacity == max(item["offered_request_rate"] for item in ordered), + "monotonicity_violations": violations, + } + + +def rank_surface( + config_results: list[dict[str, Any]], *, slo_name: str +) -> list[dict[str, Any]]: + records = [ + _capacity_record(item["loads"], config=item["config"], slo_name=slo_name) + for item in config_results + ] + records.sort( + key=lambda item: ( + -( + item["maximum_tested_feasible_request_rate_per_gpu"] + if item["maximum_tested_feasible_request_rate_per_gpu"] is not None + else -1.0 + ), + item["config"]["name"], + ) + ) + previous_value = object() + previous_rank = 0 + for index, record in enumerate(records, start=1): + value = record["maximum_tested_feasible_request_rate_per_gpu"] + if value != previous_value: + previous_rank = index + previous_value = value + record["rank"] = previous_rank + return records + + +def _validate_profile_contract(profile_paths: dict[str, Path]) -> dict[str, Any]: + payload = json.loads(profile_paths["manifest"].read_text()) + contract = payload.get("contract") or {} + if contract.get("quant_signature") != EXPECTED_QUANT_SIGNATURE: + raise ValueError(f"unexpected FP8 profile contract: {contract}") + if contract.get("profiling_precision") != "BF16": + raise ValueError(f"unexpected output/accumulation precision: {contract}") + for name, path in ( + ("linear_op.csv", profile_paths["linear"]), + ("attention.csv", profile_paths["attention"]), + ("moe.csv", profile_paths["moe"]), + ): + expected = payload["outputs"][name]["sha256"] + if sha256(path) != expected: + raise ValueError(f"profile output hash mismatch: {path}") + return contract + + +def run_frontier_load( + *, + args: argparse.Namespace, + profile_paths: dict[str, Path], + config: Any, + rate_record: dict[str, Any], +) -> dict[str, Any]: + rate = float(rate_record["offered_request_rate"]) + load_dir = args.output_root / "runs" / config.name / rate_key(rate) + result_path = load_dir / "result.json" + if result_path.is_file(): + result = json.loads(result_path.read_text()) + if result.get("status") == "completed": + return result + load_dir.mkdir(parents=True, exist_ok=True) + trace = Path(rate_record["path"]) + command = frontier_grid.build_command( + python=args.python, + frontier_source=args.frontier_source, + profile_root=args.profile_root, + profile_paths=profile_paths, + trace=trace, + config=config, + probe_dir=load_dir, + run_id=f"{config.name}_{rate_key(rate)}", + cache_root=args.output_root / "cache" / config.name, + ) + write_json(load_dir / "command.json", command) + environment = os.environ.copy() + environment.update( + { + "PYTHONPATH": str(args.frontier_source), + "WANDB_DISABLED": "true", + "VIDUR_DISABLE_WANDB": "1", + } + ) + started = time.time() + with (load_dir / "stdout.log").open("w", encoding="utf-8") as output: + completed = subprocess.run( + command, + cwd=args.frontier_source, + env=environment, + stdout=output, + stderr=subprocess.STDOUT, + check=False, + ) + if completed.returncode != 0: + raise RuntimeError( + f"Frontier failed: config={config.name}, rate={rate}, rc={completed.returncode}" + ) + request_metrics = frontier_grid.find_request_metrics(load_dir) + trace_rows = _read_trace(trace) + with request_metrics.open(encoding="utf-8", newline="") as source: + metric_rows = list(csv.DictReader(source)) + metrics_by_id = {int(row["Request Id"]): row for row in metric_rows} + if set(metrics_by_id) != set(range(len(trace_rows))): + raise ValueError(f"Frontier Request Id mismatch: {request_metrics}") + requests = [ + { + "request_id": row["source_row_index"], + "input_tokens": int(row["num_prefill_tokens"]), + "success": True, + "ttft_ms": float(metrics_by_id[index]["ttft"]), + } + for index, row in enumerate(trace_rows) + ] + score = score_requests(requests) + result = { + "status": "completed", + "config": asdict(config) | {"name": config.name}, + "offered_request_rate": rate, + "request_rate_per_gpu": rate / config.gpu_count, + "elapsed_seconds": time.time() - started, + "trace_path": str(trace), + "trace_sha256": sha256(trace), + "request_metrics_path": str(request_metrics), + "request_metrics_sha256": sha256(request_metrics), + "requests": requests, + **score, + } + write_json(result_path, result) + return result + + +def run_frontier(args: argparse.Namespace) -> Path: + protocol = json.loads(args.protocol_manifest.read_text()) + if protocol.get("schema") != PROTOCOL_SCHEMA: + raise ValueError("invalid fixed-cohort protocol manifest") + audit_protocol(args.protocol_manifest) + profile_paths = frontier_grid.resolve_profile_paths(args.profile_root) + profile_contract = _validate_profile_contract(profile_paths) + ep_equivalence = json.loads(args.ep_equivalence_artifact.read_text()) + if ( + ep_equivalence.get("schema") != "frontier-ep-batched-lane-equivalence-v1" + or not ep_equivalence.get("byte_identical") + ): + raise ValueError("batched EP lane prediction lacks byte-identical evidence") + if Path(ep_equivalence["new_source"]).resolve() != args.frontier_source.resolve(): + raise ValueError("EP equivalence was measured on a different Frontier source") + configs = frontier_grid.selected_configs(args.config) + args.output_root.mkdir(parents=True, exist_ok=True) + run_manifest = { + "schema": FRONTIER_SCHEMA, + "created_unix_s": time.time(), + "protocol_manifest": { + "path": str(args.protocol_manifest.resolve()), + "sha256": sha256(args.protocol_manifest), + }, + "frontier": { + "source": str(args.frontier_source.resolve()), + "python": str(args.python.resolve()), + "fingerprint": frontier_grid.frontier_source_fingerprint( + args.frontier_source, args.frontier_commit + ), + }, + "profiles": { + "contract": profile_contract, + "files": { + name: {"path": str(path.resolve()), "sha256": sha256(path)} + for name, path in profile_paths.items() + }, + }, + "ep_prediction_optimization_equivalence": { + "path": str(args.ep_equivalence_artifact.resolve()), + "sha256": sha256(args.ep_equivalence_artifact), + "request_metrics_byte_identical": True, + "request_metrics_sha256": ep_equivalence["new_sha256"], + }, + "kv_capacity_evidence": { + "tp4": { + "path": str(args.tp4_capacity_artifact.resolve()), + "sha256": sha256(args.tp4_capacity_artifact), + "num_gpu_blocks": 26101, + }, + "tp8": { + "path": str(args.tp8_capacity_artifact.resolve()), + "sha256": sha256(args.tp8_capacity_artifact), + "num_gpu_blocks": 62351, + }, + }, + "best_effort_contract": { + "arrival_semantics": "preserve_materialized_trace_arrivals", + "quantization": "block_fp8_w8a8_dynamic_with_bf16_output_accumulation", + "execution_profiles": "community_vllm_0.10.2_serving_entrypoints", + "moe_ep_prefill": "global_routing_then_slowest_local_ep_lane", + "moe_ep_critical_lane_cache": "exact_(total_routed_tokens,layer_id)_memoization", + "moe_ep_lane_inference": "eight_independent_rows_in_one_exact_forest_predict_call", + "kv_blocks": "measured_community_vllm_and_fixed_per_topology", + "cpu_overhead_modeling": "disabled_no_native_community_vllm_records", + "end_to_end_action_specific_calibration": False, + "real_serving_results_observed_before_freeze": False, + }, + "orchestration": { + "max_parallel_configs": args.max_parallel_configs, + "cache_isolation": "one_cache_root_per_config", + "loads_within_config": "sequential", + }, + "configs": [asdict(config) | {"name": config.name} for config in configs], + } + write_json(args.output_root / "run_manifest.json", run_manifest) + + rate_records = sorted( + protocol["rates"].values(), key=lambda item: item["offered_request_rate"] + ) + + def run_config(config: Any) -> dict[str, Any]: + loads = [] + for record in rate_records: + result = run_frontier_load( + args=args, + profile_paths=profile_paths, + config=config, + rate_record=record, + ) + loads.append(result) + print( + json.dumps( + { + "system": "Frontier", + "config": config.name, + "rate": result["offered_request_rate"], + "pass_rate": result["scores"][PRIMARY_SLO]["slo_pass_rate"], + "feasible": result["scores"][PRIMARY_SLO]["feasible"], + "elapsed_seconds": result["elapsed_seconds"], + }, + sort_keys=True, + ), + flush=True, + ) + summary = {"config": asdict(config) | {"name": config.name}, "loads": loads} + write_json(args.output_root / "results" / f"{config.name}.json", summary) + return summary + + results_by_name = {} + with ThreadPoolExecutor(max_workers=args.max_parallel_configs) as pool: + futures = {pool.submit(run_config, config): config for config in configs} + for future in as_completed(futures): + config = futures[future] + results_by_name[config.name] = future.result() + config_results = [results_by_name[config.name] for config in configs] + freeze = { + "schema": FRONTIER_SCHEMA, + "status": "frozen_before_community_run", + "created_unix_s": time.time(), + "run_manifest_sha256": sha256(args.output_root / "run_manifest.json"), + "protocol_manifest_sha256": sha256(args.protocol_manifest), + "primary_slo": PRIMARY_SLO, + "rankings": { + name: rank_surface(config_results, slo_name=name) for name in SLO_VARIANTS + }, + "config_results": config_results, + } + path = args.output_root / "frontier_ranking_frozen.json" + write_json(path, freeze) + print(path) + return path + + +def _community_study_payload( + *, + tp: int, + repo: Path, + python: Path, + vllm: Path, + model: Path, + trace: Path, + windows: Path, + port: int, +) -> dict[str, Any]: + payload = community_grid.study_payload( + tp=tp, + repo=repo, + python=python, + vllm=vllm, + model=model, + trace=trace, + windows=windows, + port=port, + ) + payload["study_id"] = f"community-qwen235b-prefill-fixed-cohort-tp{tp}-v1" + payload["slo"] = { + "target_pass_rate": TARGET_PASS_RATE, + "ttft_rule": { + key: value + for key, value in SLO_VARIANTS[PRIMARY_SLO].items() + if key != "description" + }, + } + # The dedicated runner supplies exact fixed-cohort arrivals and disables + # SLO-based early stopping; these trace values only define raw prompt loading. + payload["trace"]["early_stop_max_lag_s"] = None + payload["trace"]["early_stop_max_elapsed_s"] = None + return payload + + +def _validate_frontier_freeze(path: Path, protocol_manifest: Path) -> dict[str, Any]: + freeze = json.loads(path.read_text()) + if freeze.get("schema") != FRONTIER_SCHEMA: + raise ValueError(f"unexpected Frontier freeze schema: {freeze.get('schema')}") + if freeze.get("status") != "frozen_before_community_run": + raise ValueError("Frontier result is not marked frozen") + if freeze.get("protocol_manifest_sha256") != sha256(protocol_manifest): + raise ValueError("Frontier and community protocol manifests differ") + if len(freeze.get("config_results") or []) != len(frontier_grid.GRID): + raise ValueError("Frontier freeze is incomplete") + protocol = json.loads(protocol_manifest.read_text()) + expected_rates = { + float(record["offered_request_rate"]) + for record in protocol["rates"].values() + } + for item in freeze["config_results"]: + rates = {float(load["offered_request_rate"]) for load in item["loads"]} + if rates != expected_rates: + raise ValueError(f"Frontier freeze lacks rates for {item['config']['name']}") + return freeze + + +def prepare_community(args: argparse.Namespace) -> Path: + protocol = json.loads(args.protocol_manifest.read_text()) + if protocol.get("schema") != PROTOCOL_SCHEMA: + raise ValueError("invalid fixed-cohort protocol manifest") + _validate_frontier_freeze(args.frontier_freeze, args.protocol_manifest) + repo = args.repo.resolve() + python = args.python.resolve() + vllm = args.vllm.resolve() + model = args.model.resolve() + windows = args.windows.resolve() + trace = Path(protocol["source"]["path"]) + for path in (repo, python, vllm, model, windows, trace): + if not path.exists(): + raise FileNotFoundError(path) + if sha256(trace) != protocol["source"]["sha256"]: + raise ValueError("raw prompt trace changed after protocol freeze") + output_root = args.output_root.resolve() + output_root.mkdir(parents=True, exist_ok=True) + studies = {} + for tp in (4, 8): + path = output_root / "studies" / f"tp{tp}" / "study.json" + write_json( + path, + _community_study_payload( + tp=tp, + repo=repo, + python=python, + vllm=vllm, + model=model, + trace=trace, + windows=windows, + port=args.port, + ), + ) + studies[str(tp)] = {"path": str(path), "sha256": sha256(path)} + + configs = list(community_grid.GRID) + random.Random(args.config_order_seed).shuffle(configs) + records = [] + for execution_index, config in enumerate(configs, start=1): + rates = list(OFFERED_RATES) + random.Random(f"{args.rate_order_seed}|{config.name}").shuffle(rates) + records.append( + { + "execution_index": execution_index, + "config": asdict(config) | {"name": config.name}, + "rate_order": rates, + "study_path": studies[str(config.tp)]["path"], + "result_path": str(output_root / "results" / f"{config.name}.json"), + "engine_log_path": str(output_root / "runs" / config.name / "engine.log"), + } + ) + total_replay_seconds = sum( + float(record["duration_s"]) for record in protocol["rates"].values() + ) * len(configs) + manifest = { + "schema": COMMUNITY_SCHEMA, + "created_unix_s": time.time(), + "repository": community_grid.git_fingerprint(repo), + "protocol_manifest": { + "path": str(args.protocol_manifest.resolve()), + "sha256": sha256(args.protocol_manifest), + }, + "frontier_freeze": { + "path": str(args.frontier_freeze.resolve()), + "sha256": sha256(args.frontier_freeze), + }, + "runtime": {"python": str(python), "vllm": str(vllm)}, + "model": { + "path": str(model), + "config_sha256": sha256(model / "config.json"), + }, + "studies": studies, + "execution_contract": { + "request_mode": "raw_completion", + "completion_tokens_override": 1, + "prefix_caching": False, + "client_max_concurrency": community_grid.CLIENT_MAX_CONCURRENCY, + "slo_early_stop": False, + "one_server_launch_per_config": True, + "warmup": { + "method": "one_disjoint_request_per_input_length_bin", + "request_count": len(INPUT_BINS) - 1, + "seed": WARMUP_SEED, + "latencies_discarded": True, + }, + "rate_order_randomized_within_config": True, + "config_order_seed": args.config_order_seed, + "rate_order_seed": args.rate_order_seed, + "estimated_replay_wall_seconds_excluding_8_model_loads": total_replay_seconds, + }, + "configs": records, + } + path = output_root / "run_manifest.json" + write_json(path, manifest) + print(path) + return path + + +def _boundary_repeat_rates( + config_result: dict[str, Any], *, slo_name: str = PRIMARY_SLO +) -> list[float]: + """Select every adjacent load pair whose SLO feasibility label changes.""" + loads = sorted( + config_result["loads"], key=lambda item: float(item["offered_request_rate"]) + ) + rates = [float(item["offered_request_rate"]) for item in loads] + if len(rates) != len(set(rates)): + raise ValueError(f"duplicate offered rate in {config_result['config']['name']}") + if not rates: + raise ValueError(f"no loads in {config_result['config']['name']}") + feasibility = [bool(item["scores"][slo_name]["feasible"]) for item in loads] + selected: set[float] = set() + for index in range(len(loads) - 1): + if feasibility[index] != feasibility[index + 1]: + selected.update((rates[index], rates[index + 1])) + if not selected: + if len(rates) == 1: + selected.add(rates[0]) + elif all(feasibility): + selected.update(rates[-2:]) + else: + selected.update(rates[:2]) + return sorted(selected) + + +def _slo_has_feasible_load(config_result: dict[str, Any], *, slo_name: str) -> bool: + return any( + bool(load["scores"][slo_name]["feasible"]) + for load in config_result["loads"] + ) + + +def prepare_community_repeat(args: argparse.Namespace) -> Path: + """Prepare a fresh-server reverse-order repeat of decision-boundary loads.""" + selected_slos = list(getattr(args, "slo", [])) or list(SLO_VARIANTS) + unknown_slos = sorted(set(selected_slos) - set(SLO_VARIANTS)) + if unknown_slos: + raise ValueError(f"unknown repeat SLOs: {unknown_slos}") + selected_configs = set(getattr(args, "config", [])) + source_manifest = json.loads(args.source_manifest.read_text()) + first_pass = json.loads(args.community_freeze.read_text()) + if source_manifest.get("schema") != COMMUNITY_SCHEMA: + raise ValueError("unexpected source community manifest") + if ( + first_pass.get("schema") != COMMUNITY_SCHEMA + or first_pass.get("status") != "completed" + ): + raise ValueError("unexpected/incomplete first-pass community freeze") + if first_pass.get("run_manifest_sha256") != sha256(args.source_manifest): + raise ValueError("first-pass freeze does not match the source manifest") + source_protocol_path = Path(source_manifest["protocol_manifest"]["path"]) + if sha256(source_protocol_path) != source_manifest["protocol_manifest"]["sha256"]: + raise ValueError("source protocol changed before repeat preparation") + protocol_path = ( + args.protocol_manifest.resolve() + if args.protocol_manifest is not None + else source_protocol_path + ) + frontier_freeze_path = ( + args.frontier_freeze.resolve() + if args.frontier_freeze is not None + else Path(source_manifest["frontier_freeze"]["path"]) + ) + _validate_frontier_freeze(frontier_freeze_path, protocol_path) + protocol = json.loads(protocol_path.read_text()) + duration_by_rate = { + float(record["offered_request_rate"]): float(record["duration_s"]) + for record in protocol["rates"].values() + } + source_records = { + item["config"]["name"]: item for item in source_manifest["configs"] + } + first_results = { + item["config"]["name"]: item for item in first_pass["config_results"] + } + if set(source_records) != set(first_results): + raise ValueError("source manifest and first-pass config sets differ") + + output_root = args.output_root.resolve() + output_root.mkdir(parents=True, exist_ok=True) + first_execution_order = sorted( + source_manifest["configs"], key=lambda item: int(item["execution_index"]) + ) + if selected_configs: + unknown_configs = sorted(selected_configs - set(source_records)) + if unknown_configs: + raise ValueError(f"unknown repeat configs: {unknown_configs}") + first_execution_order = [ + item + for item in first_execution_order + if item["config"]["name"] in selected_configs + ] + records = [] + selections = [] + for execution_index, source_record in enumerate( + reversed(first_execution_order), start=1 + ): + name = source_record["config"]["name"] + excluded_unrankable_slos = [ + slo_name + for slo_name in selected_slos + if not _slo_has_feasible_load( + first_results[name], slo_name=slo_name + ) + ] + boundary_rates_by_slo = { + slo_name: ( + [] + if slo_name in excluded_unrankable_slos + else _boundary_repeat_rates( + first_results[name], slo_name=slo_name + ) + ) + for slo_name in selected_slos + } + boundary_rates = sorted( + { + rate + for rates in boundary_rates_by_slo.values() + for rate in rates + } + ) + refinement_rates = sorted( + { + rate + for rates in boundary_rates_by_slo.values() + for low, high in zip(rates, rates[1:]) + for rate in duration_by_rate + if low < rate < high and rate not in boundary_rates + } + ) + selected_rates = sorted(set(boundary_rates) | set(refinement_rates)) + source_relative_order = [ + float(rate) + for rate in source_record["rate_order"] + if float(rate) in boundary_rates + ] + if set(source_relative_order) != set(boundary_rates): + raise ValueError(f"repeat rate is absent from source order: {name}") + reversed_boundaries = list(reversed(source_relative_order)) + repeat_order = [*reversed_boundaries, *reversed(refinement_rates)] + records.append( + { + "execution_index": execution_index, + "config": source_record["config"], + "rate_order": repeat_order, + "study_path": source_record["study_path"], + "result_path": str(output_root / "results" / f"{name}.json"), + "engine_log_path": str(output_root / "runs" / name / "engine.log"), + } + ) + selections.append( + { + "config": name, + "excluded_unrankable_slos": excluded_unrankable_slos, + "boundary_rates_by_slo": boundary_rates_by_slo, + "boundary_rates": boundary_rates, + "added_refinement_rates": refinement_rates, + "selected_rates": selected_rates, + "source_relative_order": source_relative_order, + "repeat_order": repeat_order, + } + ) + total_replay_seconds = sum( + duration_by_rate[float(rate)] + for record in records + for rate in record["rate_order"] + ) + manifest = { + "schema": COMMUNITY_SCHEMA, + "created_unix_s": time.time(), + "repository": community_grid.git_fingerprint(args.repo.resolve()), + "protocol_manifest": { + "path": str(protocol_path), + "sha256": sha256(protocol_path), + }, + "frontier_freeze": { + "path": str(frontier_freeze_path), + "sha256": sha256(frontier_freeze_path), + }, + "runtime": source_manifest["runtime"], + "model": source_manifest["model"], + "studies": source_manifest["studies"], + "repeat_of": { + "source_manifest": { + "path": str(args.source_manifest.resolve()), + "sha256": sha256(args.source_manifest), + }, + "first_pass_freeze": { + "path": str(args.community_freeze.resolve()), + "sha256": sha256(args.community_freeze), + }, + }, + "execution_contract": { + "schema": REPEAT_SELECTION_SCHEMA, + "selection": "all_adjacent_pre_registered_slo_feasibility_transitions_plus_between_rate_refinement", + "selected_slos": selected_slos, + "selected_configs": sorted(selected_configs) if selected_configs else "all", + "unrankable_slo_policy": "exclude_per_config_surfaces_with_no_feasible_load", + "fallback_when_no_transition": "top_two_if_all_feasible_else_bottom_two", + "config_order": "reverse_of_first_pass", + "rate_order": "reverse_first_pass_boundary_order_then_descending_refinement_rates", + "one_fresh_server_launch_per_config": True, + "warmup": source_manifest["execution_contract"]["warmup"], + "estimated_replay_wall_seconds_excluding_model_loads": total_replay_seconds, + "selections": selections, + }, + "configs": records, + } + path = output_root / "run_manifest.json" + write_json(path, manifest) + print(path) + return path + + +def _materialize_real_requests( + *, + all_requests: list[TraceRequest], + rate_record: dict[str, Any], + cohort: list[dict[str, Any]], +) -> list[TraceRequest]: + by_id = {request.row_id: request for request in all_requests} + expected_prompt_hash = { + str(row["source_row_index"]): row["prompt_sha256"] for row in cohort + } + trace_rows = _read_trace(Path(rate_record["path"])) + requests = [] + for row in trace_rows: + request_id = row["source_row_index"] + if request_id not in by_id: + raise ValueError(f"cohort request is absent from raw trace loader: {request_id}") + request = by_id[request_id] + prompt = request.body.get("prompt") + if not isinstance(prompt, str) or sha256_text(prompt) != expected_prompt_hash[request_id]: + raise ValueError(f"raw prompt fingerprint changed: request={request_id}") + requests.append(replace(request, arrival_s=float(row["arrived_at"]))) + return requests + + +def _select_warmup_requests( + all_requests: list[TraceRequest], *, cohort_ids: set[str] +) -> list[TraceRequest]: + by_bin = [[] for _ in range(len(INPUT_BINS) - 1)] + for request in all_requests: + if request.row_id in cohort_ids or request.prompt_tokens_hint is None: + continue + input_tokens = int(request.prompt_tokens_hint) + if not 0 <= input_tokens <= 32768: + continue + by_bin[_input_bin(input_tokens)].append(request) + selected = [] + for bin_index, candidates in enumerate(by_bin): + if not candidates: + raise ValueError(f"no disjoint warmup request in input bin {bin_index}") + selected.append( + min( + candidates, + key=lambda request: sha256_text( + f"{WARMUP_SEED}|{bin_index}|{request.row_id}" + ), + ) + ) + return [replace(request, arrival_s=0.0) for request in selected] + + +def _run_real_replay( + requests: list[TraceRequest], *, recipe: Any, study: Any, max_elapsed_s: float +) -> tuple[list[Any], bool, str]: + return _replay_requests( + requests, + base_url=recipe.base_url, + timeout_s=recipe.request_timeout_s, + max_concurrency=study.trace.max_concurrency, + # target=0 makes the mathematical SLO early-stop bound unreachable + # while retaining the existing replay engine. + target_pass_rate=0.0, + max_lag_s=None, + max_elapsed_s=max_elapsed_s, + evaluate_outcome=lambda outcome: type( + "NoEarlyStopEvaluation", (), {"passed": bool(outcome.success)} + )(), + ) + + +def _real_request_records( + requests: list[TraceRequest], outcomes: list[Any] +) -> list[dict[str, Any]]: + outcomes_by_id = {outcome.request_id: outcome for outcome in outcomes} + records = [] + for request in requests: + outcome = outcomes_by_id.get(request.row_id) + records.append( + { + "request_id": request.row_id, + "arrival_s": request.arrival_s, + "input_tokens": int(request.prompt_tokens_hint or 0), + "success": bool(outcome and outcome.success), + "ttft_ms": outcome.ttft_ms if outcome is not None else None, + "tpot_ms": outcome.tpot_ms if outcome is not None else None, + "completion_tokens": outcome.completion_tokens if outcome is not None else None, + "completion_tokens_source": ( + outcome.completion_tokens_source if outcome is not None else "" + ), + "error": outcome.error if outcome is not None else "missing_outcome", + } + ) + return records + + +def run_community_config( + *, + manifest: dict[str, Any], + record: dict[str, Any], + protocol: dict[str, Any], +) -> dict[str, Any]: + result_path = Path(record["result_path"]) + if result_path.is_file(): + result = json.loads(result_path.read_text()) + if result.get("status") == "completed": + return result + config = record["config"] + study_path = Path(record["study_path"]) + study = load_study_spec(study_path) + _, all_requests = load_trace_requests(study, study_spec_path=study_path) + recipe = build_launch_recipe( + study.engine, + ConfigPatch( + flag_patch={ + "max-num-seqs": int(config["mns"]), + "max-num-batched-tokens": int(config["mbt"]), + } + ), + ) + run_dir = Path(record["engine_log_path"]).parent + run_dir.mkdir(parents=True, exist_ok=True) + write_json(run_dir / "engine_command.json", recipe.argv) + marker_env = { + "AITUNER_STUDY_ID": study.study_id, + "AITUNER_TRIAL_ID": f"fixed-cohort-{config['name']}", + } + loads = [] + load_by_rate = { + float(item["offered_request_rate"]): item + for item in protocol["rates"].values() + } + started = time.time() + with Path(record["engine_log_path"]).open("w", encoding="utf-8") as engine_log: + process = subprocess.Popen( # noqa: S603 + recipe.argv, + cwd=recipe.cwd, + env={**recipe.env, **marker_env}, + stdout=engine_log, + stderr=subprocess.STDOUT, + text=True, + start_new_session=True, + ) + previous_sigterm = _install_sigterm_as_keyboardinterrupt() + try: + _wait_for_server_or_exit( + process, + base_url=recipe.base_url, + healthcheck_path=recipe.healthcheck_path, + ready_timeout_s=recipe.ready_timeout_s, + ) + cohort_ids = { + str(row["source_row_index"]) for row in protocol["cohort"] + } + warmup_requests = _select_warmup_requests( + all_requests, cohort_ids=cohort_ids + ) + warmup_started = time.time() + warmup_outcomes, warmup_early_stopped, warmup_stop_reason = _run_real_replay( + warmup_requests, + recipe=recipe, + study=study, + max_elapsed_s=1200.0, + ) + warmup_records = _real_request_records(warmup_requests, warmup_outcomes) + warmup_result = { + "status": "completed" + if not warmup_early_stopped + and all(item["success"] for item in warmup_records) + else "failed", + "elapsed_seconds": time.time() - warmup_started, + "early_stopped": warmup_early_stopped, + "early_stop_reason": warmup_stop_reason, + "cohort_disjoint": not bool( + cohort_ids & {item["request_id"] for item in warmup_records} + ), + "requests": warmup_records, + } + write_json(run_dir / "warmup.json", warmup_result) + if warmup_result["status"] != "completed": + raise RuntimeError(f"server warmup failed: {warmup_result}") + time.sleep(2.0) + for rate in record["rate_order"]: + rate = float(rate) + load_dir = run_dir / rate_key(rate) + load_result_path = load_dir / "result.json" + if load_result_path.is_file(): + load_result = json.loads(load_result_path.read_text()) + if load_result.get("status") == "completed": + loads.append(load_result) + continue + load_dir.mkdir(parents=True, exist_ok=True) + rate_record = load_by_rate[rate] + requests = _materialize_real_requests( + all_requests=all_requests, + rate_record=rate_record, + cohort=protocol["cohort"], + ) + load_started = time.time() + outcomes, early_stopped, early_stop_reason = _run_real_replay( + requests, + recipe=recipe, + study=study, + max_elapsed_s=max(float(rate_record["duration_s"]) + 900.0, 1200.0), + ) + request_records = _real_request_records(requests, outcomes) + score = score_requests(request_records) + load_result = { + "status": "completed", + "config": config, + "offered_request_rate": rate, + "request_rate_per_gpu": rate / int(config["tp"]), + "elapsed_seconds": time.time() - load_started, + "trace_path": rate_record["path"], + "trace_sha256": rate_record["sha256"], + "early_stopped": early_stopped, + "early_stop_reason": early_stop_reason, + "requests": request_records, + **score, + } + if early_stopped: + raise RuntimeError( + f"complete replay stopped early: {config['name']} rate={rate}: " + f"{early_stop_reason}" + ) + write_json(load_result_path, load_result) + loads.append(load_result) + print( + json.dumps( + { + "system": "community_vllm", + "config": config["name"], + "rate": rate, + "pass_rate": score["scores"][PRIMARY_SLO]["slo_pass_rate"], + "feasible": score["scores"][PRIMARY_SLO]["feasible"], + "elapsed_seconds": load_result["elapsed_seconds"], + }, + sort_keys=True, + ), + flush=True, + ) + time.sleep(2.0) + finally: + _ignore_sigterm_if_main() + _terminate_process_tree(process, timeout_s=30.0, marker_env=marker_env) + _restore_sigterm(previous_sigterm) + loads.sort(key=lambda item: float(item["offered_request_rate"])) + result = { + "schema": COMMUNITY_SCHEMA, + "status": "completed", + "config": config, + "elapsed_seconds": time.time() - started, + "loads": loads, + } + write_json(result_path, result) + return result + + +def run_community(args: argparse.Namespace) -> Path: + manifest = json.loads(args.manifest.read_text()) + if manifest.get("schema") != COMMUNITY_SCHEMA: + raise ValueError(f"unexpected community manifest schema: {manifest.get('schema')}") + protocol_path = Path(manifest["protocol_manifest"]["path"]) + frontier_freeze = Path(manifest["frontier_freeze"]["path"]) + if sha256(protocol_path) != manifest["protocol_manifest"]["sha256"]: + raise ValueError("protocol changed after community manifest preparation") + if sha256(frontier_freeze) != manifest["frontier_freeze"]["sha256"]: + raise ValueError("Frontier freeze changed after community manifest preparation") + _validate_frontier_freeze(frontier_freeze, protocol_path) + protocol = json.loads(protocol_path.read_text()) + config_results = [] + for record in sorted(manifest["configs"], key=lambda item: item["execution_index"]): + result = run_community_config( + manifest=manifest, + record=record, + protocol=protocol, + ) + config_results.append(result) + print( + json.dumps( + { + "config": result["config"]["name"], + "config_completed": True, + "elapsed_seconds": result["elapsed_seconds"], + }, + sort_keys=True, + ), + flush=True, + ) + freeze = { + "schema": COMMUNITY_SCHEMA, + "status": "completed", + "created_unix_s": time.time(), + "run_manifest_sha256": sha256(args.manifest), + "protocol_manifest_sha256": sha256(protocol_path), + "frontier_freeze_sha256": sha256(frontier_freeze), + "primary_slo": PRIMARY_SLO, + "rankings": { + name: rank_surface(config_results, slo_name=name) for name in SLO_VARIANTS + }, + "config_results": config_results, + } + path = args.manifest.parent / "community_ranking_frozen.json" + write_json(path, freeze) + print(path) + return path + + +def _rank_map(ranking: list[dict[str, Any]]) -> dict[str, int]: + return {item["config"]["name"]: int(item["rank"]) for item in ranking} + + +def _capacity_map( + ranking: list[dict[str, Any]], *, field: str +) -> dict[str, float | None]: + return { + item["config"]["name"]: item[field] + for item in ranking + } + + +def _pearson(left: list[float], right: list[float]) -> float | None: + if len(left) < 2: + return None + left_mean = statistics.fmean(left) + right_mean = statistics.fmean(right) + numerator = sum((a - left_mean) * (b - right_mean) for a, b in zip(left, right)) + denominator = math.sqrt( + sum((a - left_mean) ** 2 for a in left) + * sum((b - right_mean) ** 2 for b in right) + ) + return numerator / denominator if denominator else None + + +def _average_ranks(capacities: dict[str, float | None]) -> dict[str, float]: + ordered = sorted( + capacities, + key=lambda name: ( + -( + capacities[name] + if capacities[name] is not None + else float("-inf") + ), + name, + ), + ) + ranks = {} + start = 0 + while start < len(ordered): + value = capacities[ordered[start]] + end = start + 1 + while end < len(ordered) and capacities[ordered[end]] == value: + end += 1 + average_rank = ((start + 1) + end) / 2.0 + for name in ordered[start:end]: + ranks[name] = average_rank + start = end + return ranks + + +def _pairwise_accuracy( + frontier_capacity: dict[str, float | None], real_capacity: dict[str, float | None] +) -> dict[str, Any]: + names = sorted(set(frontier_capacity) & set(real_capacity)) + comparable = 0 + correct = 0 + frontier_ties = 0 + real_ties = 0 + for left_index, left in enumerate(names): + for right in names[left_index + 1 :]: + f_left = frontier_capacity[left] + f_right = frontier_capacity[right] + r_left = real_capacity[left] + r_right = real_capacity[right] + if None in (f_left, f_right, r_left, r_right): + continue + f_sign = (f_left > f_right) - (f_left < f_right) + r_sign = (r_left > r_right) - (r_left < r_right) + frontier_ties += int(f_sign == 0) + real_ties += int(r_sign == 0) + if f_sign == 0 or r_sign == 0: + continue + comparable += 1 + correct += int(f_sign == r_sign) + return { + "comparable_non_tied_pairs": comparable, + "correct_pairs": correct, + "accuracy": correct / comparable if comparable else None, + "frontier_tied_pairs": frontier_ties, + "real_tied_pairs": real_ties, + } + + +def _objective_fidelity( + frontier_ranking: list[dict[str, Any]], + real_ranking: list[dict[str, Any]], + *, + field: str, + only_names: set[str] | None = None, +) -> dict[str, Any]: + frontier_capacity = _capacity_map(frontier_ranking, field=field) + real_capacity = _capacity_map(real_ranking, field=field) + if only_names is not None: + frontier_capacity = { + name: value for name, value in frontier_capacity.items() if name in only_names + } + real_capacity = { + name: value for name, value in real_capacity.items() if name in only_names + } + names = sorted(set(frontier_capacity) & set(real_capacity)) + frontier_valid = [value for value in frontier_capacity.values() if value is not None] + real_valid = [value for value in real_capacity.values() if value is not None] + if not frontier_valid or not real_valid: + return { + "rankable": False, + "reason": "one_or_both_surfaces_have_no_feasible_config", + "frontier_capacity": frontier_capacity, + "real_capacity": real_capacity, + } + frontier_average_ranks = _average_ranks(frontier_capacity) + real_average_ranks = _average_ranks(real_capacity) + best_frontier = max(frontier_valid) + best_real = max(real_valid) + frontier_top_set = [ + name for name, value in frontier_capacity.items() if value == best_frontier + ] + real_top_set = [name for name, value in real_capacity.items() if value == best_real] + selected_real_capacities = [ + real_capacity[name] + for name in frontier_top_set + if real_capacity.get(name) is not None + ] + regret_values = [ + (best_real - value) / best_real + for value in selected_real_capacities + if best_real != 0 + ] + return { + "rankable": True, + "capacity_field": field, + "frontier_top1_set": frontier_top_set, + "real_top1_set": real_top_set, + "top1_set_intersection": sorted(set(frontier_top_set) & set(real_top_set)), + "top1_set_match": set(frontier_top_set) == set(real_top_set), + "top1_regret_fraction_best_tie_break": min(regret_values) + if regret_values + else None, + "top1_regret_fraction_worst_tie_break": max(regret_values) + if regret_values + else None, + "spearman_rank_correlation": _pearson( + [frontier_average_ranks[name] for name in names], + [real_average_ranks[name] for name in names], + ), + "pairwise_ordering": _pairwise_accuracy(frontier_capacity, real_capacity), + "frontier_capacity": frontier_capacity, + "real_capacity": real_capacity, + } + + +def _request_residuals( + frontier_results: list[dict[str, Any]], real_results: list[dict[str, Any]] +) -> dict[str, Any]: + def index(results: list[dict[str, Any]]) -> dict[tuple[str, float, str], dict[str, Any]]: + indexed = {} + for config in results: + name = config["config"]["name"] + for load in config["loads"]: + rate = float(load["offered_request_rate"]) + for request in load["requests"]: + indexed[(name, rate, str(request["request_id"]))] = request + return indexed + + frontier_index = index(frontier_results) + real_index = index(real_results) + keys = sorted(set(frontier_index) & set(real_index)) + errors = [] + absolute_percentage_errors = [] + by_bin: dict[str, list[float]] = {} + for key in keys: + frontier_request = frontier_index[key] + real_request = real_index[key] + if not real_request["success"] or real_request.get("ttft_ms") is None: + continue + predicted = float(frontier_request["ttft_ms"]) + observed = float(real_request["ttft_ms"]) + error = predicted - observed + errors.append(error) + if observed > 0: + absolute_percentage_errors.append(abs(error) / observed) + input_tokens = int(real_request["input_tokens"]) + bin_index = _input_bin(input_tokens) + label = f"[{INPUT_BINS[bin_index]},{INPUT_BINS[bin_index + 1]})" + by_bin.setdefault(label, []).append(error) + return { + "matched_successful_requests": len(errors), + "signed_error_ms_mean": statistics.fmean(errors) if errors else None, + "mae_ms": statistics.fmean(abs(value) for value in errors) if errors else None, + "mape": ( + statistics.fmean(absolute_percentage_errors) + if absolute_percentage_errors + else None + ), + "by_input_bin": { + label: { + "count": len(values), + "signed_error_ms_mean": statistics.fmean(values), + "mae_ms": statistics.fmean(abs(value) for value in values), + } + for label, values in sorted(by_bin.items()) + }, + } + + +def _pass_rate_residuals( + frontier_results: list[dict[str, Any]], real_results: list[dict[str, Any]] +) -> dict[str, Any]: + def index( + results: list[dict[str, Any]], slo_name: str + ) -> dict[tuple[str, float], tuple[float, bool]]: + return { + (config["config"]["name"], float(load["offered_request_rate"])): ( + float(load["scores"][slo_name]["slo_pass_rate"]), + bool(load["scores"][slo_name]["feasible"]), + ) + for config in results + for load in config["loads"] + } + + variants = {} + for slo_name in SLO_VARIANTS: + frontier_index = index(frontier_results, slo_name) + real_index = index(real_results, slo_name) + keys = sorted(set(frontier_index) & set(real_index)) + errors = [frontier_index[key][0] - real_index[key][0] for key in keys] + label_matches = [frontier_index[key][1] == real_index[key][1] for key in keys] + false_positives = sum( + frontier_index[key][1] and not real_index[key][1] for key in keys + ) + false_negatives = sum( + not frontier_index[key][1] and real_index[key][1] for key in keys + ) + variants[slo_name] = { + "matched_config_load_points": len(errors), + "signed_error_mean": statistics.fmean(errors) if errors else None, + "mae": statistics.fmean(abs(value) for value in errors) if errors else None, + "max_absolute_error": max((abs(value) for value in errors), default=None), + "feasibility_label_accuracy": ( + sum(label_matches) / len(label_matches) if label_matches else None + ), + "false_feasible_points": false_positives, + "false_infeasible_points": false_negatives, + } + return variants + + +def _request_slo_classification( + frontier_results: list[dict[str, Any]], real_results: list[dict[str, Any]] +) -> dict[str, Any]: + def index( + results: list[dict[str, Any]], + ) -> dict[tuple[str, float, str], dict[str, Any]]: + return { + ( + config["config"]["name"], + float(load["offered_request_rate"]), + str(request["request_id"]), + ): request + for config in results + for load in config["loads"] + for request in load["requests"] + } + + def request_passes(request: dict[str, Any], variant: dict[str, Any]) -> bool: + return bool( + request["success"] + and request.get("ttft_ms") is not None + and float(request["ttft_ms"]) + <= slo_threshold_ms(variant, int(request["input_tokens"])) + ) + + frontier_index = index(frontier_results) + real_index = index(real_results) + keys = sorted(set(frontier_index) & set(real_index)) + variants = {} + for slo_name, variant in SLO_VARIANTS.items(): + by_load: dict[tuple[str, float], dict[str, Any]] = {} + frontier_failures: set[tuple[str, float, str]] = set() + real_failures: set[tuple[str, float, str]] = set() + for key in keys: + frontier_passes = request_passes(frontier_index[key], variant) + real_passes = request_passes(real_index[key], variant) + if not frontier_passes: + frontier_failures.add(key) + if not real_passes: + real_failures.add(key) + group = by_load.setdefault( + key[:2], + { + "matched_requests": 0, + "matching_labels": 0, + "false_slo_pass_request_ids": [], + "false_slo_fail_request_ids": [], + }, + ) + group["matched_requests"] += 1 + group["matching_labels"] += int(frontier_passes == real_passes) + if frontier_passes and not real_passes: + group["false_slo_pass_request_ids"].append(key[2]) + elif not frontier_passes and real_passes: + group["false_slo_fail_request_ids"].append(key[2]) + intersection = frontier_failures & real_failures + union = frontier_failures | real_failures + load_records = [] + for (config, rate), record in sorted(by_load.items()): + matched = int(record["matched_requests"]) + load_records.append( + { + "config": config, + "offered_request_rate": rate, + **record, + "label_accuracy": record["matching_labels"] / matched, + } + ) + matching_labels = sum(item["matching_labels"] for item in by_load.values()) + false_passes = sum( + len(item["false_slo_pass_request_ids"]) for item in by_load.values() + ) + false_fails = sum( + len(item["false_slo_fail_request_ids"]) for item in by_load.values() + ) + variants[slo_name] = { + "matched_requests": len(keys), + "matching_labels": matching_labels, + "label_accuracy": matching_labels / len(keys) if keys else None, + "false_slo_pass_requests": false_passes, + "false_slo_fail_requests": false_fails, + "frontier_failure_requests": len(frontier_failures), + "real_failure_requests": len(real_failures), + "failure_set_intersection": len(intersection), + "failure_set_union": len(union), + "failure_set_jaccard": len(intersection) / len(union) if union else 1.0, + "by_config_load": load_records, + } + return variants + + +def compare(args: argparse.Namespace) -> Path: + frontier = json.loads(args.frontier_freeze.read_text()) + community = json.loads(args.community_freeze.read_text()) + if frontier.get("schema") != FRONTIER_SCHEMA: + raise ValueError("unexpected Frontier freeze") + if community.get("schema") != COMMUNITY_SCHEMA or community.get("status") != "completed": + raise ValueError("unexpected/incomplete community freeze") + if frontier["protocol_manifest_sha256"] != community["protocol_manifest_sha256"]: + raise ValueError("comparison inputs used different protocols") + variants = {} + for slo_name in SLO_VARIANTS: + frontier_ranking = frontier["rankings"][slo_name] + real_ranking = community["rankings"][slo_name] + variants[slo_name] = { + "per_gpu_efficiency": _objective_fidelity( + frontier_ranking, + real_ranking, + field="maximum_tested_feasible_request_rate_per_gpu", + ), + "single_replica_raw_capacity": _objective_fidelity( + frontier_ranking, + real_ranking, + field="maximum_tested_feasible_request_rate", + ), + "within_topology_raw_capacity": { + f"tp{tp}": _objective_fidelity( + frontier_ranking, + real_ranking, + field="maximum_tested_feasible_request_rate", + only_names={ + item["config"]["name"] + for item in frontier_ranking + if int(item["config"]["tp"]) == tp + }, + ) + for tp in (4, 8) + }, + "frontier_ranking": frontier_ranking, + "real_ranking": real_ranking, + } + result = { + "schema": COMPARISON_SCHEMA, + "created_unix_s": time.time(), + "frontier_freeze": { + "path": str(args.frontier_freeze.resolve()), + "sha256": sha256(args.frontier_freeze), + }, + "community_freeze": { + "path": str(args.community_freeze.resolve()), + "sha256": sha256(args.community_freeze), + }, + "primary_slo": PRIMARY_SLO, + "variants": variants, + "request_level_ttft_residuals": _request_residuals( + frontier["config_results"], community["config_results"] + ), + "pass_rate_surface_residuals": _pass_rate_residuals( + frontier["config_results"], community["config_results"] + ), + "request_slo_classification": _request_slo_classification( + frontier["config_results"], community["config_results"] + ), + } + write_json(args.output, result) + print(args.output) + return args.output + + +def _index_loads( + config_results: list[dict[str, Any]], +) -> dict[tuple[str, float], dict[str, Any]]: + return { + (config["config"]["name"], float(load["offered_request_rate"])): load + for config in config_results + for load in config["loads"] + } + + +def _index_requests( + config_results: list[dict[str, Any]], +) -> dict[tuple[str, float, str], dict[str, Any]]: + return { + ( + config["config"]["name"], + float(load["offered_request_rate"]), + str(request["request_id"]), + ): request + for config in config_results + for load in config["loads"] + for request in load["requests"] + } + + +def _request_passes_slo(request: dict[str, Any], variant: dict[str, Any]) -> bool: + return bool( + request["success"] + and request.get("ttft_ms") is not None + and float(request["ttft_ms"]) + <= slo_threshold_ms(variant, int(request["input_tokens"])) + ) + + +def compare_trials(args: argparse.Namespace) -> Path: + frontier = json.loads(args.frontier_freeze.read_text()) + first = json.loads(args.first_community_freeze.read_text()) + repeat = json.loads(args.repeat_community_freeze.read_text()) + if frontier.get("schema") != FRONTIER_SCHEMA: + raise ValueError("unexpected Frontier freeze") + for label, payload in (("first", first), ("repeat", repeat)): + if payload.get("schema") != COMMUNITY_SCHEMA or payload.get("status") != "completed": + raise ValueError(f"unexpected/incomplete {label} community freeze") + + frontier_loads = _index_loads(frontier["config_results"]) + first_loads = _index_loads(first["config_results"]) + repeat_loads = _index_loads(repeat["config_results"]) + shared_load_keys = sorted(set(frontier_loads) & set(first_loads) & set(repeat_loads)) + if not shared_load_keys: + raise ValueError("trials have no shared config/load points") + for key in shared_load_keys: + trace_hashes = { + index[key].get("trace_sha256") + for index in (frontier_loads, first_loads, repeat_loads) + if index[key].get("trace_sha256") is not None + } + if len(trace_hashes) > 1: + raise ValueError(f"trace mismatch at {key}") + + frontier_requests = _index_requests(frontier["config_results"]) + first_requests = _index_requests(first["config_results"]) + repeat_requests = _index_requests(repeat["config_results"]) + shared_request_keys = sorted( + set(frontier_requests) & set(first_requests) & set(repeat_requests) + ) + if not shared_request_keys: + raise ValueError("trials have no shared requests") + for key in shared_request_keys: + lengths = { + int(index[key]["input_tokens"]) + for index in (frontier_requests, first_requests, repeat_requests) + } + if len(lengths) != 1: + raise ValueError(f"input-token mismatch at {key}") + + trial_ttft_differences = [] + for key in shared_request_keys: + first_request = first_requests[key] + repeat_request = repeat_requests[key] + if ( + first_request["success"] + and repeat_request["success"] + and first_request.get("ttft_ms") is not None + and repeat_request.get("ttft_ms") is not None + ): + trial_ttft_differences.append( + float(repeat_request["ttft_ms"]) - float(first_request["ttft_ms"]) + ) + + variants = {} + for slo_name, variant in SLO_VARIANTS.items(): + request_counts = { + "matching_real_trial_labels": 0, + "frontier_correct_first": 0, + "frontier_correct_repeat": 0, + "frontier_wrong_both_stable_real_label": 0, + "real_trial_label_flips": 0, + } + first_failures = set() + repeat_failures = set() + for key in shared_request_keys: + predicted = _request_passes_slo(frontier_requests[key], variant) + first_observed = _request_passes_slo(first_requests[key], variant) + repeat_observed = _request_passes_slo(repeat_requests[key], variant) + request_counts["matching_real_trial_labels"] += int( + first_observed == repeat_observed + ) + request_counts["frontier_correct_first"] += int(predicted == first_observed) + request_counts["frontier_correct_repeat"] += int(predicted == repeat_observed) + request_counts["frontier_wrong_both_stable_real_label"] += int( + first_observed == repeat_observed and predicted != first_observed + ) + request_counts["real_trial_label_flips"] += int( + first_observed != repeat_observed + ) + if not first_observed: + first_failures.add(key) + if not repeat_observed: + repeat_failures.add(key) + + load_records = [] + matching_load_labels = 0 + frontier_correct_first_loads = 0 + frontier_correct_repeat_loads = 0 + for config_name, rate in shared_load_keys: + predicted_score = frontier_loads[(config_name, rate)]["scores"][slo_name] + first_score = first_loads[(config_name, rate)]["scores"][slo_name] + repeat_score = repeat_loads[(config_name, rate)]["scores"][slo_name] + predicted_label = bool(predicted_score["feasible"]) + first_label = bool(first_score["feasible"]) + repeat_label = bool(repeat_score["feasible"]) + matching_load_labels += int(first_label == repeat_label) + frontier_correct_first_loads += int(predicted_label == first_label) + frontier_correct_repeat_loads += int(predicted_label == repeat_label) + load_records.append( + { + "config": config_name, + "offered_request_rate": rate, + "frontier_passed_request_count": int( + predicted_score["passed_request_count"] + ), + "first_passed_request_count": int(first_score["passed_request_count"]), + "repeat_passed_request_count": int( + repeat_score["passed_request_count"] + ), + "frontier_feasible": predicted_label, + "first_feasible": first_label, + "repeat_feasible": repeat_label, + } + ) + failure_union = first_failures | repeat_failures + variants[slo_name] = { + "shared_requests": len(shared_request_keys), + **request_counts, + "real_trial_request_label_accuracy": ( + request_counts["matching_real_trial_labels"] / len(shared_request_keys) + ), + "frontier_request_label_accuracy_first": ( + request_counts["frontier_correct_first"] / len(shared_request_keys) + ), + "frontier_request_label_accuracy_repeat": ( + request_counts["frontier_correct_repeat"] / len(shared_request_keys) + ), + "real_trial_failure_set_jaccard": ( + len(first_failures & repeat_failures) / len(failure_union) + if failure_union + else 1.0 + ), + "shared_config_load_points": len(shared_load_keys), + "real_trial_load_label_accuracy": matching_load_labels / len(shared_load_keys), + "frontier_load_label_accuracy_first": ( + frontier_correct_first_loads / len(shared_load_keys) + ), + "frontier_load_label_accuracy_repeat": ( + frontier_correct_repeat_loads / len(shared_load_keys) + ), + "config_load_records": load_records, + } + + first_primary_capacity = _capacity_map( + first["rankings"][PRIMARY_SLO], + field="maximum_tested_feasible_request_rate", + ) + first_config = { + item["config"]["name"]: item["config"] for item in first["config_results"] + } + decision_records = [] + repeat_capacity_records = [] + repeated_config_names = {name for name, _ in repeat_loads} + for config_name in sorted(set(first_primary_capacity) & repeated_config_names): + old_rates = sorted( + rate for name, rate in first_loads if name == config_name + ) + capacity = first_primary_capacity[config_name] + if capacity is None: + continue + higher_rates = [rate for rate in old_rates if rate > capacity] + next_rate = min(higher_rates) if higher_rates else None + boundary_rates = [capacity] + ([next_rate] if next_rate is not None else []) + boundary_repeated = all( + (config_name, rate) in repeat_loads for rate in boundary_rates + ) + capacity_still_feasible = bool( + boundary_repeated + and repeat_loads[(config_name, capacity)]["scores"][PRIMARY_SLO]["feasible"] + ) + next_still_infeasible = bool( + next_rate is None + or ( + boundary_repeated + and not repeat_loads[(config_name, next_rate)]["scores"][PRIMARY_SLO][ + "feasible" + ] + ) + ) + stable = boundary_repeated and capacity_still_feasible and next_still_infeasible + decision_records.append( + { + "config": config_name, + "first_capacity": capacity, + "first_next_tested_rate": next_rate, + "boundary_repeated": boundary_repeated, + "capacity_still_feasible": capacity_still_feasible, + "next_rate_still_infeasible": next_still_infeasible, + "stable_original_grid_capacity": stable, + } + ) + repeat_capacity = capacity if stable else None + repeat_capacity_records.append( + { + "config": first_config[config_name], + "maximum_tested_feasible_request_rate": repeat_capacity, + "maximum_tested_feasible_request_rate_per_gpu": ( + repeat_capacity / int(first_config[config_name]["tp"]) + if repeat_capacity is not None + else None + ), + } + ) + + result = { + "schema": TRIAL_STABILITY_SCHEMA, + "created_unix_s": time.time(), + "inputs": { + "frontier_freeze": { + "path": str(args.frontier_freeze.resolve()), + "sha256": sha256(args.frontier_freeze), + }, + "first_community_freeze": { + "path": str(args.first_community_freeze.resolve()), + "sha256": sha256(args.first_community_freeze), + }, + "repeat_community_freeze": { + "path": str(args.repeat_community_freeze.resolve()), + "sha256": sha256(args.repeat_community_freeze), + }, + }, + "shared_config_load_points": len(shared_load_keys), + "shared_requests": len(shared_request_keys), + "real_trial_ttft_ms": { + "paired_successful_requests": len(trial_ttft_differences), + "repeat_minus_first_mean": ( + statistics.fmean(trial_ttft_differences) + if trial_ttft_differences + else None + ), + "mae": ( + statistics.fmean(abs(value) for value in trial_ttft_differences) + if trial_ttft_differences + else None + ), + "absolute_error_p95": _percentile( + [abs(value) for value in trial_ttft_differences], 0.95 + ), + }, + "variants": variants, + "primary_original_grid_capacity_stability": { + "stable_configs": sum( + item["stable_original_grid_capacity"] for item in decision_records + ), + "total_configs": len(decision_records), + "records": decision_records, + "per_gpu_efficiency": _objective_fidelity( + first["rankings"][PRIMARY_SLO], + repeat_capacity_records, + field="maximum_tested_feasible_request_rate_per_gpu", + ), + "single_replica_raw_capacity": _objective_fidelity( + first["rankings"][PRIMARY_SLO], + repeat_capacity_records, + field="maximum_tested_feasible_request_rate", + ), + }, + } + write_json(args.output, result) + print(args.output) + return args.output + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + subparsers = parser.add_subparsers(dest="command", required=True) + + prepare = subparsers.add_parser("prepare-protocol") + prepare.add_argument("--trace", type=Path, required=True) + prepare.add_argument("--expected-trace-sha256", default=frontier_grid.TRACE_SHA256) + prepare.add_argument("--output-root", type=Path, required=True) + prepare.add_argument("--cohort-size", type=int, default=COHORT_SIZE) + prepare.add_argument("--seed", type=int, default=COHORT_SEED) + prepare.add_argument("--rate", action="append", type=float, default=[]) + + audit = subparsers.add_parser("audit-protocol") + audit.add_argument("--manifest", type=Path, required=True) + + frontier = subparsers.add_parser("run-frontier") + frontier.add_argument("--frontier-source", type=Path, required=True) + frontier.add_argument( + "--frontier-commit", default="d9cfeb6d8791fbf2f295dd9744c56a666171776e" + ) + frontier.add_argument("--python", type=Path, required=True) + frontier.add_argument("--profile-root", type=Path, required=True) + frontier.add_argument("--protocol-manifest", type=Path, required=True) + frontier.add_argument("--output-root", type=Path, required=True) + frontier.add_argument("--tp4-capacity-artifact", type=Path, required=True) + frontier.add_argument("--tp8-capacity-artifact", type=Path, required=True) + frontier.add_argument("--ep-equivalence-artifact", type=Path, required=True) + frontier.add_argument("--max-parallel-configs", type=int, default=1) + frontier.add_argument("--config", action="append") + + community = subparsers.add_parser("prepare-community") + community.add_argument("--repo", type=Path, required=True) + community.add_argument("--python", type=Path, required=True) + community.add_argument("--vllm", type=Path, required=True) + community.add_argument("--model", type=Path, required=True) + community.add_argument("--windows", type=Path, required=True) + community.add_argument("--protocol-manifest", type=Path, required=True) + community.add_argument("--frontier-freeze", type=Path, required=True) + community.add_argument("--output-root", type=Path, required=True) + community.add_argument("--port", type=int, default=18918) + community.add_argument("--config-order-seed", type=int, default=CONFIG_ORDER_SEED) + community.add_argument("--rate-order-seed", type=int, default=RATE_ORDER_SEED) + + run_real = subparsers.add_parser("run-community") + run_real.add_argument("--manifest", type=Path, required=True) + + repeat = subparsers.add_parser("prepare-community-repeat") + repeat.add_argument("--repo", type=Path, required=True) + repeat.add_argument("--source-manifest", type=Path, required=True) + repeat.add_argument("--community-freeze", type=Path, required=True) + repeat.add_argument("--protocol-manifest", type=Path) + repeat.add_argument("--frontier-freeze", type=Path) + repeat.add_argument("--output-root", type=Path, required=True) + repeat.add_argument("--slo", action="append", default=[]) + repeat.add_argument("--config", action="append", default=[]) + + compare_parser = subparsers.add_parser("compare") + compare_parser.add_argument("--frontier-freeze", type=Path, required=True) + compare_parser.add_argument("--community-freeze", type=Path, required=True) + compare_parser.add_argument("--output", type=Path, required=True) + + trials = subparsers.add_parser("compare-trials") + trials.add_argument("--frontier-freeze", type=Path, required=True) + trials.add_argument("--first-community-freeze", type=Path, required=True) + trials.add_argument("--repeat-community-freeze", type=Path, required=True) + trials.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + if args.command == "prepare-protocol": + if not args.rate: + args.rate = list(OFFERED_RATES) + if sorted(set(args.rate)) != list(args.rate): + raise ValueError("--rate values must be unique and ascending") + prepare_protocol(args) + elif args.command == "audit-protocol": + print(json.dumps(audit_protocol(args.manifest), indent=2)) + elif args.command == "run-frontier": + run_frontier(args) + elif args.command == "prepare-community": + prepare_community(args) + elif args.command == "run-community": + run_community(args) + elif args.command == "prepare-community-repeat": + prepare_community_repeat(args) + elif args.command == "compare": + compare(args) + elif args.command == "compare-trials": + compare_trials(args) + else: + raise AssertionError(args.command) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/fleet_fixed_cohort.toml b/runs/frontier-multicase-sufficiency-v0/best_effort/fleet_fixed_cohort.toml new file mode 100644 index 0000000..7ecc003 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/fleet_fixed_cohort.toml @@ -0,0 +1,25 @@ +version = 1 + +[paths] +state_dir = ".aituner/gpu_fleet/fixed_cohort_state" +artifacts_dir = ".aituner/gpu_fleet/fixed_cohort_artifacts" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 1024 +gpu_free_utilization_pct = 10 +prefer_pack = true + +[sync] +mode = "rsync" +local_path = "runs/frontier-multicase-sufficiency-v0/best_effort" +exclude = ["__pycache__/", "*.pyc"] + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/aituner-frontier-best-effort-20260715/runs/frontier-multicase-sufficiency-v0/best_effort" +fleet_root = "/home/admin/cpfs/wjh/.aituner_gpu_fleet_fixed_cohort" diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_fp8_profile_contract.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_fp8_profile_contract.patch new file mode 100644 index 0000000..012fd79 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_fp8_profile_contract.patch @@ -0,0 +1,89 @@ +diff --git a/frontier/config/quantization_manager.py b/frontier/config/quantization_manager.py +--- a/frontier/config/quantization_manager.py ++++ b/frontier/config/quantization_manager.py +@@ -356,9 +356,17 @@ class QuantizationManager: + target_precision = self.get_precision(op_name) + precision_match = target_precision == profiling_precision + quant_match = profiling_quant_signature == expected_quant_signature ++ # A mixed-precision profiling CSV records the model's output dtype at ++ # file level. Quantized operators in that CSV were nevertheless ++ # measured with the exact quantization scheme identified by the ++ # quant signature. Treating those samples as BF16 and scaling them ++ # again would double-apply the FP8 speedup. ++ exact_fp8_profile = quant_match and target_precision == PrecisionType.FP8 + with self._lock: +- self._operation_profiling_precision[op_name] = profiling_precision +- if precision_match: ++ self._operation_profiling_precision[op_name] = ( ++ target_precision if exact_fp8_profile else profiling_precision ++ ) ++ if precision_match or exact_fp8_profile: + self._operation_data_sources[op_name] = "profiling" + self._operation_approximation_factors.pop(op_name, None) + self._operation_speedup_factors.pop(op_name, None) +diff --git a/tests/unit/test_quantization_profile_contract.py b/tests/unit/test_quantization_profile_contract.py +new file mode 100644 +--- /dev/null ++++ b/tests/unit/test_quantization_profile_contract.py +@@ -0,0 +1,61 @@ ++from unittest.mock import MagicMock ++ ++import pytest ++ ++from frontier.config.model_config import QuantizationConfig ++from frontier.config.precision_type import PrecisionType ++from frontier.config.quantization_manager import QuantizationManager ++ ++ ++def test_exact_fp8_quant_signature_does_not_rescale_mixed_profile() -> None: ++ manager = QuantizationManager() ++ manager.load_config() ++ ++ quant_config = QuantizationConfig( ++ quant_method="fp8", ++ activation_scheme="dynamic", ++ is_checkpoint_fp8_serialized=True, ++ weight_block_size=(128, 128), ++ ) ++ quant_signature = quant_config.get_quant_signature() ++ model_config = MagicMock() ++ model_config.get_default_precision.return_value = PrecisionType.BF16 ++ model_config.get_name.return_value = "Qwen3-235B-A22B-FP8" ++ model_config.torch_dtype = "bfloat16" ++ model_config.quantization_config = quant_config ++ model_config.get_quant_signature.return_value = quant_signature ++ ++ manager.configure_from_model_config(model_config) ++ manager.register_profiling_metadata( ++ operation_names=["attn_pre_proj"], ++ profiling_precision=PrecisionType.BF16, ++ profiling_quant_signature=quant_signature, ++ expected_quant_signature=quant_signature, ++ file_path="linear_op.csv", ++ ) ++ ++ metadata = { ++ item["operation"]: item ++ for item in manager.get_operation_precision_metadata() ++ } ++ assert metadata["attn_pre_proj"]["data_source"] == "profiling" ++ assert metadata["attn_pre_proj"]["approximation_factor"] is None ++ assert manager.has_precision_mismatch("attn_pre_proj") is False ++ assert manager.adjust_compute_time("attn_pre_proj", 1.25) == pytest.approx(1.25) ++ ++ ++def test_mismatched_fp8_quant_signature_still_uses_approximation() -> None: ++ manager = QuantizationManager() ++ manager.load_config() ++ manager._operation_precisions["attn_pre_proj"] = PrecisionType.FP8 ++ ++ manager.register_profiling_metadata( ++ operation_names=["attn_pre_proj"], ++ profiling_precision=PrecisionType.BF16, ++ profiling_quant_signature="none", ++ expected_quant_signature="method=fp8", ++ file_path="linear_op.csv", ++ ) ++ ++ assert manager.has_precision_mismatch("attn_pre_proj") is True ++ assert manager.adjust_compute_time("attn_pre_proj", 1.0) == pytest.approx(0.5) diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_batched_lane_predict.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_batched_lane_predict.patch new file mode 100644 index 0000000..088aae6 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_batched_lane_predict.patch @@ -0,0 +1,82 @@ +diff --git a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +--- a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py ++++ b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +@@ -385,6 +385,78 @@ + allocation_ratios=allocation_ratios, + ) + ++ def _get_on_demand_predictions_batch( ++ self, ++ model_name: str, ++ feature_rows: List[Dict[str, float]], ++ ) -> List[float]: ++ """Predict independent feature rows in one exact forest invocation. ++ ++ ``RandomForestRegressor.predict`` applies the same trees to every row. ++ Batching the eight EP lanes removes repeated pandas/joblib call overhead; ++ it does not combine lanes or alter their features. Results populate the ++ same runtime cache used by the scalar prediction path. ++ """ ++ if not feature_rows: ++ return [] ++ ++ model_info = self._predictions.get(model_name) ++ if ( ++ not isinstance(model_info, dict) ++ or not model_info.get("_on_demand_prediction", False) ++ ): ++ raise ValueError( ++ f"Model {model_name} is not configured for on-demand prediction" ++ ) ++ model = model_info.get("_model") ++ feature_names = model_info.get("_feature_names", []) ++ if model is None: ++ raise ValueError(f"Model {model_name} has no trained model available") ++ ++ family_name = self._measurement_family_name(self._active_measurement_type) ++ runtime_cache = self._runtime_cache[family_name][model_name] ++ results: List[Optional[float]] = [None] * len(feature_rows) ++ misses: Dict[tuple, tuple] = {} ++ for index, features in enumerate(feature_rows): ++ missing = [name for name in feature_names if name not in features] ++ if missing: ++ raise ValueError( ++ f"On-demand prediction missing required features for {model_name}: " ++ f"{missing}. Provided keys: {sorted(list(features.keys()))}" ++ ) ++ cache_key = tuple(features[key] for key in sorted(features.keys())) ++ if cache_key in runtime_cache: ++ results[index] = float(runtime_cache[cache_key]) ++ continue ++ if cache_key not in misses: ++ misses[cache_key] = (features, []) ++ misses[cache_key][1].append(index) ++ ++ if misses: ++ miss_keys = list(misses) ++ frame = pd.DataFrame( ++ [ ++ [misses[key][0][name] for name in feature_names] ++ for key in miss_keys ++ ], ++ columns=feature_names, ++ ) ++ try: ++ predictions = model.predict(frame) ++ except Exception as error: ++ raise ValueError( ++ f"Batched on-demand prediction failed for {model_name}: {error}" ++ ) from error ++ for cache_key, raw_prediction in zip(miss_keys, predictions): ++ prediction = max(0.0, float(raw_prediction)) ++ runtime_cache[cache_key] = prediction ++ for index in misses[cache_key][1]: ++ results[index] = prediction ++ ++ if any(result is None for result in results): ++ raise AssertionError("batched on-demand prediction left an empty result") ++ return [float(result) for result in results] ++ + @lru_cache(maxsize=None) + def _get_critical_prefill_ep_lane_tokens( + self, diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_critical_lane.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_critical_lane.patch new file mode 100644 index 0000000..cc6d4ee --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_critical_lane.patch @@ -0,0 +1,76 @@ +diff --git a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +--- a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py ++++ b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +@@ -386,0 +387,58 @@ ++ def _get_critical_prefill_ep_lane_tokens( ++ self, ++ total_routed_tokens: int, ++ layer_id: int, ++ ) -> Dict[int, int]: ++ """Return the routed-token allocation for the slowest prefill EP lane. ++ ++ A monolithic prefill starts with a global ``tokens * topk`` routing ++ population, but each EP rank executes only the assignments for its local ++ experts. The ranks synchronize after the MoE block, so the layer cost is ++ governed by the slowest lane rather than by either the global population ++ or the average lane. ++ """ ++ total_experts = int(self._replica_config.total_expert_num) ++ lane_count = int(self._moe_ep_size) ++ if total_experts <= 0 or lane_count <= 1 or total_experts % lane_count != 0: ++ raise ValueError( ++ "Prefill EP routing requires total_expert_num to be positive and " ++ f"divisible by moe_ep_size > 1; got total_expert_num={total_experts}, " ++ f"moe_ep_size={lane_count}" ++ ) ++ ++ experts_per_lane = total_experts // lane_count ++ global_tokens = self._get_global_per_expert_tokens( ++ total_routed_tokens=total_routed_tokens, ++ layer_id=layer_id, ++ ) ++ lane_tokens: List[Dict[int, int]] = [ ++ {local_expert_id: 0 for local_expert_id in range(experts_per_lane)} ++ for _ in range(lane_count) ++ ] ++ for global_expert_id, token_count in global_tokens.items(): ++ lane_id = int(global_expert_id) // experts_per_lane ++ local_expert_id = int(global_expert_id) % experts_per_lane ++ lane_tokens[lane_id][local_expert_id] = int(token_count) ++ ++ def _lane_compute_score(allocation: Dict[int, int]) -> float: ++ features = self._build_moe_load_imbalance_features(allocation) ++ score = 0.0 ++ for operation in ("moe_shuffling", "moe_grouped_gemm"): ++ prediction = self._predictions.get(operation) ++ if isinstance(prediction, dict) and prediction.get( ++ "_on_demand_prediction", False ++ ): ++ score += float( ++ self._get_on_demand_prediction(operation, features) ++ ) ++ return score ++ ++ return max( ++ lane_tokens, ++ key=lambda allocation: ( ++ _lane_compute_score(allocation), ++ sum(allocation.values()), ++ tuple(allocation.values()), ++ ), ++ ) ++ +@@ -1023,0 +1082,13 @@ ++ ++ from frontier.entities import EPBatchGroup ++ ++ is_monolithic_prefill_ep = ( ++ self._moe_ep_size > 1 ++ and not isinstance(batch, EPBatchGroup) ++ and not bool(getattr(batch, "is_pure_decode_batch", False)) ++ ) ++ if is_monolithic_prefill_ep: ++ return self._get_critical_prefill_ep_lane_tokens( ++ total_routed_tokens=num_tokens * self._router_topk, ++ layer_id=layer_id, ++ ) diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_exact_cache.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_exact_cache.patch new file mode 100644 index 0000000..0bb9203 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_exact_cache.patch @@ -0,0 +1,17 @@ +diff --git a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +--- a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py ++++ b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +@@ -1,4 +1,5 @@ + from typing import Any, Dict, List, Optional, TYPE_CHECKING, Union ++from functools import lru_cache + import os + + import numpy as np +@@ -384,6 +385,7 @@ + allocation_ratios=allocation_ratios, + ) + ++ @lru_cache(maxsize=None) + def _get_critical_prefill_ep_lane_tokens( + self, + total_routed_tokens: int, diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_use_batched_lanes.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_use_batched_lanes.patch new file mode 100644 index 0000000..5b23cff --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_ep_prefill_use_batched_lanes.patch @@ -0,0 +1,56 @@ +diff --git a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +--- a/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py ++++ b/frontier/execution_time_predictor/sklearn_moe_execution_time_predictor.py +@@ -494,27 +494,35 @@ + local_expert_id = int(global_expert_id) % experts_per_lane + lane_tokens[lane_id][local_expert_id] = int(token_count) + +- def _lane_compute_score(allocation: Dict[int, int]) -> float: +- features = self._build_moe_load_imbalance_features(allocation) +- score = 0.0 +- for operation in ("moe_shuffling", "moe_grouped_gemm"): +- prediction = self._predictions.get(operation) +- if isinstance(prediction, dict) and prediction.get( +- "_on_demand_prediction", False +- ): +- score += float( +- self._get_on_demand_prediction(operation, features) ++ lane_features = [ ++ self._build_moe_load_imbalance_features(allocation) ++ for allocation in lane_tokens ++ ] ++ lane_scores = [0.0] * lane_count ++ for operation in ("moe_shuffling", "moe_grouped_gemm"): ++ prediction = self._predictions.get(operation) ++ if isinstance(prediction, dict) and prediction.get( ++ "_on_demand_prediction", False ++ ): ++ operation_predictions = self._get_on_demand_predictions_batch( ++ operation, lane_features ++ ) ++ lane_scores = [ ++ score + operation_prediction ++ for score, operation_prediction in zip( ++ lane_scores, operation_predictions + ) +- return score ++ ] + +- return max( +- lane_tokens, +- key=lambda allocation: ( +- _lane_compute_score(allocation), +- sum(allocation.values()), +- tuple(allocation.values()), ++ critical_lane_index = max( ++ range(lane_count), ++ key=lambda lane_index: ( ++ lane_scores[lane_index], ++ sum(lane_tokens[lane_index].values()), ++ tuple(lane_tokens[lane_index].values()), + ), + ) ++ return lane_tokens[critical_lane_index] + + def predict_monolithic_decode_shared_domain_lane_moe_times_ms( + self, diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_linear_plan.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_linear_plan.patch new file mode 100644 index 0000000..1910739 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_linear_plan.patch @@ -0,0 +1,72 @@ +diff --git a/frontier/profiling/linear_op/profiling_plan.py b/frontier/profiling/linear_op/profiling_plan.py +--- a/frontier/profiling/linear_op/profiling_plan.py ++++ b/frontier/profiling/linear_op/profiling_plan.py +@@ -112,5 +112,11 @@ def build_profiling_plan( + + ffn_sharded_enabled = tp_size in ffn_tp_set ++ if is_moe and not _supports_share_expert(model_config): ++ # Routed experts are profiled by the MoE profiler. Keeping the ++ # dense FFN surrogate enabled executes an unprofiled fake weight ++ # path and can reject otherwise-valid attention TP layouts (for ++ # example Qwen3-235B TP8 with 128x128 block FP8 weights). ++ ffn_sharded_enabled = False + + padded_n_embd = model_config.embedding_dim + padded_n_expanded_embd = model_config.mlp_hidden_dim +diff --git a/tests/unit/test_moe_linear_profiling_plan.py b/tests/unit/test_moe_linear_profiling_plan.py +new file mode 100644 +--- /dev/null ++++ b/tests/unit/test_moe_linear_profiling_plan.py +@@ -0,0 +1,52 @@ ++from types import SimpleNamespace ++ ++from frontier.profiling.linear_op.profiling_plan import build_profiling_plan ++ ++ ++def test_routed_moe_plan_does_not_execute_dense_ffn_surrogate() -> None: ++ model_config = SimpleNamespace( ++ no_tensor_parallel=False, ++ embedding_dim=4096, ++ mlp_hidden_dim=1536, ++ num_q_heads=64, ++ num_kv_heads=4, ++ model_type="qwen3_moe", ++ post_attn_norm=True, ++ is_moe=True, ++ is_step2_mini=False, ++ ) ++ ++ plan = build_profiling_plan( ++ model_config=model_config, ++ tp_size=8, ++ attn_tp=[8], ++ ffn_tp=[8], ++ disable_replicated=False, ++ is_moe=True, ++ ) ++ ++ assert plan["attn_sharded_enabled"] is True ++ assert plan["ffn_sharded_enabled"] is False ++ assert plan["ffn_enabled"] is True ++ assert plan["padded_n_expanded_embd"] == 1536 ++ assert "attn_pre_proj" in plan["enabled_ops"] ++ assert "post_attention_layernorm" in plan["enabled_ops"] ++ assert "mlp_up_proj" not in plan["enabled_ops"] ++ assert "mlp_down_proj" not in plan["enabled_ops"] ++ ++ ++def test_dense_plan_still_executes_ffn_surrogate() -> None: ++ model_config = SimpleNamespace( ++ no_tensor_parallel=False, ++ embedding_dim=4096, ++ mlp_hidden_dim=1536, ++ num_q_heads=64, ++ num_kv_heads=4, ++ model_type="qwen2", ++ post_attn_norm=True, ++ is_moe=False, ++ is_step2_mini=False, ++ ) ++ plan = build_profiling_plan(model_config, 8, [8], [8], is_moe=False) ++ assert plan["ffn_sharded_enabled"] is True ++ assert "mlp_up_proj" in plan["enabled_ops"] diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_entrypoint.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_entrypoint.patch new file mode 100644 index 0000000..164af22 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_entrypoint.patch @@ -0,0 +1,127 @@ +diff --git a/frontier/profiling/moe/moe_vllm_kernel.py b/frontier/profiling/moe/moe_vllm_kernel.py +--- a/frontier/profiling/moe/moe_vllm_kernel.py ++++ b/frontier/profiling/moe/moe_vllm_kernel.py +@@ -33,6 +33,7 @@ try: + from vllm import _custom_ops as ops + # Import vLLM 0.10.x functions + from vllm.model_executor.layers.fused_moe.fused_moe import ( ++ fused_experts, + fused_moe_kernel, + invoke_fused_moe_kernel, + moe_align_block_size, +@@ -531,82 +532,58 @@ def profile_fused_moe_kernel( + block_shape=block_shape, + ) + +- sorted_token_ids, expert_ids, num_tokens_post_padded = moe_align_block_size( +- topk_ids, +- config["BLOCK_SIZE_M"], +- align_num_experts, +- expert_map=expert_map, +- ) +- +- output_dtype = base_dtype +- intermediate_cache1 = torch.empty( +- num_tokens, +- top_k, +- w1.shape[1], +- device=device, +- dtype=output_dtype, +- ) +- intermediate_cache2 = torch.empty( +- num_tokens * top_k, +- expert_hidden_dim_per_partition, +- device=device, +- dtype=output_dtype, +- ) +- intermediate_cache3 = torch.empty( +- num_tokens, +- top_k, +- hidden_dim, +- device=device, +- dtype=output_dtype, +- ) +- output = torch.empty( +- num_tokens, +- hidden_dim, +- device=device, +- dtype=output_dtype, +- ) +- + def _step() -> None: +- _run_fused_moe_iteration( +- A=A, ++ fused_experts( ++ hidden_states=A, + w1=w1, + w2=w2, +- intermediate_cache1=intermediate_cache1, +- intermediate_cache2=intermediate_cache2, +- intermediate_cache3=intermediate_cache3, +- output=output, + topk_weights=topk_weights, +- sorted_token_ids=sorted_token_ids, +- expert_ids=expert_ids, +- num_tokens_post_padded=num_tokens_post_padded, +- top_k=top_k, +- config=config, +- expert_hidden_dim_per_partition=expert_hidden_dim_per_partition, +- block_dims=block_dims, +- A_scale=None, ++ topk_ids=topk_ids, ++ inplace=True, ++ global_num_experts=align_num_experts, ++ expert_map=expert_map, ++ use_fp8_w8a8=use_fp8, ++ per_channel_quant=per_channel_quant, + w1_scale=w1_scale, + w2_scale=w2_scale, +- use_fp8=use_fp8, +- per_channel_quant=per_channel_quant, + block_shape=block_shape, + ) + ++ def _alignment_step() -> None: ++ moe_align_block_size( ++ topk_ids, ++ config["BLOCK_SIZE_M"], ++ align_num_experts, ++ expert_map=expert_map, ++ ) ++ + for _ in range(warmup_steps): + _step() + torch.cuda.synchronize() + + if profile_method == "record_function": +- return _collect_record_function_stats( +- step_fn=_step, +- active_steps=active_steps, +- output_dir=output_dir, +- operation_name="moe_grouped_gemm", ++ raise ValueError( ++ "Serving-entrypoint MoE profiling requires cuda_event so the local " ++ "alignment component can be subtracted without double counting." + ) + +- return _collect_cuda_event_stats( ++ serving_stats = _collect_cuda_event_stats( + step_fn=_step, + active_steps=active_steps, + ) ++ alignment_stats = _collect_cuda_event_stats( ++ step_fn=_alignment_step, ++ active_steps=active_steps, ++ ) ++ return { ++ "min": max(0.0, serving_stats["min"] - alignment_stats["max"]), ++ "max": max(0.0, serving_stats["max"] - alignment_stats["min"]), ++ "mean": max(0.0, serving_stats["mean"] - alignment_stats["mean"]), ++ "median": max(0.0, serving_stats["median"] - alignment_stats["median"]), ++ "std": ( ++ serving_stats["std"] ** 2 + alignment_stats["std"] ** 2 ++ ) ** 0.5, ++ } + + + def generate_expert_weights( diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_path.patch b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_path.patch new file mode 100644 index 0000000..6702a90 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_moe_serving_path.patch @@ -0,0 +1,216 @@ +diff --git a/frontier/profiling/moe/moe_vllm_kernel.py b/frontier/profiling/moe/moe_vllm_kernel.py +--- a/frontier/profiling/moe/moe_vllm_kernel.py ++++ b/frontier/profiling/moe/moe_vllm_kernel.py +@@ -34,6 +34,7 @@ try: + import vllm + VLLM_VERSION = vllm.__version__ + ++ from vllm import _custom_ops as ops + # Import vLLM 0.10.x functions + from vllm.model_executor.layers.fused_moe.fused_moe import ( + fused_moe_kernel, +@@ -232,7 +233,7 @@ def _invoke_kernel( + """ + # Determine compute_type - for FP8, we accumulate in FP16/BF16 + if use_fp8: +- compute_type = tl.float16 # FP8 accumulates in FP16 ++ compute_type = tl.bfloat16 + else: + dtype = A.dtype + if dtype == torch.bfloat16: +@@ -275,7 +276,9 @@ def _run_fused_moe_iteration( + w1: torch.Tensor, + w2: torch.Tensor, + intermediate_cache1: torch.Tensor, + intermediate_cache2: torch.Tensor, ++ intermediate_cache3: torch.Tensor, ++ output: torch.Tensor, + topk_weights: torch.Tensor, + sorted_token_ids: torch.Tensor, + expert_ids: torch.Tensor, +@@ -292,8 +295,17 @@ def _run_fused_moe_iteration( + per_channel_quant: bool = False, + block_shape: Optional[List[int]] = None, + ) -> None: ++ first_input = A ++ first_A_scale = A_scale ++ if use_fp8: ++ group_size = block_dims[1] if block_dims else 128 ++ first_input, first_A_scale = quantize_activations_to_fp8( ++ A, ++ group_size=group_size, ++ ) ++ + _invoke_kernel( +- A=A.contiguous(), ++ A=first_input.contiguous(), + B=w1.contiguous(), + C=intermediate_cache1.contiguous(), + topk_weights=topk_weights.contiguous(), +@@ -305,15 +316,18 @@ def _run_fused_moe_iteration( + mul_routed_weight=False, + top_k=top_k, + config=config, +- A_scale=A_scale, ++ A_scale=first_A_scale, + B_scale=w1_scale, + use_fp8=use_fp8, + per_channel_quant=per_channel_quant, + block_shape=block_shape, + ) + +- intermediate_cache1_flat = intermediate_cache1.view(-1, intermediate_cache1.shape[-1]) +- intermediate_cache2_input = intermediate_cache1_flat[:, :expert_hidden_dim_per_partition].contiguous() ++ torch.ops._C.silu_and_mul( ++ intermediate_cache2, ++ intermediate_cache1.view(-1, intermediate_cache1.shape[-1]), ++ ) ++ second_input = intermediate_cache2 + + intermediate_A_scale = None + if use_fp8: +@@ -321,13 +334,13 @@ def _run_fused_moe_iteration( + group_size = block_dims[1] if block_dims else 128 +- intermediate_cache2_input, intermediate_A_scale = quantize_activations_to_fp8( +- intermediate_cache2_input, ++ second_input, intermediate_A_scale = quantize_activations_to_fp8( ++ intermediate_cache2, + group_size=group_size, + ) + + _invoke_kernel( +- A=intermediate_cache2_input, ++ A=second_input, + B=w2.contiguous(), +- C=intermediate_cache2.contiguous(), ++ C=intermediate_cache3.contiguous(), + topk_weights=topk_weights.contiguous(), + sorted_token_ids=sorted_token_ids.contiguous(), + expert_ids=expert_ids.contiguous(), +@@ -335,4 +350,6 @@ def _run_fused_moe_iteration( + ) ++ ++ ops.moe_sum(intermediate_cache3, output) + + + def _collect_cuda_event_stats(step_fn, active_steps: int) -> Dict: +@@ -493,6 +508,5 @@ def profile_fused_moe_kernel( + w1_scale = None + w2_scale = None +- A_scale = None + + block_dims = _validate_block_shape(block_shape) + if use_fp8: +@@ -509,10 +521,8 @@ def profile_fused_moe_kernel( + per_channel=per_channel_quant, + block_shape=block_shape, + ) +- group_size = block_dims[1] if block_dims else 128 +- A, A_scale = quantize_activations_to_fp8(A, group_size=group_size) + +- config_dtype = get_config_dtype_str(base_dtype) ++ config_dtype = get_config_dtype_str(base_dtype, use_fp8_w8a8=use_fp8) + config = try_get_optimal_moe_config( + w1_shape=w1.shape, + w2_shape=w2.shape, +@@ -535,13 +544,25 @@ def profile_fused_moe_kernel( + device=device, + dtype=output_dtype, + ) + intermediate_cache2 = torch.empty( +- num_tokens, +- top_k, +- hidden_dim, ++ num_tokens * top_k, ++ expert_hidden_dim_per_partition, ++ device=device, ++ dtype=output_dtype, ++ ) ++ intermediate_cache3 = torch.empty( ++ num_tokens, ++ top_k, ++ hidden_dim, + device=device, + dtype=output_dtype, + ) ++ output = torch.empty( ++ num_tokens, ++ hidden_dim, ++ device=device, ++ dtype=output_dtype, ++ ) + + def _step() -> None: + _run_fused_moe_iteration( +@@ -552,6 +571,8 @@ def profile_fused_moe_kernel( + w2=w2, + intermediate_cache1=intermediate_cache1, + intermediate_cache2=intermediate_cache2, ++ intermediate_cache3=intermediate_cache3, ++ output=output, + topk_weights=topk_weights, + sorted_token_ids=sorted_token_ids, + expert_ids=expert_ids, +@@ -562,6 +583,6 @@ def profile_fused_moe_kernel( + expert_hidden_dim_per_partition=expert_hidden_dim_per_partition, + block_dims=block_dims, +- A_scale=A_scale, ++ A_scale=None, + w1_scale=w1_scale, + w2_scale=w2_scale, + use_fp8=use_fp8, +diff --git a/frontier/profiling/moe/moe_impl.py b/frontier/profiling/moe/moe_impl.py +--- a/frontier/profiling/moe/moe_impl.py ++++ b/frontier/profiling/moe/moe_impl.py +@@ -245,10 +245,12 @@ class MoETokenShuffler(nn.Module): + def __init__( + self, + num_experts: int, + router_topk: int, + hidden_dim: int, + expert_hidden_dim: int, + dtype: torch.dtype, + use_gated: bool, + num_local_experts: Optional[int] = None, ++ use_fp8: bool = False, ++ block_shape: Optional[list[int]] = None, + ): +@@ -264,6 +266,8 @@ class MoETokenShuffler(nn.Module): + self.router_topk = router_topk + self.hidden_dim = hidden_dim + self.expert_hidden_dim = expert_hidden_dim + self.dtype = dtype + self.use_gated = use_gated ++ self.use_fp8 = use_fp8 ++ self.block_shape = block_shape + self._block_size_cache = {} +@@ -325,9 +329,12 @@ class MoETokenShuffler(nn.Module): +- config_dtype = get_config_dtype_str(dtype=self.dtype) ++ config_dtype = get_config_dtype_str( ++ dtype=self.dtype, ++ use_fp8_w8a8=self.use_fp8, ++ ) + config = try_get_optimal_moe_config( + w1_shape=w1_shape, + w2_shape=w2_shape, + top_k=self.router_topk, + dtype=config_dtype, + M=num_tokens, +- block_shape=None, ++ block_shape=self.block_shape, + ) +diff --git a/frontier/profiling/moe/moe_wrapper.py b/frontier/profiling/moe/moe_wrapper.py +--- a/frontier/profiling/moe/moe_wrapper.py ++++ b/frontier/profiling/moe/moe_wrapper.py +@@ -149,9 +149,11 @@ class MoEWrapper: + self.shuffler = MoETokenShuffler( + num_experts=self.num_experts, + num_local_experts=self.num_experts_per_device, + router_topk=self.router_topk, + hidden_dim=self.hidden_dim, + expert_hidden_dim=self.expert_hidden_dim, + dtype=self._dtype, + use_gated=self.use_gated, ++ use_fp8=self.use_fp8, ++ block_shape=self.block_shape, + ).to(dtype=self._dtype).cuda().eval() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_prefill_grid.py b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_prefill_grid.py new file mode 100644 index 0000000..b42bfa0 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/frontier_prefill_grid.py @@ -0,0 +1,740 @@ +#!/usr/bin/env python3 +"""Freeze a profile-only Frontier ranking for the Qwen235B prefill grid.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import os +import subprocess +import sys +import time +from dataclasses import asdict, dataclass +from pathlib import Path +from typing import Any, Iterable + + +SCHEMA = "frontier-qwen235b-prefill-grid-v1" +MODEL = "Qwen3-235B-A22B-FP8" +TRACE_SHA256 = "f878e9af18f94dcfaced94a8e1e6b20a2f7d97d64aa862448025660dbbd965b2" +PROFILE_RELATIVE = Path("compute/h20") / MODEL +NETWORK_RELATIVE = Path("network/h20_nccl/all_reduce.csv") +SEARCH_LOW = 0.0 +SEARCH_HIGH = 0.125 +SEARCH_PROBES = 6 +WINDOW_DURATION_S = 600.0 +MAX_INPUT_TOKENS = 32768 +MAX_MODEL_TOKENS = 40960 +BLOCK_SIZE_TOKENS = 16 +MOE_ROUTING_SEED = 42 + + +@dataclass(frozen=True) +class GridConfig: + tp: int + mns: int + mbt: int + moe_tp: int + moe_ep: int + num_gpu_blocks: int + + @property + def name(self) -> str: + return f"tp{self.tp}_mns{self.mns}_mbt{self.mbt}" + + @property + def gpu_count(self) -> int: + return self.tp + + +GRID = tuple( + GridConfig( + tp=tp, + mns=mns, + mbt=mbt, + moe_tp=4 if tp == 4 else 1, + moe_ep=1 if tp == 4 else 8, + # Measured by community vLLM on the same checkpoint/runtime. The real + # grid will use --num-gpu-blocks-override with these same values. + num_gpu_blocks=26101 if tp == 4 else 62351, + ) + for tp in (4, 8) + for mns in (64, 128) + for mbt in (8192, 16384) +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def order_hash(values: Iterable[object]) -> str: + payload = "\n".join(str(value) for value in values).encode() + return hashlib.sha256(payload).hexdigest() + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def sha256_bytes(payload: bytes) -> str: + return hashlib.sha256(payload).hexdigest() + + +def frontier_source_fingerprint( + source: Path, declared_commit: str +) -> dict[str, Any]: + model_config = source / "data" / "config" / "models" / f"{MODEL}.json" + if not model_config.is_file(): + raise FileNotFoundError(f"Frontier model config is missing: {model_config}") + device_config = source / "data" / "config" / "device" / "h20.json" + files = sorted((source / "frontier").glob("**/*.py")) + [ + model_config, + device_config, + ] + missing = [str(path) for path in files if not path.is_file()] + if missing: + raise FileNotFoundError(f"Frontier fingerprint inputs are missing: {missing}") + digest = hashlib.sha256() + for path in files: + relative = path.relative_to(source).as_posix() + digest.update(relative.encode() + b"\0") + digest.update(bytes.fromhex(sha256(path))) + fingerprint = { + "declared_upstream_commit": declared_commit, + "python_and_config_tree_sha256": digest.hexdigest(), + "fingerprinted_file_count": len(files), + "model_config": { + "path": str(model_config.resolve()), + "sha256": sha256(model_config), + }, + } + git_dir = source / ".git" + if git_dir.exists(): + commit = subprocess.run( + ["git", "rev-parse", "HEAD"], + cwd=source, + check=True, + stdout=subprocess.PIPE, + text=True, + ).stdout.strip() + status = subprocess.run( + ["git", "status", "--porcelain=v1"], + cwd=source, + check=True, + stdout=subprocess.PIPE, + text=True, + ).stdout + diff = subprocess.run( + ["git", "diff", "--binary", "HEAD"], + cwd=source, + check=True, + stdout=subprocess.PIPE, + ).stdout + fingerprint.update( + { + "git_commit": commit, + "status_porcelain": status.splitlines(), + "tracked_diff_sha256": sha256_bytes(diff), + } + ) + else: + fingerprint["git_metadata"] = "absent_source_snapshot" + return fingerprint + + +def anchor_key(anchor: float) -> str: + return f"{anchor:.12f}" + + +def anchor_filename(anchor: float) -> str: + return f"u_{anchor_key(anchor).replace('.', 'p')}.csv" + + +def binary_search_lattice() -> list[float]: + intervals = [(SEARCH_LOW, SEARCH_HIGH)] + anchors: list[float] = [] + for _ in range(SEARCH_PROBES): + next_intervals = [] + for low, high in intervals: + midpoint = (low + high) / 2.0 + anchors.append(midpoint) + next_intervals.extend(((low, midpoint), (midpoint, high))) + intervals = next_intervals + return sorted(set(anchors)) + + +def ttft_slo_ms(input_tokens: int) -> float: + return 1000.0 if input_tokens <= 8191 else 2000.0 + + +def load_trace_rows(path: Path) -> list[dict[str, Any]]: + rows = [] + with path.open(encoding="utf-8") as source: + for source_index, line in enumerate(source): + if not line.strip(): + continue + raw = json.loads(line) + input_tokens = int(raw["input_length"]) + if not 0 <= input_tokens <= MAX_INPUT_TOKENS: + continue + timestamp = float(raw["timestamp"]) + sampling_u = float(raw["sampling_u"]) + rows.append( + { + "arrived_at": timestamp, + "num_prefill_tokens": input_tokens, + "num_decode_tokens": 1, + "source_row_index": source_index, + "source_request_id": str( + raw.get("request_id") or raw.get("id") or source_index + ), + "sampling_u": sampling_u, + "slo_ttft_ms": ttft_slo_ms(input_tokens), + } + ) + # This reproduces AITuner's stable arrival-only sort. Tied timestamps retain + # source order; sampling_u must not become a secondary key. + rows.sort(key=lambda row: row["arrived_at"]) + return rows + + +def prepare_traces(trace: Path, output_root: Path, expected_sha256: str) -> Path: + actual_sha256 = sha256(trace) + if expected_sha256 and actual_sha256 != expected_sha256: + raise ValueError( + f"trace SHA256 mismatch: expected={expected_sha256}, actual={actual_sha256}" + ) + rows = load_trace_rows(trace) + if not rows: + raise ValueError("no trace rows remain after input-length filtering") + if any( + rows[index]["arrived_at"] > rows[index + 1]["arrived_at"] + for index in range(len(rows) - 1) + ): + raise ValueError("materialized trace is not monotonic by arrival") + + trace_dir = output_root / "traces" + trace_dir.mkdir(parents=True, exist_ok=True) + fields = [ + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "source_row_index", + "source_request_id", + "sampling_u", + "slo_ttft_ms", + ] + anchors: dict[str, Any] = {} + for anchor in binary_search_lattice(): + selected = [row for row in rows if row["sampling_u"] <= anchor] + target = trace_dir / anchor_filename(anchor) + with target.open("w", encoding="utf-8", newline="") as output: + writer = csv.DictWriter(output, fieldnames=fields) + writer.writeheader() + writer.writerows(selected) + key = anchor_key(anchor) + anchors[key] = { + "anchor": anchor, + "path": str(target.resolve()), + "sha256": sha256(target), + "request_count": len(selected), + "request_rate": len(selected) / WINDOW_DURATION_S, + "source_row_order_sha256": order_hash( + row["source_row_index"] for row in selected + ), + "arrival_order_sha256": order_hash( + f"{row['arrived_at']:.12f}" for row in selected + ), + "input_length_order_sha256": order_hash( + row["num_prefill_tokens"] for row in selected + ), + } + + manifest = { + "schema": SCHEMA, + "source": { + "path": str(trace.resolve()), + "sha256": actual_sha256, + "filtered_request_count": len(rows), + }, + "selection_contract": { + "input_tokens": [0, MAX_INPUT_TOKENS], + "completion_tokens_override": 1, + "stable_sort_key": "arrived_at_only", + "selection": "sampling_u <= anchor", + "window_duration_s": WINDOW_DURATION_S, + "search_low": SEARCH_LOW, + "search_high": SEARCH_HIGH, + "search_probes": SEARCH_PROBES, + "lattice_anchor_count": len(anchors), + "ttft_slo_ms": {"input_le_8191": 1000, "otherwise": 2000}, + "target_pass_rate": 0.95, + }, + "anchors": anchors, + } + manifest_path = output_root / "trace_manifest.json" + write_json(manifest_path, manifest) + return manifest_path + + +def resolve_profile_paths(profile_root: Path) -> dict[str, Path]: + paths = { + "linear": profile_root / PROFILE_RELATIVE / "linear_op.csv", + "attention": profile_root / PROFILE_RELATIVE / "attention.csv", + "moe": profile_root / PROFILE_RELATIVE / "moe.csv", + "all_reduce": profile_root / NETWORK_RELATIVE, + "manifest": profile_root / "profile_manifest.json", + } + missing = [f"{name}={path}" for name, path in paths.items() if not path.is_file()] + if missing: + raise FileNotFoundError("missing profile inputs: " + ", ".join(missing)) + return paths + + +def build_command( + *, + python: Path, + frontier_source: Path, + profile_root: Path, + profile_paths: dict[str, Path], + trace: Path, + config: GridConfig, + probe_dir: Path, + run_id: str, + cache_root: Path, +) -> list[str]: + return [ + str(python), + "-m", + "frontier.main", + "--simulation_mode", + "offline", + # Frontier's offline default rewrites every request arrival to time 0. + # Preserve the trace arrival process so the simulator and serving run + # exercise the same queueing workload. + "--offline_use_generated_request_arrivals", + "--sys_arch", + "co-location", + "--cluster_config_num_replicas", + "1", + "--replica_config_model_name", + MODEL, + "--replica_config_attn_tensor_parallel_size", + str(config.tp), + "--replica_config_attn_data_parallel_size", + "1", + "--replica_config_moe_tensor_parallel_size", + str(config.moe_tp), + "--replica_config_moe_expert_parallel_size", + str(config.moe_ep), + "--replica_config_total_expert_num", + "128", + "--replica_config_router_topk", + "8", + "--replica_config_moe_routing_mode", + "simulation", + "--replica_config_moe_routing_seed", + str(MOE_ROUTING_SEED), + "--replica_config_num_pipeline_stages", + "1", + "--replica_config_device", + "h20", + "--replica_config_network_device", + "h20_dgx", + "--cc_backend_config_type", + "vidur", + "--vidur_cc_backend_config_profiling_data_dir", + str(profile_root), + "--vidur_cc_backend_config_cache_dir", + str(cache_root / "collectives"), + "--vidur_cc_backend_config_all_reduce_input_file", + str(profile_paths["all_reduce"]), + "--replica_scheduler_config_type", + "vllm_v1", + "--decode_cuda_graph_mode", + "none", + "--vllm_v1_scheduler_config_batch_size_cap", + str(config.mns), + "--vllm_v1_scheduler_config_block_size", + str(BLOCK_SIZE_TOKENS), + "--vllm_v1_scheduler_config_num_blocks", + str(config.num_gpu_blocks), + "--vllm_v1_scheduler_config_num_blocks_mode", + "explicit", + "--vllm_v1_scheduler_config_max_tokens_in_batch", + str(config.mbt), + "--vllm_v1_scheduler_config_enable_chunked_prefill", + "--no-vllm_v1_scheduler_config_enable_prefix_caching", + "--request_generator_config_type", + "trace_replay", + "--trace_request_generator_config_trace_file", + str(trace), + "--trace_request_generator_config_time_scale_factor", + "1", + "--trace_request_generator_config_prefill_scale_factor", + "1", + "--trace_request_generator_config_decode_scale_factor", + "1", + "--trace_request_generator_config_max_tokens", + str(MAX_MODEL_TOKENS), + "--no-random_forrest_execution_time_predictor_config_enable_dummy_mode", + "--random_forrest_execution_time_predictor_config_linear_op_input_file", + str(profile_paths["linear"]), + "--random_forrest_execution_time_predictor_config_atten_input_file", + str(profile_paths["attention"]), + "--random_forrest_execution_time_predictor_config_moe_input_file", + str(profile_paths["moe"]), + "--random_forrest_execution_time_predictor_config_all_reduce_input_file", + str(profile_paths["all_reduce"]), + "--random_forrest_execution_time_predictor_config_prediction_max_prefill_chunk_size", + "16384", + "--random_forrest_execution_time_predictor_config_prediction_max_tokens_per_request", + str(MAX_INPUT_TOKENS + 1), + "--random_forrest_execution_time_predictor_config_prediction_max_batch_size", + "128", + "--random_forrest_execution_time_predictor_config_skip_cpu_overhead_modeling", + "--metrics_config_cache_dir", + str(cache_root / "execution"), + "--metrics_config_output_dir", + str(probe_dir / "metrics"), + "--metrics_config_run_id", + run_id, + "--metrics_config_write_metrics", + "--metrics_config_store_request_metrics", + "--no-metrics_config_store_plots", + "--no-metrics_config_enable_chrome_trace", + "--no-metrics_config_write_json_trace", + ] + + +def score_request_metrics(trace: Path, request_metrics: Path) -> dict[str, Any]: + with trace.open(encoding="utf-8", newline="") as source: + trace_rows = list(csv.DictReader(source)) + with request_metrics.open(encoding="utf-8", newline="") as source: + metric_rows = list(csv.DictReader(source)) + if len(metric_rows) != len(trace_rows): + raise ValueError( + f"request count mismatch: trace={len(trace_rows)}, metrics={len(metric_rows)}" + ) + metrics_by_id = {int(row["Request Id"]): row for row in metric_rows} + expected_ids = set(range(len(trace_rows))) + if set(metrics_by_id) != expected_ids: + raise ValueError("Frontier Request Ids do not match trace row positions") + + passed = 0 + ttfts = [] + for request_id, trace_row in enumerate(trace_rows): + ttft = float(metrics_by_id[request_id]["ttft"]) + threshold = float(trace_row["slo_ttft_ms"]) + ttfts.append(ttft) + passed += int(ttft <= threshold) + count = len(trace_rows) + pass_rate = passed / count if count else 0.0 + ordered = sorted(ttfts) + + def percentile(fraction: float) -> float | None: + if not ordered: + return None + index = round((len(ordered) - 1) * fraction) + return ordered[index] + + return { + "request_count": count, + "passed_request_count": passed, + "slo_pass_rate": pass_rate, + "feasible": pass_rate >= 0.95, + "ttft_ms": { + "min": min(ordered) if ordered else None, + "p50": percentile(0.50), + "p95": percentile(0.95), + "p99": percentile(0.99), + "max": max(ordered) if ordered else None, + }, + } + + +def find_request_metrics(probe_dir: Path) -> Path: + candidates = list((probe_dir / "metrics").glob("**/request_metrics.csv")) + if len(candidates) != 1: + raise ValueError( + f"expected one request_metrics.csv under {probe_dir}, got {candidates}" + ) + return candidates[0] + + +def run_probe( + *, + python: Path, + frontier_source: Path, + profile_root: Path, + profile_paths: dict[str, Path], + trace_record: dict[str, Any], + config: GridConfig, + probe_index: int, + output_root: Path, + cache_root: Path, +) -> dict[str, Any]: + anchor = float(trace_record["anchor"]) + probe_dir = output_root / "runs" / config.name / f"probe_{probe_index}_{anchor_filename(anchor)[:-4]}" + result_path = probe_dir / "result.json" + if result_path.is_file(): + result = json.loads(result_path.read_text()) + if result.get("status") == "completed": + return result + probe_dir.mkdir(parents=True, exist_ok=True) + trace = Path(trace_record["path"]) + run_id = f"{config.name}_probe{probe_index}_{anchor_filename(anchor)[:-4]}" + command = build_command( + python=python, + frontier_source=frontier_source, + profile_root=profile_root, + profile_paths=profile_paths, + trace=trace, + config=config, + probe_dir=probe_dir, + run_id=run_id, + cache_root=cache_root, + ) + write_json(probe_dir / "command.json", command) + environment = os.environ.copy() + environment.update( + { + "PYTHONPATH": str(frontier_source), + "WANDB_DISABLED": "true", + "VIDUR_DISABLE_WANDB": "1", + } + ) + started = time.time() + with (probe_dir / "stdout.log").open("w", encoding="utf-8") as output: + completed = subprocess.run( + command, + cwd=frontier_source, + env=environment, + stdout=output, + stderr=subprocess.STDOUT, + check=False, + ) + elapsed = time.time() - started + if completed.returncode != 0: + result = { + "status": "failed", + "returncode": completed.returncode, + "elapsed_seconds": elapsed, + "config": asdict(config), + "anchor": anchor, + "trace_sha256": trace_record["sha256"], + } + write_json(result_path, result) + raise RuntimeError(f"Frontier probe failed: {config.name}, anchor={anchor}") + + request_metrics = find_request_metrics(probe_dir) + score = score_request_metrics(trace, request_metrics) + result = { + "status": "completed", + "elapsed_seconds": elapsed, + "config": asdict(config), + "anchor": anchor, + "trace_path": str(trace), + "trace_sha256": trace_record["sha256"], + "request_metrics_path": str(request_metrics), + "request_metrics_sha256": sha256(request_metrics), + "request_rate": score["request_count"] / WINDOW_DURATION_S, + "request_rate_per_gpu": score["request_count"] + / WINDOW_DURATION_S + / config.gpu_count, + **score, + } + write_json(result_path, result) + return result + + +def selected_configs(names: list[str] | None) -> list[GridConfig]: + if not names: + return list(GRID) + by_name = {config.name: config for config in GRID} + unknown = sorted(set(names) - set(by_name)) + if unknown: + raise ValueError(f"unknown configs: {unknown}; available={sorted(by_name)}") + return [by_name[name] for name in names] + + +def run_grid(args: argparse.Namespace) -> None: + trace_manifest = json.loads(args.trace_manifest.read_text()) + if trace_manifest.get("schema") != SCHEMA: + raise ValueError(f"unexpected trace manifest schema: {trace_manifest.get('schema')}") + profile_paths = resolve_profile_paths(args.profile_root) + args.output_root.mkdir(parents=True, exist_ok=True) + cache_root = args.output_root / "cache" + configs = selected_configs(args.config) + run_manifest = { + "schema": SCHEMA, + "frontier": { + "source": str(args.frontier_source.resolve()), + "python": str(args.python.resolve()), + "fingerprint": frontier_source_fingerprint( + args.frontier_source, args.frontier_commit + ), + }, + "trace_manifest": { + "path": str(args.trace_manifest.resolve()), + "sha256": sha256(args.trace_manifest), + }, + "profiles": { + name: {"path": str(path.resolve()), "sha256": sha256(path)} + for name, path in profile_paths.items() + }, + "kv_capacity_evidence": { + "tp4": { + "path": str(args.tp4_capacity_artifact.resolve()), + "sha256": sha256(args.tp4_capacity_artifact), + "num_gpu_blocks": 26101, + }, + "tp8": { + "path": str(args.tp8_capacity_artifact.resolve()), + "sha256": sha256(args.tp8_capacity_artifact), + "num_gpu_blocks": 62351, + }, + }, + "contract": { + "metric": "maximum SLO-feasible request_rate_per_gpu", + "target_pass_rate": 0.95, + "cpu_overhead_modeling": "skipped_no_community_vllm_native_records", + "end_to_end_calibration": False, + "moe_routing_mode": "simulation", + "moe_routing_seed": MOE_ROUTING_SEED, + "kv_blocks_source": "community_vllm_measured_and_fixed_per_topology", + "network_device": "h20_dgx", + }, + "configs": [asdict(config) | {"name": config.name} for config in configs], + } + write_json(args.output_root / "run_manifest.json", run_manifest) + + all_results = {} + for config in configs: + low = SEARCH_LOW + high = SEARCH_HIGH + probes = [] + best = None + for probe_index in range(SEARCH_PROBES): + anchor = (low + high) / 2.0 + record = trace_manifest["anchors"].get(anchor_key(anchor)) + if record is None: + raise ValueError(f"trace manifest lacks anchor {anchor_key(anchor)}") + result = run_probe( + python=args.python, + frontier_source=args.frontier_source, + profile_root=args.profile_root, + profile_paths=profile_paths, + trace_record=record, + config=config, + probe_index=probe_index, + output_root=args.output_root, + cache_root=cache_root, + ) + probes.append(result) + print( + json.dumps( + { + "config": config.name, + "probe": probe_index, + "anchor": anchor, + "pass_rate": result["slo_pass_rate"], + "feasible": result["feasible"], + "elapsed_seconds": result["elapsed_seconds"], + }, + sort_keys=True, + ), + flush=True, + ) + if result["feasible"]: + low = anchor + best = result + else: + high = anchor + summary = { + "config": asdict(config) | {"name": config.name}, + "probes": probes, + "capacity_interval": [low, high], + "best_feasible": best, + } + write_json(args.output_root / "results" / f"{config.name}.json", summary) + all_results[config.name] = summary + + ranked = sorted( + all_results.values(), + key=lambda item: ( + -( + item["best_feasible"]["request_rate_per_gpu"] + if item["best_feasible"] is not None + else -1.0 + ), + item["config"]["name"], + ), + ) + freeze = { + "schema": SCHEMA, + "run_manifest_sha256": sha256(args.output_root / "run_manifest.json"), + "ranking": [ + { + "rank": index + 1, + "config": item["config"], + "capacity_interval_sampling_u": item["capacity_interval"], + "best_feasible": item["best_feasible"], + } + for index, item in enumerate(ranked) + ], + } + write_json(args.output_root / "frontier_ranking_frozen.json", freeze) + print(json.dumps(freeze, indent=2), flush=True) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + subparsers = parser.add_subparsers(dest="command", required=True) + prepare = subparsers.add_parser("prepare") + prepare.add_argument("--trace", type=Path, required=True) + prepare.add_argument("--output-root", type=Path, required=True) + prepare.add_argument("--expected-trace-sha256", default=TRACE_SHA256) + + run = subparsers.add_parser("run") + run.add_argument("--frontier-source", type=Path, required=True) + run.add_argument( + "--frontier-commit", + default="d9cfeb6d8791fbf2f295dd9744c56a666171776e", + ) + run.add_argument("--python", type=Path, required=True) + run.add_argument("--profile-root", type=Path, required=True) + run.add_argument("--trace-manifest", type=Path, required=True) + run.add_argument("--output-root", type=Path, required=True) + run.add_argument("--tp4-capacity-artifact", type=Path, required=True) + run.add_argument("--tp8-capacity-artifact", type=Path, required=True) + run.add_argument("--config", action="append") + return parser.parse_args() + + +def main() -> None: + args = parse_args() + if args.command == "prepare": + manifest = prepare_traces( + args.trace, args.output_root, args.expected_trace_sha256 + ) + print(manifest) + return + if args.command == "run": + run_grid(args) + return + raise AssertionError(args.command) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort.toml b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort.toml new file mode 100644 index 0000000..afd140d --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort.toml @@ -0,0 +1,10 @@ +# Immutable one-job queue for the blind Qwen235B fixed-cohort rank experiment. +version = 1 + +[[jobs]] +name = "qwen235b-fixed-cohort-community-v1-blind-20260715" +gpus = 8 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd ../../.. && /tmp/wjh-frontier-vllm0102-smoke/.venv/bin/python runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py run-community --manifest /home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v1-blind/run_manifest.json" +env = { PYTHONPATH = "/home/admin/cpfs/wjh/aituner/aituner-frontier-best-effort-20260715/src" } diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_repeat.toml b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_repeat.toml new file mode 100644 index 0000000..0655160 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_repeat.toml @@ -0,0 +1,10 @@ +# Reverse-order fresh-server boundary repeat for the frozen blind surface. +version = 1 + +[[jobs]] +name = "qwen235b-fixed-cohort-community-v2-boundary-reverse-20260715" +gpus = 8 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd ../../.. && /tmp/wjh-frontier-vllm0102-smoke/.venv/bin/python runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py run-community --manifest /home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v2-boundary-reverse/run_manifest.json" +env = { PYTHONPATH = "/home/admin/cpfs/wjh/aituner/aituner-frontier-best-effort-20260715/src" } diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_targeted.toml b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_targeted.toml new file mode 100644 index 0000000..74f2c95 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/jobs_fixed_cohort_targeted.toml @@ -0,0 +1,10 @@ +# Third fresh-server trial for primary-SLO boundaries disputed after refinement. +version = 1 + +[[jobs]] +name = "qwen235b-fixed-cohort-community-v3-primary-disputed-20260715" +gpus = 8 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd ../../.. && /tmp/wjh-frontier-vllm0102-smoke/.venv/bin/python runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_rank.py run-community --manifest /home/admin/cpfs/wjh/frontier-community-qwen235-smoke-20260715/rankable-fixed-cohort-v1/community-v3-primary-disputed/run_manifest.json" +env = { PYTHONPATH = "/home/admin/cpfs/wjh/aituner/aituner-frontier-best-effort-20260715/src" } diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/moe_path_parity.py b/runs/frontier-multicase-sufficiency-v0/best_effort/moe_path_parity.py new file mode 100644 index 0000000..b8c1a66 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/moe_path_parity.py @@ -0,0 +1,220 @@ +#!/usr/bin/env python3 +"""Compare patched Frontier MoE decomposition with vLLM's serving path.""" + +from __future__ import annotations + +import argparse +import json +from pathlib import Path + +import torch + +from frontier.profiling.moe.moe_vllm_kernel import ( + profile_fused_moe_kernel, + quantize_weights_to_fp8, +) +from vllm.model_executor.layers.fused_moe.fused_moe import ( + fused_experts, + get_config_dtype_str, + moe_align_block_size, + try_get_optimal_moe_config, +) + + +def _measure(step, warmup: int, active: int) -> dict[str, float]: + for _ in range(warmup): + step() + torch.cuda.synchronize() + + samples = [] + for _ in range(active): + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + step() + end.record() + torch.cuda.synchronize() + samples.append(start.elapsed_time(end)) + values = torch.tensor(samples) + return { + "min": float(values.min()), + "median": float(values.median()), + "mean": float(values.mean()), + "max": float(values.max()), + "std": float(values.std()), + } + + +def _routing(num_tokens: int, top_k: int, num_experts: int, seed: int): + generator = torch.Generator(device="cuda") + generator.manual_seed(seed) + topk_ids = torch.randint( + num_experts, + (num_tokens, top_k), + generator=generator, + device="cuda", + dtype=torch.int64, + ) + topk_weights = torch.rand( + (num_tokens, top_k), + generator=generator, + device="cuda", + dtype=torch.float32, + ) + topk_weights /= topk_weights.sum(dim=-1, keepdim=True) + return topk_weights, topk_ids + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--tokens", nargs="+", type=int, default=[16, 256, 1024]) + parser.add_argument("--tp", type=int, default=4) + parser.add_argument("--ep", type=int, default=1) + parser.add_argument("--warmup", type=int, default=2) + parser.add_argument("--active", type=int, default=20) + parser.add_argument("--seed", type=int, default=20260715) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + + hidden_dim = 4096 + expert_hidden_dim = 1536 + global_num_experts = 128 + top_k = 8 + block_shape = [128, 128] + if global_num_experts % args.ep: + raise ValueError("EP must divide 128 experts") + if expert_hidden_dim % args.tp: + raise ValueError("TP must divide the expert intermediate dimension") + + local_num_experts = global_num_experts // args.ep + local_intermediate = expert_hidden_dim // args.tp + device = torch.device("cuda") + + torch.manual_seed(args.seed) + w1_bf16 = torch.randn( + local_num_experts, + 2 * local_intermediate, + hidden_dim, + dtype=torch.bfloat16, + device=device, + ) + w2_bf16 = torch.randn( + local_num_experts, + hidden_dim, + local_intermediate, + dtype=torch.bfloat16, + device=device, + ) + w1, w1_scale = quantize_weights_to_fp8(w1_bf16, block_shape=block_shape) + w2, w2_scale = quantize_weights_to_fp8(w2_bf16, block_shape=block_shape) + del w1_bf16, w2_bf16 + torch.cuda.empty_cache() + + rows = [] + for index, num_tokens in enumerate(args.tokens): + topk_weights, topk_ids = _routing( + num_tokens, + top_k, + global_num_experts, + args.seed + index, + ) + hidden_states = torch.randn( + num_tokens, + hidden_dim, + dtype=torch.bfloat16, + device=device, + ) + + frontier_grouped = profile_fused_moe_kernel( + num_tokens=num_tokens, + num_experts=local_num_experts, + hidden_dim=hidden_dim, + expert_hidden_dim=expert_hidden_dim, + top_k=top_k, + topk_weights=topk_weights, + topk_ids=topk_ids, + tensor_parallel_size=args.tp, + dtype=torch.bfloat16, + warmup_steps=args.warmup, + active_steps=args.active, + use_fp8=True, + per_channel_quant=False, + block_shape=block_shape, + global_num_experts=global_num_experts, + ) + + config = try_get_optimal_moe_config( + w1_shape=w1.shape, + w2_shape=w2.shape, + top_k=top_k, + dtype=get_config_dtype_str( + torch.bfloat16, + use_fp8_w8a8=True, + ), + M=num_tokens, + block_shape=block_shape, + ) + + def align_step() -> None: + moe_align_block_size( + topk_ids, + config["BLOCK_SIZE_M"], + global_num_experts, + ) + + alignment = _measure(align_step, args.warmup, args.active) + + def serving_step() -> None: + fused_experts( + hidden_states=hidden_states, + w1=w1, + w2=w2, + topk_weights=topk_weights, + topk_ids=topk_ids, + inplace=True, + use_fp8_w8a8=True, + per_channel_quant=False, + global_num_experts=global_num_experts, + w1_scale=w1_scale, + w2_scale=w2_scale, + block_shape=block_shape, + ) + + serving = _measure(serving_step, args.warmup, args.active) + decomposed_ms = frontier_grouped["median"] + alignment["median"] + rows.append( + { + "num_tokens": num_tokens, + "block_size_m": config["BLOCK_SIZE_M"], + "frontier_grouped_ms": frontier_grouped, + "frontier_alignment_ms": alignment, + "frontier_decomposed_median_ms": decomposed_ms, + "vllm_fused_experts_ms": serving, + "decomposed_over_serving": decomposed_ms / serving["median"], + } + ) + + result = { + "contract": "Frontier grouped_gemm + shuffling alignment vs vLLM fused_experts", + "model_shape": { + "hidden_dim": hidden_dim, + "expert_hidden_dim": expert_hidden_dim, + "global_num_experts": global_num_experts, + "local_num_experts": local_num_experts, + "top_k": top_k, + "tp": args.tp, + "ep": args.ep, + "dtype": "block_fp8_w8a8_bf16_output", + "block_shape": block_shape, + }, + "warmup": args.warmup, + "active": args.active, + "rows": rows, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(result, indent=2) + "\n", encoding="utf-8") + print(json.dumps(result, indent=2)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/profile_allreduce.py b/runs/frontier-multicase-sufficiency-v0/best_effort/profile_allreduce.py new file mode 100644 index 0000000..484a2e7 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/profile_allreduce.py @@ -0,0 +1,256 @@ +#!/usr/bin/env python3 +"""Profile the NCCL all-reduce paths consumed by the Qwen235B experiment.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import os +import statistics +import sys +from datetime import datetime, timezone +from pathlib import Path + +import torch +import torch.distributed as dist + + +DEFAULT_TOKENS = [ + 1, + 2, + 3, + 4, + 6, + 8, + 12, + 16, + 24, + 32, + 48, + 63, + 64, + 65, + 96, + 127, + 128, + 129, + 192, + 255, + 256, + 257, + 384, + 511, + 512, + 513, + 768, + 1023, + 1024, + 1025, + 1536, + 2047, + 2048, + 2049, + 3072, + 4095, + 4096, + 4097, + 6144, + 8191, + 8192, + 8193, + 12288, + 16383, + 16384, +] + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--hidden-size", type=int, default=4096) + parser.add_argument("--world-sizes", type=int, nargs="+", default=[4, 8]) + parser.add_argument("--tokens", type=int, nargs="+", default=DEFAULT_TOKENS) + parser.add_argument("--warmup-iterations", type=int, default=5) + parser.add_argument("--measured-iterations", type=int, default=30) + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def profile_one_size( + *, + group: dist.ProcessGroup, + group_size: int, + num_tokens: int, + hidden_size: int, + warmup_iterations: int, + measured_iterations: int, +) -> list[float]: + tensor = torch.ones( + (num_tokens, hidden_size), dtype=torch.bfloat16, device="cuda" + ) + for _ in range(warmup_iterations): + dist.all_reduce(tensor, group=group) + dist.barrier(group=group) + torch.cuda.synchronize() + + local_samples: list[float] = [] + for _ in range(measured_iterations): + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + dist.all_reduce(tensor, group=group) + end.record() + end.synchronize() + local_samples.append(float(start.elapsed_time(end))) + + samples_by_rank: list[list[float] | None] = [None] * group_size + dist.all_gather_object(samples_by_rank, local_samples, group=group) + del tensor + + # Collective completion is limited by the slowest rank. Preserve that + # critical-path statistic instead of reporting only rank 0. + return [ + max(float(rank_samples[index]) for rank_samples in samples_by_rank if rank_samples) + for index in range(measured_iterations) + ] + + +def profile_group( + *, + group: dist.ProcessGroup, + group_ranks: list[int], + args: argparse.Namespace, +) -> list[dict[str, object]]: + global_rank = dist.get_rank() + if global_rank not in group_ranks: + return [] + + group_size = len(group_ranks) + group_rank = dist.get_rank(group=group) + rows: list[dict[str, object]] = [] + trial_orders = (("forward", args.tokens), ("reverse", list(reversed(args.tokens)))) + for trial_id, (order_name, tokens) in enumerate(trial_orders): + for num_tokens in tokens: + samples = profile_one_size( + group=group, + group_size=group_size, + num_tokens=num_tokens, + hidden_size=args.hidden_size, + warmup_iterations=args.warmup_iterations, + measured_iterations=args.measured_iterations, + ) + if group_rank != 0: + continue + ordered = sorted(samples) + p95_index = int(0.95 * (len(ordered) - 1)) + rows.append( + { + "time_stats.all_reduce.min": min(samples), + "time_stats.all_reduce.max": max(samples), + "time_stats.all_reduce.mean": statistics.fmean(samples), + "time_stats.all_reduce.median": statistics.median(samples), + "time_stats.all_reduce.std": statistics.stdev(samples), + "time_stats.all_reduce.p95": ordered[p95_index], + "rank": 0, + "num_workers": group_size, + "size": num_tokens * args.hidden_size * 2, + "collective": "all_reduce", + "devices_per_node": group_size, + "max_devices_per_node": dist.get_world_size(), + "profiling_precision": "BF16", + "measurement_type": "CUDA_EVENT", + "backend": "nccl", + "num_tokens": num_tokens, + "hidden_size": args.hidden_size, + "trial_id": trial_id, + "order": order_name, + "warmup_iterations": args.warmup_iterations, + "measured_iterations": args.measured_iterations, + } + ) + return rows + + +def main() -> None: + args = parse_args() + if args.hidden_size <= 0: + raise ValueError("hidden-size must be positive") + if args.warmup_iterations <= 0 or args.measured_iterations < 2: + raise ValueError("warmup must be positive and measured iterations must be >= 2") + if sorted(set(args.tokens)) != sorted(args.tokens) or min(args.tokens) <= 0: + raise ValueError("tokens must be unique positive integers") + + local_rank = int(os.environ["LOCAL_RANK"]) + torch.cuda.set_device(local_rank) + dist.init_process_group(backend="nccl") + global_rank = dist.get_rank() + global_world_size = dist.get_world_size() + if max(args.world_sizes) > global_world_size: + raise ValueError( + f"requested world size {max(args.world_sizes)} exceeds torchrun world size " + f"{global_world_size}" + ) + + groups: dict[int, tuple[dist.ProcessGroup, list[int]]] = {} + for world_size in sorted(set(args.world_sizes)): + ranks = list(range(world_size)) + group = dist.group.WORLD if world_size == global_world_size else dist.new_group(ranks) + groups[world_size] = (group, ranks) + + all_rows: list[dict[str, object]] = [] + for world_size in args.world_sizes: + group, ranks = groups[world_size] + all_rows.extend(profile_group(group=group, group_ranks=ranks, args=args)) + dist.barrier() + + if global_rank == 0: + expected_rows = len(set(args.world_sizes)) * len(args.tokens) * 2 + if len(all_rows) != expected_rows: + raise RuntimeError( + f"profile row count mismatch: expected={expected_rows}, actual={len(all_rows)}" + ) + args.output.parent.mkdir(parents=True, exist_ok=True) + with args.output.open("w", newline="", encoding="utf-8") as output_file: + writer = csv.DictWriter(output_file, fieldnames=list(all_rows[0])) + writer.writeheader() + writer.writerows(all_rows) + + manifest = { + "generated_at_utc": datetime.now(timezone.utc).isoformat(), + "command": [sys.executable, *sys.argv], + "output": str(args.output.resolve()), + "output_sha256": sha256(args.output), + "rows": len(all_rows), + "world_sizes": args.world_sizes, + "tokens": args.tokens, + "hidden_size": args.hidden_size, + "dtype": "torch.bfloat16", + "backend": dist.get_backend(), + "torch_version": torch.__version__, + "cuda_version": torch.version.cuda, + "nccl_version": list(torch.cuda.nccl.version()), + "gpu": torch.cuda.get_device_name(0), + "warmup_iterations": args.warmup_iterations, + "measured_iterations": args.measured_iterations, + "sample_semantics": "per-iteration maximum CUDA-event time across participating ranks", + } + manifest_path = args.output.with_suffix(args.output.suffix + ".manifest.json") + manifest_path.write_text( + json.dumps(manifest, indent=2) + "\n", encoding="utf-8" + ) + print(json.dumps(manifest, indent=2)) + + dist.destroy_process_group() + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/test_community_prefill_grid.py b/runs/frontier-multicase-sufficiency-v0/best_effort/test_community_prefill_grid.py new file mode 100644 index 0000000..5357a24 --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/test_community_prefill_grid.py @@ -0,0 +1,66 @@ +from __future__ import annotations + +import importlib.util +import json +import sys +from pathlib import Path + + +SCRIPT = Path(__file__).with_name("community_prefill_grid.py") +SPEC = importlib.util.spec_from_file_location("community_prefill_grid", SCRIPT) +MODULE = importlib.util.module_from_spec(SPEC) +assert SPEC.loader is not None +sys.modules[SPEC.name] = MODULE +SPEC.loader.exec_module(MODULE) + + +def test_grid_matches_frozen_frontier_surface() -> None: + assert len(MODULE.GRID) == 8 + assert {item.name for item in MODULE.GRID} == { + f"tp{tp}_mns{mns}_mbt{mbt}" + for tp in (4, 8) + for mns in (64, 128) + for mbt in (8192, 16384) + } + assert {item.num_gpu_blocks for item in MODULE.GRID if item.tp == 4} == {26101} + assert {item.num_gpu_blocks for item in MODULE.GRID if item.tp == 8} == {62351} + + +def test_study_uses_raw_prompt_and_fixed_serving_contract(tmp_path: Path) -> None: + payload = MODULE.study_payload( + tp=8, + repo=tmp_path, + python=tmp_path / "python", + vllm=tmp_path / "vllm", + model=tmp_path / "model", + trace=tmp_path / "trace.jsonl", + windows=tmp_path / "windows.json", + port=18918, + ) + assert payload["trace"]["request_mode"] == "raw_completion" + assert payload["trace"]["completion_tokens_override"] == 1 + assert payload["trace"]["replay_time_scale"] == 1.0 + assert payload["trace"]["max_concurrency"] == 256 + assert payload["trace"]["max_concurrency"] > max(item.mns for item in MODULE.GRID) + assert payload["engine"]["base_flags"]["enable-expert-parallel"] is True + assert payload["engine"]["base_flags"]["disable-custom-all-reduce"] is True + assert payload["engine"]["base_flags"]["num-gpu-blocks-override"] == 62351 + assert payload["engine"]["base_flags"]["enable-prefix-caching"] is False + + +def test_capacity_interval_replays_binary_decisions() -> None: + probes = [ + {"threshold": 0.0625, "feasible": False}, + {"threshold": 0.03125, "feasible": False}, + {"threshold": 0.015625, "feasible": True}, + {"threshold": 0.0234375, "feasible": False}, + {"threshold": 0.01953125, "feasible": True}, + {"threshold": 0.021484375, "feasible": True}, + ] + assert MODULE.capacity_interval(probes) == [0.021484375, 0.0234375] + + +def test_manifest_json_round_trip(tmp_path: Path) -> None: + path = tmp_path / "x.json" + MODULE.write_json(path, {"schema": MODULE.SCHEMA}) + assert json.loads(path.read_text()) == {"schema": MODULE.SCHEMA} diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/test_fixed_cohort_rank.py b/runs/frontier-multicase-sufficiency-v0/best_effort/test_fixed_cohort_rank.py new file mode 100644 index 0000000..b78c9ad --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/test_fixed_cohort_rank.py @@ -0,0 +1,400 @@ +from __future__ import annotations + +import argparse +import csv +import importlib.util +import json +import sys +from pathlib import Path + + +SCRIPT = Path(__file__).with_name("fixed_cohort_rank.py") +SPEC = importlib.util.spec_from_file_location("fixed_cohort_rank", SCRIPT) +MODULE = importlib.util.module_from_spec(SPEC) +assert SPEC.loader is not None +sys.modules[SPEC.name] = MODULE +SPEC.loader.exec_module(MODULE) + + +def _write_trace(path: Path) -> None: + lengths = [100, 1500, 3000, 6000, 12000, 24000] + rows = [] + for index in range(120): + rows.append( + { + "timestamp": index * 0.5, + "sampling_u": (index + 1) / 121, + "input_length": lengths[index % len(lengths)], + "output_length": 10, + "prompt": f"raw prompt {index}", + } + ) + path.write_text("".join(json.dumps(row) + "\n" for row in rows)) + + +def test_prepare_protocol_keeps_identical_cohort_and_scales_only_arrivals( + tmp_path: Path, +) -> None: + trace = tmp_path / "trace.jsonl" + _write_trace(trace) + manifest_path = MODULE.prepare_protocol( + argparse.Namespace( + trace=trace, + expected_trace_sha256="", + output_root=tmp_path / "protocol", + cohort_size=24, + seed=7, + rate=[0.2, 0.4], + ) + ) + manifest = json.loads(manifest_path.read_text()) + assert manifest["selection"]["cohort_bin_quotas"] == [4, 4, 4, 4, 4, 4] + assert manifest["load_contract"]["binary_search"] is False + assert manifest["load_contract"]["monotonicity_assumed"] is False + + materialized = [] + for record in manifest["rates"].values(): + with Path(record["path"]).open(newline="") as source: + rows = list(csv.DictReader(source)) + materialized.append(rows) + duration = float(rows[-1]["arrived_at"]) - float(rows[0]["arrived_at"]) + assert len(rows) / duration == record["offered_request_rate"] + assert [row["source_row_index"] for row in materialized[0]] == [ + row["source_row_index"] for row in materialized[1] + ] + assert [row["num_prefill_tokens"] for row in materialized[0]] == [ + row["num_prefill_tokens"] for row in materialized[1] + ] + assert MODULE.audit_protocol(manifest_path)["status"] == "passed" + + +def test_score_requests_applies_all_slos_to_same_outcomes() -> None: + requests = [ + {"input_tokens": 100, "success": True, "ttft_ms": 900.0}, + {"input_tokens": 16000, "success": True, "ttft_ms": 2500.0}, + {"input_tokens": 24000, "success": False, "ttft_ms": None}, + ] + score = MODULE.score_requests(requests) + assert score["scores"]["linear_8k_primary"]["passed_request_count"] == 2 + assert score["scores"]["legacy_step_1s_2s"]["passed_request_count"] == 1 + assert score["ttft_ms"]["count"] == 2 + + +def test_rank_surface_records_nonmonotone_feasibility_without_binary_search() -> None: + config_results = [ + { + "config": {"name": "a", "tp": 4}, + "loads": [ + { + "offered_request_rate": 0.2, + "scores": {"linear_8k_primary": {"feasible": False, "slo_pass_rate": 0.9}}, + }, + { + "offered_request_rate": 0.4, + "scores": {"linear_8k_primary": {"feasible": True, "slo_pass_rate": 0.95}}, + }, + ], + }, + { + "config": {"name": "b", "tp": 8}, + "loads": [ + { + "offered_request_rate": 0.2, + "scores": {"linear_8k_primary": {"feasible": True, "slo_pass_rate": 1.0}}, + }, + { + "offered_request_rate": 0.4, + "scores": {"linear_8k_primary": {"feasible": False, "slo_pass_rate": 0.9}}, + }, + ], + }, + ] + ranking = MODULE.rank_surface(config_results, slo_name="linear_8k_primary") + a = next(item for item in ranking if item["config"]["name"] == "a") + assert a["maximum_tested_feasible_request_rate"] == 0.4 + assert a["monotonicity_violations"] == [[0.2, 0.4]] + + +def test_average_ranks_use_midrank_for_capacity_ties() -> None: + assert MODULE._average_ranks({"a": 0.2, "b": 0.2, "c": 0.1}) == { + "a": 1.5, + "b": 1.5, + "c": 3.0, + } + + +def test_community_study_uses_affine_primary_slo(tmp_path: Path) -> None: + payload = MODULE._community_study_payload( + tp=8, + repo=tmp_path, + python=tmp_path / "python", + vllm=tmp_path / "vllm", + model=tmp_path / "model", + trace=tmp_path / "trace.jsonl", + windows=tmp_path / "windows.json", + port=18918, + ) + assert payload["slo"]["ttft_rule"] == { + "kind": "linear_ms", + "intercept_ms": 1000.0, + "per_token_ms": 0.125, + } + assert payload["trace"]["request_mode"] == "raw_completion" + assert payload["trace"]["completion_tokens_override"] == 1 + + +def test_warmup_is_disjoint_and_covers_each_input_bin() -> None: + lengths = [100, 1500, 3000, 6000, 12000, 24000] + requests = [] + cohort_ids = set() + for bin_index, length in enumerate(lengths): + for suffix in ("cohort", "warmup"): + row_id = f"{bin_index}-{suffix}" + if suffix == "cohort": + cohort_ids.add(row_id) + requests.append( + MODULE.TraceRequest( + row_id=row_id, + arrival_s=10.0, + sampling_u=0.5, + body={"prompt": row_id}, + prompt_tokens_hint=length, + completion_tokens_hint=1, + ) + ) + selected = MODULE._select_warmup_requests(requests, cohort_ids=cohort_ids) + assert len(selected) == len(MODULE.INPUT_BINS) - 1 + assert not cohort_ids & {request.row_id for request in selected} + assert [MODULE._input_bin(int(request.prompt_tokens_hint)) for request in selected] == list( + range(len(MODULE.INPUT_BINS) - 1) + ) + assert {request.arrival_s for request in selected} == {0.0} + + +def test_objective_fidelity_separates_raw_and_per_gpu_rankings() -> None: + frontier = [ + { + "config": {"name": "tp8", "tp": 8}, + "maximum_tested_feasible_request_rate": 0.4, + "maximum_tested_feasible_request_rate_per_gpu": 0.05, + }, + { + "config": {"name": "tp4", "tp": 4}, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + }, + ] + real = [ + { + "config": {"name": "tp8", "tp": 8}, + "maximum_tested_feasible_request_rate": 0.2, + "maximum_tested_feasible_request_rate_per_gpu": 0.025, + }, + { + "config": {"name": "tp4", "tp": 4}, + "maximum_tested_feasible_request_rate": 0.3, + "maximum_tested_feasible_request_rate_per_gpu": 0.075, + }, + ] + raw = MODULE._objective_fidelity( + frontier, + real, + field="maximum_tested_feasible_request_rate", + ) + efficiency = MODULE._objective_fidelity( + frontier, + real, + field="maximum_tested_feasible_request_rate_per_gpu", + ) + assert raw["frontier_top1_set"] == ["tp8"] + assert raw["real_top1_set"] == ["tp4"] + assert abs(raw["top1_regret_fraction_worst_tie_break"] - 1 / 3) < 1e-12 + assert efficiency["frontier_top1_set"] == efficiency["real_top1_set"] == ["tp4"] + + +def test_objective_fidelity_marks_null_capacity_unrankable() -> None: + ranking = [ + { + "config": {"name": "a", "tp": 4}, + "maximum_tested_feasible_request_rate": None, + } + ] + result = MODULE._objective_fidelity( + ranking, + ranking, + field="maximum_tested_feasible_request_rate", + ) + assert result["rankable"] is False + + +def test_boundary_repeat_rates_keep_every_transition_and_censored_fallback() -> None: + def result(feasibility: list[bool]) -> dict[str, object]: + return { + "config": {"name": "test"}, + "loads": [ + { + "offered_request_rate": rate, + "scores": {MODULE.PRIMARY_SLO: {"feasible": feasible}}, + } + for rate, feasible in zip((0.1, 0.2, 0.3, 0.4), feasibility) + ], + } + + assert MODULE._boundary_repeat_rates(result([True, True, False, False])) == [ + 0.2, + 0.3, + ] + assert MODULE._boundary_repeat_rates(result([True, False, True, False])) == [ + 0.1, + 0.2, + 0.3, + 0.4, + ] + assert MODULE._boundary_repeat_rates(result([True, True, True, True])) == [ + 0.3, + 0.4, + ] + assert MODULE._boundary_repeat_rates(result([False, False, False, False])) == [ + 0.1, + 0.2, + ] + + sensitivity = result([True, True, False, False]) + for load, feasible in zip(sensitivity["loads"], [True, True, True, False]): + load["scores"]["linear_6k"] = {"feasible": feasible} + assert MODULE._boundary_repeat_rates( + sensitivity, slo_name="linear_6k" + ) == [0.3, 0.4] + assert MODULE._slo_has_feasible_load( + sensitivity, slo_name="linear_6k" + ) + assert not MODULE._slo_has_feasible_load( + result([False, False, False, False]), + slo_name=MODULE.PRIMARY_SLO, + ) + + +def test_request_slo_classification_exposes_aggregate_error_cancellation() -> None: + def results(ttfts: list[float]) -> list[dict[str, object]]: + return [ + { + "config": {"name": "a"}, + "loads": [ + { + "offered_request_rate": 0.1, + "requests": [ + { + "request_id": str(index), + "input_tokens": 0, + "success": True, + "ttft_ms": ttft, + } + for index, ttft in enumerate(ttfts) + ], + } + ], + } + ] + + classification = MODULE._request_slo_classification( + results([900.0, 1100.0]), results([1100.0, 900.0]) + )[MODULE.PRIMARY_SLO] + assert classification["frontier_failure_requests"] == 1 + assert classification["real_failure_requests"] == 1 + assert classification["failure_set_jaccard"] == 0.0 + assert classification["label_accuracy"] == 0.0 + assert classification["false_slo_pass_requests"] == 1 + assert classification["false_slo_fail_requests"] == 1 + + +def test_frontier_freeze_validation_uses_manifest_rates(tmp_path: Path) -> None: + protocol_path = tmp_path / "protocol.json" + protocol = { + "rates": { + "low": {"offered_request_rate": 0.2}, + "high": {"offered_request_rate": 0.4}, + } + } + protocol_path.write_text(json.dumps(protocol)) + freeze_path = tmp_path / "freeze.json" + freeze_path.write_text( + json.dumps( + { + "schema": MODULE.FRONTIER_SCHEMA, + "status": "frozen_before_community_run", + "protocol_manifest_sha256": MODULE.sha256(protocol_path), + "config_results": [ + { + "config": {"name": config.name}, + "loads": [ + {"offered_request_rate": 0.2}, + {"offered_request_rate": 0.4}, + ], + } + for config in MODULE.frontier_grid.GRID + ], + } + ) + ) + validated = MODULE._validate_frontier_freeze(freeze_path, protocol_path) + assert validated["protocol_manifest_sha256"] == MODULE.sha256(protocol_path) + + +def test_compare_trials_separates_real_label_flips_from_stable_frontier_errors( + tmp_path: Path, +) -> None: + config = {"name": "a", "tp": 4} + + def config_result(ttfts: list[float]) -> dict[str, object]: + requests = [ + { + "request_id": str(index), + "input_tokens": 0, + "success": True, + "ttft_ms": ttft, + } + for index, ttft in enumerate(ttfts) + ] + load = { + "offered_request_rate": 0.1, + "requests": requests, + **MODULE.score_requests(requests), + } + return {"config": config, "loads": [load]} + + frontier_result = config_result([900.0, 1100.0]) + first_result = config_result([900.0, 900.0]) + repeat_result = config_result([900.0, 1100.0]) + + def freeze(path: Path, schema: str, result: dict[str, object]) -> Path: + payload = { + "schema": schema, + "status": "completed", + "config_results": [result], + "rankings": { + name: MODULE.rank_surface([result], slo_name=name) + for name in MODULE.SLO_VARIANTS + }, + } + path.write_text(json.dumps(payload)) + return path + + frontier_path = freeze(tmp_path / "frontier.json", MODULE.FRONTIER_SCHEMA, frontier_result) + first_path = freeze(tmp_path / "first.json", MODULE.COMMUNITY_SCHEMA, first_result) + repeat_path = freeze(tmp_path / "repeat.json", MODULE.COMMUNITY_SCHEMA, repeat_result) + output = tmp_path / "stability.json" + MODULE.compare_trials( + argparse.Namespace( + frontier_freeze=frontier_path, + first_community_freeze=first_path, + repeat_community_freeze=repeat_path, + output=output, + ) + ) + result = json.loads(output.read_text()) + primary = result["variants"][MODULE.PRIMARY_SLO] + assert primary["shared_requests"] == 2 + assert primary["real_trial_label_flips"] == 1 + assert primary["frontier_wrong_both_stable_real_label"] == 0 + assert primary["real_trial_load_label_accuracy"] == 0.0 + assert result["primary_original_grid_capacity_stability"]["stable_configs"] == 0 diff --git a/runs/frontier-multicase-sufficiency-v0/best_effort/test_frontier_prefill_grid.py b/runs/frontier-multicase-sufficiency-v0/best_effort/test_frontier_prefill_grid.py new file mode 100644 index 0000000..0e41e8a --- /dev/null +++ b/runs/frontier-multicase-sufficiency-v0/best_effort/test_frontier_prefill_grid.py @@ -0,0 +1,89 @@ +from __future__ import annotations + +import csv +import importlib.util +import json +import sys +from pathlib import Path + + +SCRIPT = Path(__file__).with_name("frontier_prefill_grid.py") +SPEC = importlib.util.spec_from_file_location("frontier_prefill_grid", SCRIPT) +MODULE = importlib.util.module_from_spec(SPEC) +assert SPEC.loader is not None +sys.modules[SPEC.name] = MODULE +SPEC.loader.exec_module(MODULE) + + +def test_binary_search_lattice_covers_six_probe_tree() -> None: + anchors = MODULE.binary_search_lattice() + assert len(anchors) == 63 + assert anchors[0] == 0.001953125 + assert anchors[-1] == 0.123046875 + assert 0.0625 in anchors + + +def test_prepare_preserves_stable_arrival_order(tmp_path: Path) -> None: + trace = tmp_path / "trace.jsonl" + rows = [ + {"timestamp": 1.0, "sampling_u": 0.02, "input_length": 9000, "id": "a"}, + {"timestamp": 0.5, "sampling_u": 0.03, "input_length": 100, "id": "b"}, + {"timestamp": 1.0, "sampling_u": 0.01, "input_length": 200, "id": "c"}, + {"timestamp": 0.2, "sampling_u": 0.01, "input_length": 40000, "id": "x"}, + ] + trace.write_text("".join(json.dumps(row) + "\n" for row in rows)) + manifest_path = MODULE.prepare_traces(trace, tmp_path / "out", "") + manifest = json.loads(manifest_path.read_text()) + record = manifest["anchors"][MODULE.anchor_key(0.0625)] + with Path(record["path"]).open(newline="") as source: + selected = list(csv.DictReader(source)) + assert [row["source_request_id"] for row in selected] == ["b", "a", "c"] + assert [float(row["slo_ttft_ms"]) for row in selected] == [1000, 2000, 1000] + + +def test_score_uses_trace_slo_and_request_id_mapping(tmp_path: Path) -> None: + trace = tmp_path / "trace.csv" + with trace.open("w", newline="") as output: + writer = csv.DictWriter( + output, + fieldnames=["arrived_at", "num_prefill_tokens", "num_decode_tokens", "slo_ttft_ms"], + ) + writer.writeheader() + writer.writerows( + [ + {"arrived_at": 0, "num_prefill_tokens": 10, "num_decode_tokens": 1, "slo_ttft_ms": 1000}, + {"arrived_at": 1, "num_prefill_tokens": 9000, "num_decode_tokens": 1, "slo_ttft_ms": 2000}, + ] + ) + metrics = tmp_path / "request_metrics.csv" + with metrics.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=["Request Id", "ttft"]) + writer.writeheader() + writer.writerows([{"Request Id": 1, "ttft": 1900}, {"Request Id": 0, "ttft": 1100}]) + score = MODULE.score_request_metrics(trace, metrics) + assert score["passed_request_count"] == 1 + assert score["slo_pass_rate"] == 0.5 + assert score["feasible"] is False + + +def test_build_command_preserves_trace_arrivals(tmp_path: Path) -> None: + config = MODULE.GRID[0] + command = MODULE.build_command( + python=tmp_path / "python", + frontier_source=tmp_path / "frontier", + profile_root=tmp_path / "profiles", + profile_paths={ + "linear": tmp_path / "linear.csv", + "attention": tmp_path / "attention.csv", + "moe": tmp_path / "moe.csv", + "all_reduce": tmp_path / "all_reduce.csv", + }, + trace=tmp_path / "trace.csv", + config=config, + probe_dir=tmp_path / "probe", + run_id="arrival-contract", + cache_root=tmp_path / "cache", + ) + + assert command[command.index("--simulation_mode") + 1] == "offline" + assert "--offline_use_generated_request_arrivals" in command diff --git a/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py b/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py new file mode 100644 index 0000000..a7e01c7 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py @@ -0,0 +1,476 @@ +#!/usr/bin/env python3 +"""Compare the frozen Frontier surface with conservative real capacities.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +import re +from collections import defaultdict +from datetime import datetime +from pathlib import Path +from typing import Any + + +RUN_PATTERN = re.compile(r"qwen30-prefill-real-tp(?P\d+)-mns(?P\d+)-") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--fleet-artifacts", type=Path, required=True) + parser.add_argument( + "--simulator-manifest", type=Path, action="append", required=True + ) + parser.add_argument("--output-root", type=Path, required=True) + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def load_json(path: Path) -> dict[str, Any]: + return json.loads(path.read_text()) + + +def sign(value: float) -> int: + return (value > 0) - (value < 0) + + +def kendall_tau_b(real: list[float], simulated: list[float]) -> dict[str, Any]: + if len(real) != len(simulated): + raise ValueError("ranking vectors have different lengths") + concordant = discordant = real_only_ties = simulated_only_ties = both_ties = 0 + for left in range(len(real)): + for right in range(left + 1, len(real)): + real_sign = sign(real[left] - real[right]) + sim_sign = sign(simulated[left] - simulated[right]) + if real_sign == 0 and sim_sign == 0: + both_ties += 1 + elif real_sign == 0: + real_only_ties += 1 + elif sim_sign == 0: + simulated_only_ties += 1 + elif real_sign == sim_sign: + concordant += 1 + else: + discordant += 1 + denominator = math.sqrt( + (concordant + discordant + real_only_ties) + * (concordant + discordant + simulated_only_ties) + ) + tau = (concordant - discordant) / denominator if denominator else None + return { + "kendall_tau_b": tau, + "concordant": concordant, + "discordant": discordant, + "real_only_ties": real_only_ties, + "simulator_only_ties": simulated_only_ties, + "both_ties": both_ties, + } + + +def result_files_by_anchor(root: Path) -> dict[tuple[str, str], Path]: + selected: dict[tuple[str, str], Path] = {} + digests: dict[tuple[str, str], str] = {} + for path in root.glob("artifacts/**/round*/results/r*.json"): + if path.name.startswith("warmup_"): + continue + round_name = path.parent.parent.name + key = (round_name, path.name) + digest = sha256(path) + if key in digests and digests[key] != digest: + raise RuntimeError( + f"conflicting duplicate real anchor {key} under {root}" + ) + digests[key] = digest + if key not in selected or len(path.parts) < len(selected[key].parts): + selected[key] = path + return selected + + +def find_real_runs(root: Path) -> dict[str, Path]: + candidates: dict[str, list[Path]] = defaultdict(list) + for path in root.iterdir(): + if not path.is_dir(): + continue + match = RUN_PATTERN.search(path.name) + if not match: + continue + name = f"tp{int(match.group('tp'))}_mns{int(match.group('mns'))}" + candidates[name].append(path) + selected: dict[str, Path] = {} + for name, paths in candidates.items(): + complete = [ + path + for path in paths + if (path / "remote_run" / "exit_code").is_file() + and (path / "remote_run" / "exit_code").read_text().strip() == "0" + ] + measured_counts = { + path: len(result_files_by_anchor(path)) + for path in complete + } + if not measured_counts: + raise RuntimeError(f"no successful run for {name}") + maximum = max(measured_counts.values()) + richest = [path for path, count in measured_counts.items() if count == maximum] + if len(richest) != 1: + raise RuntimeError(f"ambiguous richest successful run for {name}: {richest}") + selected[name] = richest[0] + if len(selected) != 12: + raise RuntimeError(f"expected 12 real configs, got {sorted(selected)}") + return selected + + +def campaign_resources(root: Path) -> dict[str, Any]: + runs = [] + gpu_hours = 0.0 + for path in sorted(root.iterdir()): + match = RUN_PATTERN.search(path.name) + exit_code = path / "remote_run" / "exit_code" + started_at = path / "remote_run" / "started_at" + finished_at = path / "remote_run" / "finished_at" + if ( + not path.is_dir() + or not match + or not exit_code.is_file() + or exit_code.read_text().strip() != "0" + or not started_at.is_file() + or not finished_at.is_file() + ): + continue + started = datetime.fromisoformat(started_at.read_text().strip()) + finished = datetime.fromisoformat(finished_at.read_text().strip()) + duration_seconds = (finished - started).total_seconds() + tp = int(match.group("tp")) + run_gpu_hours = duration_seconds * tp / 3600.0 + gpu_hours += run_gpu_hours + runs.append( + { + "run": path.name, + "tp": tp, + "duration_seconds": duration_seconds, + "gpu_hours": run_gpu_hours, + } + ) + return { + "successful_fleet_jobs": len(runs), + "gpu_hours": gpu_hours, + "runs": runs, + } + + +def parse_real_config(name: str, run_root: Path) -> dict[str, Any]: + result_files = sorted(result_files_by_anchor(run_root).values()) + if len(result_files) not in {10, 16}: + raise RuntimeError(f"expected 10 base or 16 refined anchors for {name}, got {len(result_files)}") + by_rate: dict[float, list[dict[str, Any]]] = defaultdict(list) + for path in result_files: + payload = load_json(path) + rate = float(payload["workload"]["offered_request_rate"]) + by_rate[rate].append( + { + "path": str(path.resolve()), + "sha256": sha256(path), + "summary": payload["summary"], + } + ) + tp = int(name.split("_")[0][2:]) + base_rates = [4.0, 8.0, 16.0, 32.0, 64.0] + refined_rates = sorted({*base_rates, *(tp * value for value in (5.0, 6.0, 7.0))}) + if tuple(sorted(by_rate)) not in {tuple(base_rates), tuple(refined_rates)}: + raise RuntimeError(f"unexpected rate grid for {name}: {sorted(by_rate)}") + anchors = [] + for rate, rounds in sorted(by_rate.items()): + if len(rounds) != 2: + raise RuntimeError(f"expected two rounds for {name}@{rate}, got {len(rounds)}") + round_feasible = [bool(row["summary"]["slo"]["feasible"]) for row in rounds] + anchors.append( + { + "rate": rate, + "rounds": rounds, + "conservative_feasible": all(round_feasible), + "round_feasible": round_feasible, + "round_ttft_p95_ms": [ + float(row["summary"]["ttft_p95_ms"]) for row in rounds + ], + } + ) + feasible = [row["rate"] for row in anchors if row["conservative_feasible"]] + capacity = max(feasible, default=0.0) + return { + "name": name, + "tp": tp, + "mns": int(name.split("_mns")[1]), + "anchors": anchors, + "capacity": capacity, + "capacity_per_gpu": capacity / tp, + "source_run": str(run_root.resolve()), + } + + +def parse_simulator(manifests: list[Path]) -> tuple[dict[str, Any], list[dict[str, str]]]: + configs: dict[str, Any] = {} + sources = [] + for path in manifests: + payload = load_json(path) + if payload["status"] not in { + "complete", + "frozen_before_real", + "partial_not_decision_bearing", + }: + raise RuntimeError(f"simulator manifest has invalid status: {path}") + sources.append({"path": str(path.resolve()), "sha256": sha256(path)}) + result_by_name = { + result["config"]["name"]: result for result in payload["config_results"] + } + for capacity in payload["capacity"]: + name = capacity["config"]["name"] + result = result_by_name.get(name) + if result is None: + raise RuntimeError(f"missing simulator config result {name}") + entry = configs.setdefault( + name, + { + "name": name, + "tp": int(capacity["config"]["tp"]), + "mns": int(capacity["config"]["mns"]), + "anchor_by_rate": {}, + }, + ) + for load in result["loads"]: + rate = float(load["offered_request_rate"]) + if rate in entry["anchor_by_rate"]: + raise RuntimeError(f"duplicate simulator anchor {name}@{rate}") + entry["anchor_by_rate"][rate] = { + "rate": rate, + "feasible": bool(load["score"]["feasible"]), + "pass_rate": float(load["score"]["pass_rate"]), + "ttft_p95_ms": float(load["score"]["ttft_p95_ms"]), + } + if len(configs) != 12: + raise RuntimeError(f"expected 12 simulator configs, got {sorted(configs)}") + for entry in configs.values(): + entry["anchors"] = [ + entry["anchor_by_rate"][rate] for rate in sorted(entry["anchor_by_rate"]) + ] + del entry["anchor_by_rate"] + feasible = [row["rate"] for row in entry["anchors"] if row["feasible"]] + entry["capacity"] = max(feasible, default=0.0) + entry["capacity_per_gpu"] = entry["capacity"] / entry["tp"] + return configs, sources + + +def compare(real: dict[str, Any], simulated: dict[str, Any]) -> dict[str, Any]: + names = sorted(real, key=lambda name: (real[name]["tp"], real[name]["mns"])) + if set(names) != set(simulated): + raise RuntimeError("real and simulator config sets differ") + real_scores = [real[name]["capacity_per_gpu"] for name in names] + sim_scores = [simulated[name]["capacity_per_gpu"] for name in names] + real_best = max(real_scores) + sim_best = max(sim_scores) + real_top = [name for name in names if real[name]["capacity_per_gpu"] == real_best] + sim_top = [ + name for name in names if simulated[name]["capacity_per_gpu"] == sim_best + ] + worst_sim_choice = min(real[name]["capacity_per_gpu"] for name in sim_top) + best_sim_choice = max(real[name]["capacity_per_gpu"] for name in sim_top) + tau = kendall_tau_b(real_scores, sim_scores) + + pairwise = {"all": {"comparable": 0, "correct": 0}, "within_tp": {}} + for left in range(len(names)): + for right in range(left + 1, len(names)): + real_sign = sign(real_scores[left] - real_scores[right]) + sim_sign = sign(sim_scores[left] - sim_scores[right]) + if real_sign: + pairwise["all"]["comparable"] += 1 + pairwise["all"]["correct"] += int(real_sign == sim_sign) + if real[names[left]]["tp"] == real[names[right]]["tp"] and real_sign: + key = f"tp{real[names[left]]['tp']}" + bucket = pairwise["within_tp"].setdefault( + key, {"comparable": 0, "correct": 0} + ) + bucket["comparable"] += 1 + bucket["correct"] += int(real_sign == sim_sign) + for bucket in [pairwise["all"], *pairwise["within_tp"].values()]: + bucket["accuracy"] = ( + bucket["correct"] / bucket["comparable"] + if bucket["comparable"] + else None + ) + + confusion = {"real_pass_sim_pass": 0, "real_pass_sim_fail": 0, + "real_fail_sim_pass": 0, "real_fail_sim_fail": 0} + anchor_grid_coverage = { + "shared": 0, + "real_only": 0, + "simulator_only": 0, + } + for name in names: + real_anchors = {row["rate"]: row for row in real[name]["anchors"]} + sim_anchors = {row["rate"]: row for row in simulated[name]["anchors"]} + shared = set(real_anchors) & set(sim_anchors) + if not shared: + raise RuntimeError(f"anchor grids do not overlap for {name}") + anchor_grid_coverage["shared"] += len(shared) + anchor_grid_coverage["real_only"] += len(set(real_anchors) - shared) + anchor_grid_coverage["simulator_only"] += len(set(sim_anchors) - shared) + for rate in shared: + real_pass = real_anchors[rate]["conservative_feasible"] + sim_pass = sim_anchors[rate]["feasible"] + key = f"real_{'pass' if real_pass else 'fail'}_sim_{'pass' if sim_pass else 'fail'}" + confusion[key] += 1 + return { + "config_order": names, + "real_top_set": real_top, + "simulator_top_set": sim_top, + "top_set_exact_match": real_top == sim_top, + "top_set_overlap": sorted(set(real_top) & set(sim_top)), + "top1_regret_best": (real_best - best_sim_choice) / real_best, + "top1_regret_worst": (real_best - worst_sim_choice) / real_best, + "real_best_capacity_per_gpu": real_best, + "simulator_best_capacity_per_gpu": sim_best, + "kendall": tau, + "pairwise_non_tied": pairwise, + "anchor_confusion": confusion, + "anchor_grid_coverage": anchor_grid_coverage, + } + + +def write_csv(path: Path, rows: list[dict[str, Any]]) -> None: + with path.open("w", newline="") as handle: + writer = csv.DictWriter( + handle, fieldnames=list(rows[0]), lineterminator="\n" + ) + writer.writeheader() + writer.writerows(rows) + + +def plot(path: Path, rows: list[dict[str, Any]], metrics: dict[str, Any]) -> None: + import matplotlib.pyplot as plt + import numpy as np + + labels = [f"TP{row['tp']}\nMNS{row['mns']}" for row in rows] + x = np.arange(len(rows)) + width = 0.36 + figure, axes = plt.subplots( + 1, 2, figsize=(13.5, 5.0), gridspec_kw={"width_ratios": [3.3, 1.0]} + ) + axis = axes[0] + axis.bar(x - width / 2, [row["real"] for row in rows], width, label="Real vLLM") + axis.bar( + x + width / 2, + [row["simulator"] for row in rows], + width, + label="Frontier simulator", + ) + axis.set_xticks(x, labels, fontsize=8) + axis.set_ylabel("Max tested SLO-feasible request rate / GPU") + axis.set_title("Qwen3-30B-A3B prefill-only: config ranking") + axis.grid(axis="y", alpha=0.25) + axis.legend(frameon=False, ncols=2) + for separator in (3.5, 7.5): + axis.axvline(separator, color="0.75", linewidth=0.8) + tau = metrics["kendall"]["kendall_tau_b"] + annotation = f"worst regret={metrics['top1_regret_worst'] * 100:.1f}%" + annotation += ( + f"\nKendall τ-b={tau:.3f}" + if tau is not None + else "\nKendall τ-b=undefined" + ) + axis.text( + 0.01, + 0.82, + annotation, + transform=axis.transAxes, + va="top", + fontsize=9, + bbox={"facecolor": "white", "edgecolor": "0.8", "alpha": 0.9}, + ) + + confusion = metrics["anchor_confusion"] + matrix = np.array( + [ + [confusion["real_pass_sim_pass"], confusion["real_pass_sim_fail"]], + [confusion["real_fail_sim_pass"], confusion["real_fail_sim_fail"]], + ] + ) + image = axes[1].imshow(matrix, cmap="Blues", vmin=0) + axes[1].set_xticks([0, 1], ["Sim pass", "Sim fail"]) + axes[1].set_yticks([0, 1], ["Real pass", "Real fail"]) + axes[1].set_title(f"{int(matrix.sum())} anchor decisions") + for row in range(2): + for column in range(2): + axes[1].text(column, row, int(matrix[row, column]), ha="center", va="center") + figure.colorbar(image, ax=axes[1], fraction=0.047, pad=0.04) + figure.suptitle("ISL=2048, OSL=1, TTFT≤1256 ms, 95% pass gate; two real rounds") + figure.tight_layout() + figure.savefig(path, dpi=180, bbox_inches="tight") + plt.close(figure) + + +def main() -> None: + args = parse_args() + real_runs = find_real_runs(args.fleet_artifacts.resolve()) + real = {name: parse_real_config(name, path) for name, path in real_runs.items()} + simulated, simulator_sources = parse_simulator( + [path.resolve() for path in args.simulator_manifest] + ) + metrics = compare(real, simulated) + rows = [ + { + "config": name, + "tp": real[name]["tp"], + "mns": real[name]["mns"], + "real": real[name]["capacity_per_gpu"], + "simulator": simulated[name]["capacity_per_gpu"], + } + for name in metrics["config_order"] + ] + resources = campaign_resources(args.fleet_artifacts.resolve()) + resources["fresh_server_anchors"] = sum( + len(config["anchors"]) * 2 for config in real.values() + ) + resources["measured_requests"] = resources["fresh_server_anchors"] * 64 + resources["warmup_requests"] = sum( + min(32, max(4, math.ceil(anchor["rate"] * 2.0))) * 2 + for config in real.values() + for anchor in config["anchors"] + ) + args.output_root.mkdir(parents=True, exist_ok=True) + payload = { + "schema": "qwen30-prefill-fidelity-comparison-v1", + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "contract": { + "model": "Qwen3-30B-A3B", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": False, + "ttft_slo_ms": 1256.0, + "target_pass_rate": 0.95, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + }, + "metrics": metrics, + "real_campaign_resources": resources, + "real": real, + "simulator": simulated, + "simulator_sources": simulator_sources, + } + (args.output_root / "comparison.json").write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n" + ) + write_csv(args.output_root / "capacity.csv", rows) + plot(args.output_root / "qwen30-prefill-ranking.png", rows, metrics) + print(json.dumps(metrics, indent=2, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-phase-factorial-v0/experiment-card.md b/runs/frontier-phase-factorial-v0/experiment-card.md new file mode 100644 index 0000000..b2623c6 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/experiment-card.md @@ -0,0 +1,56 @@ +# 实验 EXP-SIMFID-PHASE-FACTORIAL：prefill-only 是否是 simulator ranking 的容易区间？ + +> **状态：** 已完成（2026-07-17） + +## Claim 与决策 + +- **Parent claim：** Frontier 的 config-ranking fidelity 由 workload execution phase 决定；prefill-only 可能比 decode/mixed 更容易由 isolated operator profiles 组合。 +- **目的：** 用跨 model 的 phase factorial 区分 phase-complexity explanation 与 model/runtime/profile-specific explanation。 +- **Competing hypotheses：** H-phase：prefill-only 是低状态反馈的 compatibility envelope，因此 30B prefill-only 也能达到低-regret ranking，而 decode/mixed 更容易失真；H-stack：30B 的失败主要来自 FA3/CUDA-graph/routing/profile composition 等 stack-specific mismatch，因此即使 prefill-only 也可能失败；H-margin：mixed 的 differential 可以错误，但只要 topology margin 足够仍会选对 top set。 +- **事前预测：** 若 H-phase 成立，30B prefill-only 满足 regret ≤5%、Kendall τ-b ≥0.8，并明显优于同模型 mixed；若 H-stack 成立，30B prefill-only 仍过不了 gate；若 H-margin 成立，235B mixed 可保持 top set，但会漏掉 MNS/MBT pair directions。 +- **判定规则：** 30B prefill-only fail → “prefill-only 是充分条件”被否证，下一步优先做 same-model execution-context ablation；30B prefill-only pass 且新增 decode-heavy mixed fail → 支持 phase hypothesis；30B prefill-only 与 235B mixed 都 pass → phase 不能单独解释，转向 margin-aware compatibility envelope。 + +## Setup + +- **自变量：** model×phase：已有 Qwen3-235B-A22B-FP8 mixed；新增 Qwen3-30B-A3B BF16 prefill-only。 +- **控制变量：** dash0 H20、community vLLM 与各自 frozen Frontier profile、同一 config 内 real/sim 的 request shape、arrival lattice、SLO、prefix policy、MNS/MBT/TP 与随机种子。 +- **30B system context：** community vLLM 0.20.0+cu129，BF16 weights/activation/KV，TP∈{1,2,4}，MNS∈{8,16,32,64}，MBT=8192，chunked prefill on，prefix off；real 保留 runtime 默认 CUDA graph，Frontier profile-only 不做 E2E calibration。 +- **30B workload：** fixed ISL=2048、OSL=1，64 个不同 token-chain prompts，uniform open-loop QPS；fresh server per `(config, rate, round)`，target-rate warmup 与 measured requests 分离。 +- **30B SLO：** TTFT≤1256 ms，至少 61/64 requests 通过；primary score 为最大共同 tested feasible req/s / 实际 TP GPUs。 +- **Boundary refinement rule：** base grid `{4,8,16,32,64}` 先定位每个 TP 的 pass→fail 区间；若除以 TP 后的离散容量产生无法区分的 top tie，则在查看最终 ranking 前追加共同 per-GPU lattice `5/6/7 req/s/GPU`，即 TP1 测 `{5,6,7}`、TP2 测 `{10,12,14}`、TP4 测 `{20,24,28}`。refinement 不替换或删除 base anchors。 +- **235B baseline：** 已冻结 `ISL=2048, OSL=128`、8 configs、68 个 fresh-server anchors；primary sensitivity TTFT≤1256 ms、TPOT≤150 ms。 +- **Baselines：** real community vLLM；Frontier same-stack profile-only；historical Qwen30 mixed profile-only；historical frozen per-TP calibration 只作为 upper bound，不参与本 case 拟合。 +- **Metrics：** top set、worst tie-break regret、Kendall τ-b、exact/non-tied pair direction、anchor confusion、absolute capacity、TTFT p50/p95、real trial variance与GPU-hour。 + +## 预期产物与 review + +- **预期数据：** 30B frozen simulator surface；real config-rate anchors；两模型 phase comparison table；failure mechanism breakdown。 +- **Figure prototype：** `mock-phase-factorial.png`；x=model×phase，左轴=worst regret，右侧 annotation=τ-b；虚线是 5% regret gate。mock 只表达可区分趋势，不进入结论。 +- **人工 review：** 已批准。用户要求 smoke 通过后推进实验；先做这两个 case，再根据 hypothesis verdict 扩展。 +- **Review 意见：** 不把 existing 235B mixed top-set match 隐藏掉；不把“全 config 并列”算作成功 hit；30B prefill-only 必须使用相同 primary ranking objective。 + +## Benchmark design audit + +| Risk | Verdict | 处理 | +|---|---|---| +| Selective benchmarking | PASS for initial screen | 同时报已有 235B mixed success和内部 pairwise failure；后续 expansion 由预注册 verdict 触发 | +| Simplified workload | NEEDS EVIDENCE | fixed-shape 只用于 phase isolation，不外推 trace-faithful mixed | +| Calibration=evaluation | PASS | 新 case 不用 serving E2E 数据拟合 scale | +| Missing significance | PASS for ranking screen | 96 个 config-rate cells 均做两个独立 fresh-server rounds；两轮都 pass 才算 feasible | +| Relative-only result | PASS by design | 同时报 req/s/GPU、TTFT distribution、rank/regret | + +## 复现信息 + +- **Code：** AITuner branch `codex/fidelity-prefix-pilot-20260714`；Frontier upstream `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + frozen known patches。 +- **Environment：** 只使用 dash0 8×H20；Qwen30 venv `/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1`；model `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`。 +- **产物路径：** local/remote `runs/frontier-phase-factorial-v0/`；raw GPU artifacts 由 fleet harvest，condensed JSON/CSV 进入结果目录。 +- **已知 deviation：** 235B 为 FP8/vLLM0.10.2/FlashInfer eager，30B 为 BF16/vLLM0.20/FA3/default CUDA graph；因此跨模型只检验 hypothesis consistency，causal phase claim 最终仍需 same-model phase pair。初始 continuous fleet monitor 未在 fresh-server 间隙保留 controller-level GPU reservation，产生重叠 launch；该 attempt 整体移动到 `invalid-overlap-*`，不进入统计。后续一次试图在“其余 4 卡”上并行 refinement 时，探针再次命中 TP4 fresh-server 空窗，把 4 个 TP1 jobs 放到了同一 GPU set。这四个 jobs 尚未产生 measured result；但当时 TP4/MNS64 已产生的单个 anchor 也按污染处理。两者都整体移到 `invalid-overlap-20260716T1750Z`，TP4/MNS64 从空目录重跑。此后的 barrier waves 不在运行中追加 job；但 Wave 3 的 harvest monitor 在未及时返回 launch 状态时已发射 MNS16/32，紧接的 monitor retry 又在它们的启动空窗发射 MNS64。三者都未产生 measured result，整体移到 `invalid-overlap-20260716T1836Z`。最终 TP4 waves 使用只含本波 jobs 的独立 queue state，不再依赖 pending-job 探针调度。 + +## 结果 + +- **观察事实：** 235B fixed-shape mixed 中 real/sim 的四个 TP4 top configs 完全一致，worst regret=0、τ-b=0.8944；但 20 个 real non-tie pairs 只保持 16 个，10/34 anchors false-infeasible，TP8 MNS×MBT interaction 被漏掉。30B prefill-only 中真机 capacity/GPU 为 TP1=7、TP2=7、TP4=8，Frontier 为 TP1=8、TP2=8、TP4=6；real top set 是四个 TP4 configs，simulator top set 是全部八个 TP1/TP2 configs，无交集。worst regret=12.5%、τ-b=-1.0，32 个 real non-tie pairs 中 0 个同序。96 个 anchor labels 中有 8 个 false-feasible 和 8 个 false-infeasible。 +- **实验成本：** 接受 24 个 fleet jobs、192 个 fresh-server anchors、12,288 个 measured requests 和 4,512 个 warmups，消耗 12.0744 H20-GPU-hours。 +- **异常与排除：** fleet controller 在 fresh-server 空窗期没有保留 GPU reservation，产生了三批重叠 launch。污染 attempt 不进入 accepted artifact root，未产生 measured result 的重叠 jobs 也不被计数；同一波中已产生的 TP4/MNS64 单 anchor 同样按污染丢弃，从空目录重跑。最终 TP4 refinement 使用彼此独立的 queue states。analyzer 按 `(round, filename)` 去重相同 harvest copy，如果 hash 冲突则直接报错。 +- **Interpretation 与剩余 alternatives：** `H-phase` 的强形式（prefill-only 是 fidelity 充分条件）被否证；`H-margin` 与数据更一致。235B 的真机 TP4/TP8 最优 margin 为 2×，足以掩盖内部 residual；30B 的 8-vs-7 margin 被 TP-dependent saturation residual 穿过。但这是跨 stack comparison，不能把差异因果归结为 model size。 +- **Claim update：** “prefill-only 容易，decode/mixed 困难”的强假设被否证。新的可证伪命题是：config-ranking fidelity 取决于 scheduler-state-conditioned action residual 是否大于 real decision margin。 +- **下一步：** 不继续扩展跨模型 phase cases。在 Qwen30 prefill-only 上依次做 measured-collective injection、batch-composition-conditioned pure-prefill attention/step profile、TP1@8/TP2@16/TP4@32 的 scheduler batch/queue/per-step trace 对齐，最后再测 routing/graph。 diff --git a/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml b/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml new file mode 100644 index 0000000..5553f93 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-base-rerun" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-exclusive.toml b/runs/frontier-phase-factorial-v0/fleet-exclusive.toml new file mode 100644 index 0000000..7de87b2 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-exclusive.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-exclusive" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml new file mode 100644 index 0000000..a7e6466 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine-tp4-wave3" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml new file mode 100644 index 0000000..3cba17e --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine-tp4-wave4" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine.toml b/runs/frontier-phase-factorial-v0/fleet-refine.toml new file mode 100644 index 0000000..786d090 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet.toml b/runs/frontier-phase-factorial-v0/fleet.toml new file mode 100644 index 0000000..76a9e93 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 1024 +gpu_free_utilization_pct = 10 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml b/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml new file mode 100644 index 0000000..2eb2f2d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml @@ -0,0 +1,21 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "4 8 16 32 64" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_full.toml b/runs/frontier-phase-factorial-v0/jobs_full.toml new file mode 100644 index 0000000..05d7621 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_full.toml @@ -0,0 +1,241 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "8" +RATES = "4 8 16 32 64" +SERVER_PORT = "8720" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "16" +RATES = "4 8 16 32 64" +SERVER_PORT = "8721" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "32" +RATES = "4 8 16 32 64" +SERVER_PORT = "8722" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "64" +RATES = "4 8 16 32 64" +SERVER_PORT = "8723" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "8" +RATES = "4 8 16 32 64" +SERVER_PORT = "8724" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "16" +RATES = "4 8 16 32 64" +SERVER_PORT = "8725" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "32" +RATES = "4 8 16 32 64" +SERVER_PORT = "8726" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "64" +RATES = "4 8 16 32 64" +SERVER_PORT = "8727" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "8" +RATES = "4 8 16 32 64" +SERVER_PORT = "8728" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "16" +RATES = "4 8 16 32 64" +SERVER_PORT = "8729" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "32" +RATES = "4 8 16 32 64" +SERVER_PORT = "8730" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v2-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "4 8 16 32 64" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_refine.toml b/runs/frontier-phase-factorial-v0/jobs_refine.toml new file mode 100644 index 0000000..55fd01d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine.toml @@ -0,0 +1,240 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns8-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "8" +RATES = "5 6 7" +SERVER_PORT = "8720" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" +[[jobs]] +name = "qwen30-prefill-real-tp1-mns16-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "16" +RATES = "5 6 7" +SERVER_PORT = "8721" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns32-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "32" +RATES = "5 6 7" +SERVER_PORT = "8722" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns64-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "64" +RATES = "5 6 7" +SERVER_PORT = "8723" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns8-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "8" +RATES = "10 12 14" +SERVER_PORT = "8724" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns16-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "16" +RATES = "10 12 14" +SERVER_PORT = "8725" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns32-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "32" +RATES = "10 12 14" +SERVER_PORT = "8726" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns64-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "64" +RATES = "10 12 14" +SERVER_PORT = "8727" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns8-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "8" +RATES = "20 24 28" +SERVER_PORT = "8728" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns16-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "16" +RATES = "20 24 28" +SERVER_PORT = "8729" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns32-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "32" +RATES = "20 24 28" +SERVER_PORT = "8730" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "20 24 28" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml new file mode 100644 index 0000000..b830690 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml @@ -0,0 +1,41 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns16-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "16" +RATES = "20 24 28" +SERVER_PORT = "8729" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns32-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "32" +RATES = "20 24 28" +SERVER_PORT = "8730" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml new file mode 100644 index 0000000..b69872d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml @@ -0,0 +1,21 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "20 24 28" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_smoke.toml b/runs/frontier-phase-factorial-v0/jobs_smoke.toml new file mode 100644 index 0000000..5741e2d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_smoke.toml @@ -0,0 +1,21 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-smoke-tp1-mns8-20260717-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 1800 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-smoke-tp1-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "8" +RATES = "4" +SERVER_PORT = "8718" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-smoke-tp1-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/mock-phase-factorial.png b/runs/frontier-phase-factorial-v0/mock-phase-factorial.png new file mode 100644 index 0000000..6c3c2bd Binary files /dev/null and b/runs/frontier-phase-factorial-v0/mock-phase-factorial.png differ diff --git a/runs/frontier-phase-factorial-v0/plot_mock_phase_factorial.py b/runs/frontier-phase-factorial-v0/plot_mock_phase_factorial.py new file mode 100644 index 0000000..fdfc37d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/plot_mock_phase_factorial.py @@ -0,0 +1,45 @@ +#!/usr/bin/env python3 +"""Render the preregistered phase-factorial hypothesis schematic (mock data).""" + +from __future__ import annotations + +from pathlib import Path + +import matplotlib + +matplotlib.use("Agg") +import matplotlib.pyplot as plt + + +def main() -> None: + labels = ("235B\nprefill-only", "235B\nmixed", "30B\nprefill-only", "30B\nmixed") + x = range(len(labels)) + phase_hypothesis = (1.0, 18.0, 2.0, 28.0) + stack_hypothesis = (1.0, 4.0, 24.0, 30.0) + + fig, ax = plt.subplots(figsize=(8.6, 4.8), constrained_layout=True) + ax.plot(x, phase_hypothesis, marker="o", lw=2, label="H-phase (mock)") + ax.plot(x, stack_hypothesis, marker="s", lw=2, label="H-stack (mock)") + ax.axhline(5, color="black", ls="--", lw=1.2, label="5% regret gate") + ax.set_xticks(list(x), labels) + ax.set_ylabel("Worst selected-config regret (%) — MOCK DATA") + ax.set_title("Schematic only: predictions that distinguish phase vs stack explanations") + ax.set_ylim(0, 35) + ax.grid(axis="y", alpha=0.25) + ax.legend() + ax.text( + 0.01, + 0.98, + "MOCK DATA / NOT A RESULT", + transform=ax.transAxes, + va="top", + color="crimson", + weight="bold", + ) + output = Path(__file__).with_name("mock-phase-factorial.png") + fig.savefig(output, dpi=180) + print(output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-phase-factorial-v0/qwen30_prefill_client.py b/runs/frontier-phase-factorial-v0/qwen30_prefill_client.py new file mode 100644 index 0000000..cff055e --- /dev/null +++ b/runs/frontier-phase-factorial-v0/qwen30_prefill_client.py @@ -0,0 +1,253 @@ +#!/usr/bin/env python3 +"""Open-loop fixed-shape workload for one real offered-load anchor.""" + +from __future__ import annotations + +import argparse +import concurrent.futures +import hashlib +import http.client +import json +import math +import time +from pathlib import Path +from typing import Any + + +TARGET_PASS_RATE = 0.95 + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--host", default="127.0.0.1") + parser.add_argument("--port", type=int, required=True) + parser.add_argument("--served-model", required=True) + parser.add_argument("--model-path", type=Path, required=True) + parser.add_argument("--rate", type=float, required=True) + parser.add_argument("--requests", type=int, default=64) + parser.add_argument("--input-tokens", type=int, default=2048) + parser.add_argument("--output-tokens", type=int, default=1) + parser.add_argument("--ttft-slo-ms", type=float) + parser.add_argument("--tpot-slo-ms", type=float, default=150.0) + parser.add_argument("--timeout-seconds", type=float, default=900.0) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def percentile(values: list[float], fraction: float) -> float | None: + if not values: + return None + ordered = sorted(values) + index = min(len(ordered) - 1, max(0, math.ceil(fraction * len(ordered)) - 1)) + return ordered[index] + + +def ttft_slo_ms(input_tokens: int) -> float: + return 1000.0 + 1000.0 * input_tokens / 8000.0 + + +def run_request( + *, + request_index: int, + scheduled_at: float, + benchmark_start: float, + args: argparse.Namespace, + prompt_ids: list[int], +) -> dict[str, Any]: + delay = scheduled_at - time.perf_counter() + if delay > 0: + time.sleep(delay) + admitted = time.perf_counter() + record: dict[str, Any] = { + "request_index": request_index, + "scheduled_s": scheduled_at - benchmark_start, + "admitted_s": admitted - benchmark_start, + "admission_lag_ms": (admitted - scheduled_at) * 1000.0, + "success": False, + } + connection = http.client.HTTPConnection(args.host, args.port, timeout=args.timeout_seconds) + body = { + "model": args.served_model, + "prompt": prompt_ids, + "min_tokens": args.output_tokens, + "max_tokens": args.output_tokens, + "ignore_eos": True, + "temperature": 0, + "stream": True, + "stream_options": {"include_usage": True}, + "return_token_ids": True, + } + try: + started = time.perf_counter() + connection.request( + "POST", + "/v1/completions", + body=json.dumps(body, separators=(",", ":")).encode(), + headers={"Content-Type": "application/json"}, + ) + response = connection.getresponse() + if response.status != 200: + raise RuntimeError( + f"HTTP {response.status}: {response.read().decode(errors='replace')}" + ) + first_token_at = None + last_token_at = None + streamed_tokens = 0 + usage = None + while True: + raw = response.readline() + if not raw: + break + line = raw.decode(errors="replace").strip() + if not line.startswith("data:"): + continue + data = line[5:].strip() + if data == "[DONE]": + break + payload = json.loads(data) + if payload.get("usage"): + usage = payload["usage"] + emitted = 0 + for choice in payload.get("choices") or []: + token_ids = choice.get("token_ids") or [] + emitted += len(token_ids) if token_ids else int(bool(choice.get("text"))) + if emitted: + now = time.perf_counter() + first_token_at = first_token_at or now + last_token_at = now + streamed_tokens += emitted + finished = time.perf_counter() + if first_token_at is None or usage is None: + raise RuntimeError("missing streaming token or usage") + prompt_tokens = int(usage["prompt_tokens"]) + completion_tokens = int(usage["completion_tokens"]) + if prompt_tokens != args.input_tokens or completion_tokens != args.output_tokens: + raise RuntimeError(f"usage mismatch: {prompt_tokens}+{completion_tokens}") + ttft = (first_token_at - started) * 1000.0 + tpot = ( + (last_token_at - first_token_at) * 1000.0 / (completion_tokens - 1) + if completion_tokens > 1 + else None + ) + record.update( + { + "success": True, + "prompt_tokens": prompt_tokens, + "completion_tokens": completion_tokens, + "streamed_token_count": streamed_tokens, + "ttft_ms": ttft, + "tpot_ms": tpot, + "e2e_ms": (finished - started) * 1000.0, + "slo_pass": ttft <= args.ttft_slo_ms + and (tpot is None or tpot <= args.tpot_slo_ms), + } + ) + except Exception as error: # Failed requests remain in the SLO denominator. + record["error"] = f"{type(error).__name__}: {error}" + record["slo_pass"] = False + finally: + connection.close() + return record + + +def main() -> None: + args = parse_args() + if min( + args.rate, + args.requests, + args.input_tokens, + args.output_tokens, + args.tpot_slo_ms, + ) <= 0: + raise ValueError("rate, requests, tokens, and SLO must be positive") + if args.ttft_slo_ms is None: + args.ttft_slo_ms = ttft_slo_ms(args.input_tokens) + if args.ttft_slo_ms <= 0: + raise ValueError("TTFT SLO must be positive") + from transformers import AutoTokenizer + + tokenizer = AutoTokenizer.from_pretrained(args.model_path, trust_remote_code=True) + excluded = set(tokenizer.all_special_ids) + candidates = [ + token_id for token_id in range(tokenizer.vocab_size) if token_id not in excluded + ] + if len(candidates) < args.requests + 1: + raise RuntimeError("tokenizer has too few non-special token IDs") + base_id = candidates[0] + prompts = [ + [candidates[index + 1], *([base_id] * (args.input_tokens - 1))] + for index in range(args.requests) + ] + prompt_hash = hashlib.sha256( + "\n".join(",".join(map(str, prompt)) for prompt in prompts).encode() + ).hexdigest() + + benchmark_start = time.perf_counter() + 2.0 + with concurrent.futures.ThreadPoolExecutor(max_workers=args.requests) as pool: + futures = [ + pool.submit( + run_request, + request_index=index, + scheduled_at=benchmark_start + index / args.rate, + benchmark_start=benchmark_start, + args=args, + prompt_ids=prompts[index], + ) + for index in range(args.requests) + ] + requests = [future.result() for future in futures] + requests.sort(key=lambda row: int(row["request_index"])) + completed = [row for row in requests if row["success"]] + passed = sum(bool(row["slo_pass"]) for row in requests) + ttfts = [float(row["ttft_ms"]) for row in completed] + tpots = [ + float(row["tpot_ms"]) + for row in completed + if row["tpot_ms"] is not None + ] + pass_rate = passed / len(requests) + payload = { + "schema": "qwen30-fixed-rate-anchor-v2", + "workload": { + "offered_request_rate": args.rate, + "request_count": args.requests, + "input_tokens": args.input_tokens, + "output_tokens": args.output_tokens, + "prefix_caching": False, + "arrival": "open_loop_uniform", + "last_scheduled_arrival_s": (args.requests - 1) / args.rate, + "prompt_vector_sha256": prompt_hash, + }, + "summary": { + "completed": len(completed), + "failed": len(requests) - len(completed), + "ttft_p50_ms": percentile(ttfts, 0.50), + "ttft_p95_ms": percentile(ttfts, 0.95), + "ttft_max_ms": max(ttfts) if ttfts else None, + "tpot_p50_ms": percentile(tpots, 0.50), + "tpot_p95_ms": percentile(tpots, 0.95), + "tpot_max_ms": max(tpots) if tpots else None, + "admission_lag_max_ms": max( + float(row["admission_lag_ms"]) for row in requests + ), + "slo": { + "ttft_threshold_ms": args.ttft_slo_ms, + "tpot_threshold_ms": ( + args.tpot_slo_ms if args.output_tokens > 1 else None + ), + "passed": passed, + "pass_rate": pass_rate, + "feasible": pass_rate >= TARGET_PASS_RATE, + }, + }, + "requests": requests, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps(payload["summary"], sort_keys=True), flush=True) + if len(completed) != args.requests: + raise SystemExit(2) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-phase-factorial-v0/remote-sync-marker/README.md b/runs/frontier-phase-factorial-v0/remote-sync-marker/README.md new file mode 100644 index 0000000..9dfd754 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/remote-sync-marker/README.md @@ -0,0 +1,6 @@ +# Phase-factorial fleet sync marker + +Experiment source is synchronized through Git into the clean dash0 checkout +`/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1`. This small +directory exists only to satisfy the fleet orchestrator's explicit sync phase +without copying local simulator caches or raw metrics to the GPU host. diff --git a/runs/frontier-phase-factorial-v0/results/final/capacity.csv b/runs/frontier-phase-factorial-v0/results/final/capacity.csv new file mode 100644 index 0000000..de7a968 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/results/final/capacity.csv @@ -0,0 +1,13 @@ +config,tp,mns,real,simulator +tp1_mns8,1,8,7.0,8.0 +tp1_mns16,1,16,7.0,8.0 +tp1_mns32,1,32,7.0,8.0 +tp1_mns64,1,64,7.0,8.0 +tp2_mns8,2,8,7.0,8.0 +tp2_mns16,2,16,7.0,8.0 +tp2_mns32,2,32,7.0,8.0 +tp2_mns64,2,64,7.0,8.0 +tp4_mns8,4,8,8.0,6.0 +tp4_mns16,4,16,8.0,6.0 +tp4_mns32,4,32,8.0,6.0 +tp4_mns64,4,64,8.0,6.0 diff --git a/runs/frontier-phase-factorial-v0/results/final/comparison.json b/runs/frontier-phase-factorial-v0/results/final/comparison.json new file mode 100644 index 0000000..db46764 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/results/final/comparison.json @@ -0,0 +1,5860 @@ +{ + "contract": { + "input_tokens": 2048, + "model": "Qwen3-30B-A3B", + "output_tokens": 1, + "prefix_caching": false, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "metrics": { + "anchor_confusion": { + "real_fail_sim_fail": 24, + "real_fail_sim_pass": 8, + "real_pass_sim_fail": 8, + "real_pass_sim_pass": 56 + }, + "config_order": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "kendall": { + "both_ties": 34, + "concordant": 0, + "discordant": 32, + "kendall_tau_b": -1.0, + "real_only_ties": 0, + "simulator_only_ties": 0 + }, + "pairwise_non_tied": { + "all": { + "accuracy": 0.0, + "comparable": 32, + "correct": 0 + }, + "within_tp": {} + }, + "real_best_capacity_per_gpu": 8.0, + "real_top_set": [ + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "simulator_best_capacity_per_gpu": 8.0, + "simulator_top_set": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64" + ], + "top1_regret_best": 0.125, + "top1_regret_worst": 0.125, + "top_set_exact_match": false, + "top_set_overlap": [] + }, + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "real": { + "tp1_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.15452707000077, + 152.64549700077623 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r4p00.json", + "sha256": "d82e386ec3effdf57c148171311928298405da0e4f896584bf05db2485bb6bd4", + "summary": { + "admission_lag_max_ms": 0.11588609777390957, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.6234939750284, + "ttft_p50_ms": 151.05855697765946, + "ttft_p95_ms": 153.15452707000077 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r4p00.json", + "sha256": "1565be89131e3c8a9c352c6f29694ef2793734b0623a2db1cbea7e0b0e8a2bf6", + "summary": { + "admission_lag_max_ms": 0.11451996397227049, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.18561100866646, + "ttft_p50_ms": 151.18564700242132, + "ttft_p95_ms": 152.64549700077623 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.8308589477092, + 152.43642206769437 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r5p00.json", + "sha256": "e4447d94a326c541412b40157cfe283a6350680e3285b5d10fed6b8de5c0cd93", + "summary": { + "admission_lag_max_ms": 0.17109804321080446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.35138496104628, + "ttft_p50_ms": 150.91424097772688, + "ttft_p95_ms": 152.8308589477092 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r5p00.json", + "sha256": "572c17e309c6c9f23c490979e7acb885475f142bc530ecfee88916b048259ea2", + "summary": { + "admission_lag_max_ms": 0.1321269664913416, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.11930498946458, + "ttft_p50_ms": 150.50161199178547, + "ttft_p95_ms": 152.43642206769437 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.56597595382482, + 152.16859406791627 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r6p00.json", + "sha256": "68c8b39154d46f1ceda17f47e8cbd84a8c8a9f088c5e82d8f0ac25a7180dc14c", + "summary": { + "admission_lag_max_ms": 0.11641765013337135, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.50876995641738, + "ttft_p50_ms": 150.41159105021507, + "ttft_p95_ms": 153.56597595382482 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r6p00.json", + "sha256": "a970a2cdf7034026a56af3ea4da59856fd588da35d66ef4f9e15481a892edcc9", + "summary": { + "admission_lag_max_ms": 0.11551892384886742, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 168.65782206878066, + "ttft_p50_ms": 150.03317291848361, + "ttft_p95_ms": 152.16859406791627 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 300.90478900820017, + 293.0445579113439 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r7p00.json", + "sha256": "1aa74c18adbff53b36ae8a6f07b5b0865f6b68fb3e7862a70b36f67c18f0bba0", + "summary": { + "admission_lag_max_ms": 0.10505830869078636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.15459000505507, + "ttft_p50_ms": 263.44375393819064, + "ttft_p95_ms": 300.90478900820017 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r7p00.json", + "sha256": "e14bc9dba35d453862839eeed77262a1ed9b8394404ca90c60ec8cb26a1aa73b", + "summary": { + "admission_lag_max_ms": 0.10963028762489557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 295.1545650139451, + "ttft_p50_ms": 243.97043196950108, + "ttft_p95_ms": 293.0445579113439 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1323.4397460473701, + 1315.5151740647852 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r8p00.json", + "sha256": "9aeee7453ffe5098d203e02b84f5009ed347c478b736e89297bd3695dd81d477", + "summary": { + "admission_lag_max_ms": 0.1251589274033904, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1477.699184906669, + "ttft_p50_ms": 776.2576789828017, + "ttft_p95_ms": 1323.4397460473701 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r8p00.json", + "sha256": "7630f4901cf4ec4ea955738b119a1c32aba53f9ca1edb8819556fc60ae1fda09", + "summary": { + "admission_lag_max_ms": 0.11316593736410141, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1404.566578916274, + "ttft_p50_ms": 761.7035870207474, + "ttft_p95_ms": 1315.5151740647852 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4892.253074911423, + 4893.758205929771 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r16p00.json", + "sha256": "d3a2a85cf06a4f080edc9a3809f05493ac55b0aa1d890806ffbc551c8eca14f8", + "summary": { + "admission_lag_max_ms": 0.39830803871154785, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.215195022523, + "ttft_p50_ms": 2742.7029330283403, + "ttft_p95_ms": 4892.253074911423 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r16p00.json", + "sha256": "67110284c38a9a60228f00b1d2b0de9477d6dec660f1c91eb4f39e1b79e272a5", + "summary": { + "admission_lag_max_ms": 0.11306500528007746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.818040987477, + "ttft_p50_ms": 2742.512951954268, + "ttft_p95_ms": 4893.758205929771 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6650.756464921869, + 6751.845097984187 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r32p00.json", + "sha256": "affc62b27382400497bb82e9da31ed9891b60155318fc4f250dca0704ccc5e02", + "summary": { + "admission_lag_max_ms": 0.11010200250893831, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6920.621327939443, + "ttft_p50_ms": 3650.839229929261, + "ttft_p95_ms": 6650.756464921869 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r32p00.json", + "sha256": "df0c8d703b023692814be41f47da4db4ce1b7eb0c3423a742c10884b6447583f", + "summary": { + "admission_lag_max_ms": 0.1288279891014099, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6897.979240980931, + "ttft_p50_ms": 3730.34580796957, + "ttft_p95_ms": 6751.845097984187 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7732.2855240199715, + 7744.056030060165 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r64p00.json", + "sha256": "fbfad78129c843c85905c3f94f278f468a0a019c0cdf0b62140d0fea8f0aaffc", + "summary": { + "admission_lag_max_ms": 0.11090096086263657, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7836.245508980937, + "ttft_p50_ms": 4328.587644966319, + "ttft_p95_ms": 7732.2855240199715 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r64p00.json", + "sha256": "0f005c642327708e1ceb058f71f806b4d46ea666f45695605552e8f234521608", + "summary": { + "admission_lag_max_ms": 0.10862003546208143, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7847.871358972043, + "ttft_p50_ms": 4332.9262170009315, + "ttft_p95_ms": 7744.056030060165 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp1_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.44141807872802, + 152.2580359596759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r4p00.json", + "sha256": "211fa276e4a99a061126882085b9850549b453520ac3414dd2d332c89e52b431", + "summary": { + "admission_lag_max_ms": 0.09491504170000553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 169.56262104213238, + "ttft_p50_ms": 150.8563900133595, + "ttft_p95_ms": 153.44141807872802 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r4p00.json", + "sha256": "17428902a5b20e9084f20792963f0620a1af06cdc1afb76ee99fd15dd905b89e", + "summary": { + "admission_lag_max_ms": 0.12033001985400915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.99700601119548, + "ttft_p50_ms": 150.5822929320857, + "ttft_p95_ms": 152.2580359596759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.05012692604214, + 152.00478909537196 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r5p00.json", + "sha256": "b0eb19ff7ab8a26e34e91f213a983e58689c63071285e27b3719bda0a765ff32", + "summary": { + "admission_lag_max_ms": 0.11265103239566088, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.48999004345387, + "ttft_p50_ms": 150.21017810795456, + "ttft_p95_ms": 152.05012692604214 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r5p00.json", + "sha256": "3b857dff42dd16ebee95356ceacd6b58ff7bd0a9b91f3d6208d489b84985697f", + "summary": { + "admission_lag_max_ms": 0.11149293277412653, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.33813393954188, + "ttft_p50_ms": 150.45341395307332, + "ttft_p95_ms": 152.00478909537196 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.74227899499238, + 152.82704099081457 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r6p00.json", + "sha256": "7615fa664a99f02f9f6d4f1fb0f2d6cba11842772c086634b939bc08cd0f2f4d", + "summary": { + "admission_lag_max_ms": 0.1096969936043024, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.2669020127505, + "ttft_p50_ms": 150.6725640501827, + "ttft_p95_ms": 152.74227899499238 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r6p00.json", + "sha256": "345e3cd3f8cd4d7c7a5e9343f937dd9b1b5928500133830f3046b743da6df64e", + "summary": { + "admission_lag_max_ms": 0.12192933354526758, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.12304301187396, + "ttft_p50_ms": 150.68741200957447, + "ttft_p95_ms": 152.82704099081457 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 302.5522669777274, + 294.5978350471705 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r7p00.json", + "sha256": "3f54a3e3f2279a5325a173ff714d600fb1a2cdfcfba1e8e5c980f7b21b99be59", + "summary": { + "admission_lag_max_ms": 0.11122168507426977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.24335307907313, + "ttft_p50_ms": 262.4604320153594, + "ttft_p95_ms": 302.5522669777274 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r7p00.json", + "sha256": "0832169cdf1bef49c5006464c6cbcd50f4567e0ab90bec98d722f64f7b476763", + "summary": { + "admission_lag_max_ms": 0.1059951027855277, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 297.1431679325178, + "ttft_p50_ms": 249.51713997870684, + "ttft_p95_ms": 294.5978350471705 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1321.9368209829554, + 1318.7565059633926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r8p00.json", + "sha256": "a0f5900ac9d2c47dcc7d7ac08d3f3d23e1d6f529fdd0200989afa819dd3df055", + "summary": { + "admission_lag_max_ms": 0.17414195463061333, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1475.8616799954325, + "ttft_p50_ms": 775.7647309917957, + "ttft_p95_ms": 1321.9368209829554 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r8p00.json", + "sha256": "c0fc3c85095ddacc66849e6f58be0851abab6d4022ee4eb7120c8cb8ea9e5d2c", + "summary": { + "admission_lag_max_ms": 2.749277977272868, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1407.0930189918727, + "ttft_p50_ms": 763.8335369993001, + "ttft_p95_ms": 1318.7565059633926 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4893.3289990527555, + 4893.17379205022 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r16p00.json", + "sha256": "4e9ccaf5384f622b1da7c58d3de803c86099dd9deae91e363aabccc14aeabc05", + "summary": { + "admission_lag_max_ms": 0.11893699411302805, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.85822511185, + "ttft_p50_ms": 2742.8449960425496, + "ttft_p95_ms": 4893.3289990527555 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r16p00.json", + "sha256": "c207e415186aed1148e16085145bcdfd189314e7b92f74ea72fb2e4972ef4eae", + "summary": { + "admission_lag_max_ms": 0.09541702456772327, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.2590809213, + "ttft_p50_ms": 2740.232800017111, + "ttft_p95_ms": 4893.17379205022 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6653.632597066462, + 6647.155530983582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r32p00.json", + "sha256": "780f563a2739d892761905031f82ca8bd0ac4f4eb485c454390dcb2bcdfc1b81", + "summary": { + "admission_lag_max_ms": 0.105265062302351, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6922.717701992951, + "ttft_p50_ms": 3652.452490059659, + "ttft_p95_ms": 6653.632597066462 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r32p00.json", + "sha256": "2a5eb8fb37f03e759de149d24f9b23c812a26afe72ffaaffa030688904634e43", + "summary": { + "admission_lag_max_ms": 0.10375294368714094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6917.111590038985, + "ttft_p50_ms": 3649.0042679943144, + "ttft_p95_ms": 6647.155530983582 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7747.191094094887, + 7839.257381972857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r64p00.json", + "sha256": "b94a39a91842ab39d545c473fc4f872ec79eb1888943c3e03ab03c18961a9105", + "summary": { + "admission_lag_max_ms": 0.6591080455109477, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7851.370444986969, + "ttft_p50_ms": 4335.578468977474, + "ttft_p95_ms": 7747.191094094887 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r64p00.json", + "sha256": "69ab47305c4bb14f3b4ed5bc0b18a37b8a5e7c1cdbee9232e6d2b9616ed10f84", + "summary": { + "admission_lag_max_ms": 0.10753795504570007, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.125, + "passed": 8, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7885.349316056818, + "ttft_p50_ms": 4009.1016669757664, + "ttft_p95_ms": 7839.257381972857 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp1_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 156.76817402709275, + 152.9719429090619 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r4p00.json", + "sha256": "df798dd7a1cde3cfaf698286f44bbac2b0730a2c806f125268bdf7226e2bb47c", + "summary": { + "admission_lag_max_ms": 0.1147780567407608, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 162.9874319769442, + "ttft_p50_ms": 152.67050196416676, + "ttft_p95_ms": 156.76817402709275 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r4p00.json", + "sha256": "df80057224d69ce2a79f937aa45c2093d66d131765f3c7bfcc2289a9aaff3d6a", + "summary": { + "admission_lag_max_ms": 0.2131829969584942, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.917470946908, + "ttft_p50_ms": 150.96506499685347, + "ttft_p95_ms": 152.9719429090619 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.4973110537976, + 155.23677493911237 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r5p00.json", + "sha256": "e614ffaf879859407fc6420847f0140a81bda9a94fcc438fb6782c9019bf5378", + "summary": { + "admission_lag_max_ms": 0.1283780438825488, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.08639602921903, + "ttft_p50_ms": 152.1085740532726, + "ttft_p95_ms": 153.4973110537976 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r5p00.json", + "sha256": "13e825eb5ced3d0cb2b5dd68d9243fdf1cee8779b90e9aad77be04053be22be2", + "summary": { + "admission_lag_max_ms": 0.1683390000835061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.6691420264542, + "ttft_p50_ms": 152.42570801638067, + "ttft_p95_ms": 155.23677493911237 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.53963600844145, + 152.85827894695103 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r6p00.json", + "sha256": "ac530f98f57ec452538ebe2311291d0a7c83bbd5554e82f53e78a2311802e411", + "summary": { + "admission_lag_max_ms": 0.13490929268300533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.18225098680705, + "ttft_p50_ms": 150.29372205026448, + "ttft_p95_ms": 152.53963600844145 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r6p00.json", + "sha256": "2f1445ee0890b0972252d10180ef6f00cbad45ee73bd174f21b370cdf950b654", + "summary": { + "admission_lag_max_ms": 0.16514293383806944, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.88926689513028, + "ttft_p50_ms": 150.2534201135859, + "ttft_p95_ms": 152.85827894695103 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 346.96662798523903, + 295.1050230767578 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r7p00.json", + "sha256": "c4765794d442daf7eeb64a2f587792701ebe7659c7702a3bbbfc8900801fff0e", + "summary": { + "admission_lag_max_ms": 0.16206549480557442, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 354.881492909044, + "ttft_p50_ms": 310.15722593292594, + "ttft_p95_ms": 346.96662798523903 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r7p00.json", + "sha256": "30addb463839e23a7356a5cf6d4060ec1eebfa2b9564fa9500be5bcb7aecb537", + "summary": { + "admission_lag_max_ms": 0.10446517262607813, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 296.6241450048983, + "ttft_p50_ms": 246.06937903445214, + "ttft_p95_ms": 295.1050230767578 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1376.107804942876, + 1321.1636180058122 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r8p00.json", + "sha256": "89e2975f9c7319e477789ea77015541d5235efa61d90fb8bd5b0420a58f7f186", + "summary": { + "admission_lag_max_ms": 0.1416490413248539, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.890625, + "passed": 57, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1501.330335973762, + "ttft_p50_ms": 802.2130889585242, + "ttft_p95_ms": 1376.107804942876 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r8p00.json", + "sha256": "9ca6935f1320033c3ec4b72a35c5d9909ccd7b0fff2e4b2c92c6a26f52887df1", + "summary": { + "admission_lag_max_ms": 0.12422900181263685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1410.047525074333, + "ttft_p50_ms": 766.8001820566133, + "ttft_p95_ms": 1321.1636180058122 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4950.155022088438, + 4892.750842031091 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r16p00.json", + "sha256": "4fce2749435bf2e8c77999e11be0a788cd85f459a647760f3db34f1dbc0aee85", + "summary": { + "admission_lag_max_ms": 0.10520406067371368, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.171875, + "passed": 11, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 5031.207721913233, + "ttft_p50_ms": 2775.8131299633533, + "ttft_p95_ms": 4950.155022088438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r16p00.json", + "sha256": "33ded0192d8986dbcf0a7f5ab6329a440994243d4e528d6c1b278bb6b92720fd", + "summary": { + "admission_lag_max_ms": 0.14005706179887056, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.664474112913, + "ttft_p50_ms": 2743.070787983015, + "ttft_p95_ms": 4892.750842031091 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6750.775643973611, + 6754.7571890754625 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r32p00.json", + "sha256": "44030c4d3374cecebe67b886b428ba8014c87d941c00a178d6049bc57c809f46", + "summary": { + "admission_lag_max_ms": 0.7358170114457607, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6899.517056066543, + "ttft_p50_ms": 3728.6347220651805, + "ttft_p95_ms": 6750.775643973611 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r32p00.json", + "sha256": "f397bad36552c6df5bf5bcee702cfa968ddf3e20b20e9976554ddfbe838ef3b2", + "summary": { + "admission_lag_max_ms": 0.7476879982277751, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6901.606322033331, + "ttft_p50_ms": 3729.3128239689395, + "ttft_p95_ms": 6754.7571890754625 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7800.616211956367, + 7748.703246936202 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r64p00.json", + "sha256": "61eadc2af60386d6972d69fe7b083874e705864162073d54d1bde2bc0e6de7e7", + "summary": { + "admission_lag_max_ms": 0.11387700214982033, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7905.77321103774, + "ttft_p50_ms": 4366.223473916762, + "ttft_p95_ms": 7800.616211956367 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r64p00.json", + "sha256": "db02e9a814154a9259cbe3fefdf15700923458f7e1208596e861b69b6a6ddd41", + "summary": { + "admission_lag_max_ms": 0.14900392852723598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7852.748095989227, + "ttft_p50_ms": 4337.09563605953, + "ttft_p95_ms": 7748.703246936202 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp1_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.07429495826364, + 153.62840006127954 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r4p00.json", + "sha256": "23d5ab491b70c02f51c2f06dec0407edc0627819a33b9bcef1349d974a4e20fe", + "summary": { + "admission_lag_max_ms": 0.14362391084432602, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 171.46877304185182, + "ttft_p50_ms": 151.62539994344115, + "ttft_p95_ms": 154.07429495826364 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r4p00.json", + "sha256": "563aa230e0f5e420cf453dfd9b829d7df493919192d638cbe76f85eb77b6d2e7", + "summary": { + "admission_lag_max_ms": 0.11323392391204834, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.62100906390697, + "ttft_p50_ms": 151.69239102397114, + "ttft_p95_ms": 153.62840006127954 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.20477103907615, + 153.81942794192582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r5p00.json", + "sha256": "6f2a04a5c11eed5826e2a6e1ce6e6f8d98504b31d0fec30d9440df2cdb62e56d", + "summary": { + "admission_lag_max_ms": 0.10794296395033598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.37187200970948, + "ttft_p50_ms": 152.59731002151966, + "ttft_p95_ms": 154.20477103907615 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r5p00.json", + "sha256": "27a0bfa65a66e5cc77e10a5cf39a846f0fd1dd266c10926dae9db7a80c581d6e", + "summary": { + "admission_lag_max_ms": 0.1204309519380331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.40366806089878, + "ttft_p50_ms": 152.0398510619998, + "ttft_p95_ms": 153.81942794192582 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.00951700191945, + 153.5524509381503 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r6p00.json", + "sha256": "4562fae05d510cd50f2c9d7479916d056c8ac4d04851d2388b98f8410192c23a", + "summary": { + "admission_lag_max_ms": 0.10347133502364159, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.3970319693908, + "ttft_p50_ms": 150.414744974114, + "ttft_p95_ms": 153.00951700191945 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r6p00.json", + "sha256": "bd13283e2807150396b278f46ff630ae9820895e0c054a9efcfd8fc213947d47", + "summary": { + "admission_lag_max_ms": 0.13600767124444246, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 159.63539096992463, + "ttft_p50_ms": 151.83425601571798, + "ttft_p95_ms": 153.5524509381503 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 336.8800369789824, + 329.65395506471395 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r7p00.json", + "sha256": "fd5afe52238d6ddb72bdf7f7b142fddbeb3f62e160e4a1b3a639b308717964e6", + "summary": { + "admission_lag_max_ms": 0.14346837997436523, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 341.1952300230041, + "ttft_p50_ms": 294.81438896618783, + "ttft_p95_ms": 336.8800369789824 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r7p00.json", + "sha256": "b0fdc1ef617313261c490010adeb2f4ee93bbb053f1db10734ed5a63fd37fa76", + "summary": { + "admission_lag_max_ms": 0.11047185398638248, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 334.06115300022066, + "ttft_p50_ms": 281.1476809438318, + "ttft_p95_ms": 329.65395506471395 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + true + ], + "round_ttft_p95_ms": [ + 1289.2397560644895, + 1254.356418037787 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r8p00.json", + "sha256": "8877ed44b90f421afd22ec110fe6337581c67cd973b44378bcb7528fc8624de3", + "summary": { + "admission_lag_max_ms": 0.09899691212922335, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.9375, + "passed": 60, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1454.3529660440981, + "ttft_p50_ms": 813.0663179326802, + "ttft_p95_ms": 1289.2397560644895 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r8p00.json", + "sha256": "7ab24f5ab33bf166589af7e2f5c2a274ceef6a64f0029a96ec1879b42332123d", + "summary": { + "admission_lag_max_ms": 0.11694896966218948, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 0.953125, + "passed": 61, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1347.0055809011683, + "ttft_p50_ms": 744.0344559727237, + "ttft_p95_ms": 1254.356418037787 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4903.538785991259, + 4900.28300601989 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r16p00.json", + "sha256": "363d9b43164b5451bc3dd668caa0ce141d15114087cf444e7fd79d12b7563fd6", + "summary": { + "admission_lag_max_ms": 0.11420203372836113, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4982.48829995282, + "ttft_p50_ms": 2749.9175920384005, + "ttft_p95_ms": 4903.538785991259 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r16p00.json", + "sha256": "8aec70dfd6b69f4903bccc02c5d5474165ed07d9dd010c860ff9b447429da950", + "summary": { + "admission_lag_max_ms": 0.19414699636399746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4979.007771937177, + "ttft_p50_ms": 2749.115635990165, + "ttft_p95_ms": 4900.28300601989 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6760.187636013143, + 6534.733187989332 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r32p00.json", + "sha256": "7684ac0d2d6f851218b88fb37db3a5b54d877a6f1ead3bb604a6cb4d9a61cf3f", + "summary": { + "admission_lag_max_ms": 0.43534801807254553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6907.940509961918, + "ttft_p50_ms": 3735.540736000985, + "ttft_p95_ms": 6760.187636013143 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r32p00.json", + "sha256": "947974a9ae04e170b8c1aad32ddbda8c1000f3902a9b572c6338a7d0d4f23146", + "summary": { + "admission_lag_max_ms": 1.221106038428843, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6798.48151106853, + "ttft_p50_ms": 3595.78589303419, + "ttft_p95_ms": 6534.733187989332 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7754.704045015387, + 7756.428897031583 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r64p00.json", + "sha256": "abfc2a158087f0ad75b560dd420604a9f81359ffdf39f0ce8e2ea25b451e6904", + "summary": { + "admission_lag_max_ms": 0.13654399663209915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7857.916672946885, + "ttft_p50_ms": 4341.527819051407, + "ttft_p95_ms": 7754.704045015387 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r64p00.json", + "sha256": "192c0c850a52d2b64780bc6321cbfd10faddcc0ff10741ddad283b7d789b9def", + "summary": { + "admission_lag_max_ms": 0.11000398080796003, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7859.458995983005, + "ttft_p50_ms": 4341.894763987511, + "ttft_p95_ms": 7756.428897031583 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp1_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 99.1211449727416, + 101.08128399588168 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r4p00.json", + "sha256": "f149882da390114f2a2a140a3bd2ca6f15b6f7761591ea7b49b5a961c9d18df5", + "summary": { + "admission_lag_max_ms": 0.10085199028253555, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.28417506348342, + "ttft_p50_ms": 95.90165293775499, + "ttft_p95_ms": 99.1211449727416 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r4p00.json", + "sha256": "154dfaf2729001a8ef5b7430290f4ca906df1d4758e867d5b350489410f36d5c", + "summary": { + "admission_lag_max_ms": 0.11863000690937042, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 105.76263803523034, + "ttft_p50_ms": 96.27402503974736, + "ttft_p95_ms": 101.08128399588168 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.1850419882685, + 97.03529300168157 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r8p00.json", + "sha256": "9338d5cb994b0c12a6420f030ae2a9335a04986b4a81e31f4781ac838aeef2d4", + "summary": { + "admission_lag_max_ms": 0.12352492194622755, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.26434096693993, + "ttft_p50_ms": 95.07436107378453, + "ttft_p95_ms": 97.1850419882685 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r8p00.json", + "sha256": "a35b67970948976482ee835e595ee5631fa921607276d5ea7928460514f0d06b", + "summary": { + "admission_lag_max_ms": 0.11170096695423126, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.72453604172915, + "ttft_p50_ms": 94.03157397173345, + "ttft_p95_ms": 97.03529300168157 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.10817302018404, + 94.37024802900851 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r10p00.json", + "sha256": "f67e0f4f6b41861fcbcbee04b4efbf0dee46e273239ec94ef22d650ba6afa22a", + "summary": { + "admission_lag_max_ms": 0.10012194979935884, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.40214702207595, + "ttft_p50_ms": 94.17163301259279, + "ttft_p95_ms": 96.10817302018404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r10p00.json", + "sha256": "5a64b52f0298e9ecda2fddb8230cc27ab4b9c1db01575f68b5331816b324d1d7", + "summary": { + "admission_lag_max_ms": 0.10465399827808142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.07159405015409, + "ttft_p50_ms": 93.24954205658287, + "ttft_p95_ms": 94.37024802900851 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 447.0475659472868, + 454.8899739747867 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r12p00.json", + "sha256": "c6e27b9078281a54feeb1fab09616385a70de819791ce14f46b001bf540a80ab", + "summary": { + "admission_lag_max_ms": 0.12051803059875965, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 455.56937297806144, + "ttft_p50_ms": 280.09015799034387, + "ttft_p95_ms": 447.0475659472868 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r12p00.json", + "sha256": "861f9c4186c33e9ee6365d19c53e98c10615c5ee529a84e69f84dc292f53ac4f", + "summary": { + "admission_lag_max_ms": 0.11322635691612959, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 457.4526828946546, + "ttft_p50_ms": 290.1579710887745, + "ttft_p95_ms": 454.8899739747867 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1000.5082850111648, + 1004.6314439969137 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r14p00.json", + "sha256": "2dcaf1087edd1f12fbfe0618a6ab636b22484756a5729919bc0e51de153f8f4a", + "summary": { + "admission_lag_max_ms": 0.1881921198219061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1089.2393708927557, + "ttft_p50_ms": 611.2175739835948, + "ttft_p95_ms": 1000.5082850111648 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r14p00.json", + "sha256": "16937b31f72298c7a50ec3c7b9c1b9c798f305d83469b3d3c058efc4d53135da", + "summary": { + "admission_lag_max_ms": 0.10659568943083286, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1096.6680150013417, + "ttft_p50_ms": 617.3943240428343, + "ttft_p95_ms": 1004.6314439969137 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1458.4544260287657, + 1471.3564389385283 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r16p00.json", + "sha256": "5576a7fd06977f51f1c43bb73c3c474cc21a93cd288b422ec3d09db671b3afe7", + "summary": { + "admission_lag_max_ms": 0.135264010168612, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1583.2410319708288, + "ttft_p50_ms": 904.5663280412555, + "ttft_p95_ms": 1458.4544260287657 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r16p00.json", + "sha256": "2ccf16426046c412829442f72932268d5b27ce051a4312d135d32b240b48ec7c", + "summary": { + "admission_lag_max_ms": 0.11741300113499165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.78125, + "passed": 50, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1537.9131329245865, + "ttft_p50_ms": 896.1611289996654, + "ttft_p95_ms": 1471.3564389385283 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3348.027400090359, + 3348.575382027775 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r32p00.json", + "sha256": "d258760d15f1fdf2adefca72f7a3505c0b839036799dcb074450f7f526f76718", + "summary": { + "admission_lag_max_ms": 0.18610397819429636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3442.3487790627405, + "ttft_p50_ms": 1801.6119640087709, + "ttft_p95_ms": 3348.027400090359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r32p00.json", + "sha256": "edf5cccc1400639a94b5430897a04f20d2214bb1ffee7f4bfa62514150179a6a", + "summary": { + "admission_lag_max_ms": 0.1144439447671175, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3443.0311269825324, + "ttft_p50_ms": 1801.435066969134, + "ttft_p95_ms": 3348.575382027775 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4286.911493982188, + 4287.898648995906 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r64p00.json", + "sha256": "2a6303efb826637c3c6448d6f766f7de4fc4d7b63db163a48f72c20df68394f4", + "summary": { + "admission_lag_max_ms": 0.10997895151376724, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.311317006126, + "ttft_p50_ms": 2396.4726959820837, + "ttft_p95_ms": 4286.911493982188 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r64p00.json", + "sha256": "6fafe3bf206583b8689be1b2b57d6b92cbd1ba54f7fb230ca05f6c68b3ab9757", + "summary": { + "admission_lag_max_ms": 0.12669397983700037, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.538130956702, + "ttft_p50_ms": 2394.112486974336, + "ttft_p95_ms": 4287.898648995906 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp2_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.2925309324637, + 96.13083000294864 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r4p00.json", + "sha256": "812f8a108a9390fb759b46e5bd41e3ddad66bc144d47bf4024d834e42b47c47e", + "summary": { + "admission_lag_max_ms": 0.11386198457330465, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.76773096807301, + "ttft_p50_ms": 95.72782705072314, + "ttft_p95_ms": 97.2925309324637 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r4p00.json", + "sha256": "70f922eb7651d46d81d7d394b8453c5b778f921069cf48fb2ae2677afa94a842", + "summary": { + "admission_lag_max_ms": 0.30313804745674133, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 129.06646099872887, + "ttft_p50_ms": 94.53632694203407, + "ttft_p95_ms": 96.13083000294864 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 95.42673197574914, + 98.94769196398556 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r8p00.json", + "sha256": "24097d38b0ffb45f6cdbd93f0aad04dce3b7da41ff45e948eb3a794902b6547f", + "summary": { + "admission_lag_max_ms": 0.11571904178708792, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.04934400785714, + "ttft_p50_ms": 94.34181300457567, + "ttft_p95_ms": 95.42673197574914 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r8p00.json", + "sha256": "7251f076d34ba9484b08fa3fc4b898d2ba8d1199cb1eaeaaffc19930f8c889b9", + "summary": { + "admission_lag_max_ms": 0.1177559606730938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.93720292206854, + "ttft_p50_ms": 94.77824496570975, + "ttft_p95_ms": 98.94769196398556 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.28185909241438, + 95.9081610199064 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r10p00.json", + "sha256": "85796457715a25c15fc909f6def32776fbd49956eb7890df1c4abcef8c5afe78", + "summary": { + "admission_lag_max_ms": 0.14786410611122847, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.46025596279651, + "ttft_p50_ms": 93.54631893802434, + "ttft_p95_ms": 96.28185909241438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r10p00.json", + "sha256": "0bf4d4d006bf0b5bd8b989ceeeb4bacb0e3ee1b8ea66f30b3deb120a14871cf0", + "summary": { + "admission_lag_max_ms": 0.11162098962813616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.54603894241154, + "ttft_p50_ms": 93.77408598084003, + "ttft_p95_ms": 95.9081610199064 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 439.0145930228755, + 429.9719400005415 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r12p00.json", + "sha256": "68de6e36bd7c5624cab107fe5cb21bb817711691e545439a958c926a27e3fe29", + "summary": { + "admission_lag_max_ms": 0.10475562885403633, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 445.90794993564487, + "ttft_p50_ms": 270.3119309153408, + "ttft_p95_ms": 439.0145930228755 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r12p00.json", + "sha256": "d8d3a44a07b46f1673817ece2948fd97c30f41bd451d93a2fa02edf074536c63", + "summary": { + "admission_lag_max_ms": 0.4323029424995184, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 434.6071509644389, + "ttft_p50_ms": 265.2625450864434, + "ttft_p95_ms": 429.9719400005415 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1014.2896320903674, + 999.8802930349484 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r14p00.json", + "sha256": "10196c2f9a9b69c2e47f64d06bf5ee6863af4cd2b612301622f0e304949a942a", + "summary": { + "admission_lag_max_ms": 0.09005970787256956, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1105.1907669752836, + "ttft_p50_ms": 621.6445650206879, + "ttft_p95_ms": 1014.2896320903674 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r14p00.json", + "sha256": "c10cab85588a3563bed36fc4480462635d34d240938beaa0e5034f13186578a6", + "summary": { + "admission_lag_max_ms": 0.11327012907713652, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1090.6395060010254, + "ttft_p50_ms": 623.5586369875818, + "ttft_p95_ms": 999.8802930349484 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1488.3801550604403, + 1484.6198010491207 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r16p00.json", + "sha256": "ef5046c49c32178855bed6a0488ea21e8da22f786ddaabc503bea2d1025bb2f3", + "summary": { + "admission_lag_max_ms": 0.7892940193414688, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1613.616122980602, + "ttft_p50_ms": 905.2404230460525, + "ttft_p95_ms": 1488.3801550604403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r16p00.json", + "sha256": "de02c1d49414cffe7e20023abf4d502a4efc6d2e01fcb3e49e928081aaef77b6", + "summary": { + "admission_lag_max_ms": 0.10548799764364958, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1550.9842830942944, + "ttft_p50_ms": 904.9503420246765, + "ttft_p95_ms": 1484.6198010491207 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3391.811708919704, + 3362.9186518955976 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r32p00.json", + "sha256": "876c5444d17c4ce81eca907ae17d5cf7062ffa67fa266b491580e52f712614ec", + "summary": { + "admission_lag_max_ms": 0.12557301670312881, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3486.074637970887, + "ttft_p50_ms": 1840.204154024832, + "ttft_p95_ms": 3391.811708919704 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r32p00.json", + "sha256": "02b4cd93ce1f9816e822022fb608ba8d4c89265f673f6d27c8fa3d4f1523ffe0", + "summary": { + "admission_lag_max_ms": 0.11014204937964678, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3456.965696066618, + "ttft_p50_ms": 1812.343467026949, + "ttft_p95_ms": 3362.9186518955976 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4306.351027917117, + 4302.602957002819 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r64p00.json", + "sha256": "13ba282f179b2d6ce13a0aa2fd04eb29f82e579b7995e3406cb0afd69a2d5ee2", + "summary": { + "admission_lag_max_ms": 0.3526840591803193, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4359.505696920678, + "ttft_p50_ms": 2411.1130589153618, + "ttft_p95_ms": 4306.351027917117 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r64p00.json", + "sha256": "767622d62f0757ef2d8c7651ea470bfa2e62edad0ed4e75f66f84a0b722e35da", + "summary": { + "admission_lag_max_ms": 0.10626309085637331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4355.409953044727, + "ttft_p50_ms": 2409.5974090741947, + "ttft_p95_ms": 4302.602957002819 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp2_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.54812794178724, + 95.08875198662281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r4p00.json", + "sha256": "6e6a45d0c1010094c3395969815fd8cf9c35b919db202d772fb514744f981ebc", + "summary": { + "admission_lag_max_ms": 0.11306197848170996, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.27749610505998, + "ttft_p50_ms": 93.01548008807003, + "ttft_p95_ms": 96.54812794178724 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r4p00.json", + "sha256": "58aa71d694704cfea15c37cb29420dbac1dc505045c29a731e8be904846cca5f", + "summary": { + "admission_lag_max_ms": 0.11853000614792109, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 98.74943597242236, + "ttft_p50_ms": 93.57558900956064, + "ttft_p95_ms": 95.08875198662281 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.24941800534725, + 96.21043398510665 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r8p00.json", + "sha256": "311350af236cb0306aa635f05f8999b66bf5080bad7cc1f6703fffe0be31aee6", + "summary": { + "admission_lag_max_ms": 0.19771000370383263, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52211803849787, + "ttft_p50_ms": 92.4453770276159, + "ttft_p95_ms": 94.24941800534725 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r8p00.json", + "sha256": "60fe989a36a526e76f6f5c98e217b9feaf081bc680b16b9d53650bf085047c06", + "summary": { + "admission_lag_max_ms": 0.3933720290660858, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.97230398282409, + "ttft_p50_ms": 93.26327103190124, + "ttft_p95_ms": 96.21043398510665 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.98661107197404, + 92.9473100695759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r10p00.json", + "sha256": "d0a04444a0e41bf76f71ac6ea3815bfbe76f7a51f8ddf669976bc3a4896dd0c1", + "summary": { + "admission_lag_max_ms": 0.10270101483911276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 103.49742788821459, + "ttft_p50_ms": 94.55623896792531, + "ttft_p95_ms": 97.98661107197404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r10p00.json", + "sha256": "f330f26f45f6039c48150a0b6b146ca3dbb9ea8b2dff086a9efd5938ab981eb7", + "summary": { + "admission_lag_max_ms": 0.11239596642553806, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.4133670097217, + "ttft_p50_ms": 91.2688790122047, + "ttft_p95_ms": 92.9473100695759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 260.68170997314155, + 319.04888805001974 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r12p00.json", + "sha256": "fed5eacf18018402b0c2675b8c5595f92693a9bddec30f354d989098f9fe4965", + "summary": { + "admission_lag_max_ms": 0.11784560047090054, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 345.19017208367586, + "ttft_p50_ms": 211.41935500781983, + "ttft_p95_ms": 260.68170997314155 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r12p00.json", + "sha256": "f2be27cbeff3e5c1a95050418f2dc3c712a4648cada8cdb3d36382cfffc9593a", + "summary": { + "admission_lag_max_ms": 0.12930459342896938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 326.86293800361454, + "ttft_p50_ms": 229.8808820778504, + "ttft_p95_ms": 319.04888805001974 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 895.7287209341303, + 912.1733580250293 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r14p00.json", + "sha256": "c2e157808e0308cc443f3abdf5114898cef66c435688d24321f80bef2d966208", + "summary": { + "admission_lag_max_ms": 0.14902697876095772, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 962.3544370988384, + "ttft_p50_ms": 576.2168710352853, + "ttft_p95_ms": 895.7287209341303 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r14p00.json", + "sha256": "749ef7d31bf8782d42719b75ee1a2e4a443f097510af721d02cf53df191bc66e", + "summary": { + "admission_lag_max_ms": 0.42896345257759094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 986.0872740391642, + "ttft_p50_ms": 569.7114520007744, + "ttft_p95_ms": 912.1733580250293 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1364.6171110449359, + 1366.6857269126922 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r16p00.json", + "sha256": "7545ea9210348019de88ede5d2cba7527913534274deee034fa6811c173a6e72", + "summary": { + "admission_lag_max_ms": 0.12612901628017426, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1435.8308439841494, + "ttft_p50_ms": 848.9949750946835, + "ttft_p95_ms": 1364.6171110449359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r16p00.json", + "sha256": "d75e27137f6a7811126d8c957bb5a081944fc1bb7643852cdd785111773f6010", + "summary": { + "admission_lag_max_ms": 0.1258660340681672, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1436.2916270038113, + "ttft_p50_ms": 846.2129919789732, + "ttft_p95_ms": 1366.6857269126922 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3243.9196659252048, + 3214.524411014281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r32p00.json", + "sha256": "366ebddc167338ed096f373de5667740721422c13d2db1f887ed09186afafd19", + "summary": { + "admission_lag_max_ms": 0.09669095743447542, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3337.068362045102, + "ttft_p50_ms": 1751.3202880509198, + "ttft_p95_ms": 3243.9196659252048 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r32p00.json", + "sha256": "01b8f5081e36a64d5776285c8c5f423b53b1ad684c4b8f95a8caa5d997c6bf09", + "summary": { + "admission_lag_max_ms": 0.8310690755024552, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3277.297410997562, + "ttft_p50_ms": 1790.5721209244803, + "ttft_p95_ms": 3214.524411014281 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4319.725521025248, + 4367.031275993213 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r64p00.json", + "sha256": "92202709fd6bbd558ff6f1986df213848254110c4274447a7bd5879153290d46", + "summary": { + "admission_lag_max_ms": 0.10392803233116865, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4360.145599930547, + "ttft_p50_ms": 2403.5812759539112, + "ttft_p95_ms": 4319.725521025248 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r64p00.json", + "sha256": "9660d15d31bc70c5550ed29477f563313d4d81baf0e561126aebd3b7d097cfa3", + "summary": { + "admission_lag_max_ms": 0.17276196740567684, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.25, + "passed": 16, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4414.442001027055, + "ttft_p50_ms": 2260.2788450894877, + "ttft_p95_ms": 4367.031275993213 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp2_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.17482794076204, + 98.03147299680859 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r4p00.json", + "sha256": "0e315f6d9e0087ad553e874b0c16c2e6de79eb6eca22eb118dc5deca52e6efe3", + "summary": { + "admission_lag_max_ms": 0.11708191595971584, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.92840604577214, + "ttft_p50_ms": 95.13642196543515, + "ttft_p95_ms": 97.17482794076204 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r4p00.json", + "sha256": "baa394a24d2a3db40a3a7ae6dae0c5ce49ff05499a1e17eaa0bca4b479ae33bf", + "summary": { + "admission_lag_max_ms": 0.10946101974695921, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52333807107061, + "ttft_p50_ms": 93.19468098692596, + "ttft_p95_ms": 98.03147299680859 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.67980605829507, + 93.80627400241792 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r8p00.json", + "sha256": "303df27c6d4604d5af65f9056acacb7c4373fb6eb7edb6e5d42689eeaa384aaa", + "summary": { + "admission_lag_max_ms": 0.11111306957900524, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 113.2329449756071, + "ttft_p50_ms": 93.05417106952518, + "ttft_p95_ms": 97.67980605829507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r8p00.json", + "sha256": "cdbb1bca17b24b67e18ef3dd4997494d6bd2a887b3b5d1264fbf411e688c53bc", + "summary": { + "admission_lag_max_ms": 0.12463307939469814, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.12245299387723, + "ttft_p50_ms": 92.26030297577381, + "ttft_p95_ms": 93.80627400241792 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.45723297540098, + 93.57301203999668 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r10p00.json", + "sha256": "fe17ef2417319307d651bc10b01a618234d61bab797ad87a7332cab20d78ecb2", + "summary": { + "admission_lag_max_ms": 0.3267280990257859, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.28593400400132, + "ttft_p50_ms": 92.22474799025804, + "ttft_p95_ms": 94.45723297540098 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r10p00.json", + "sha256": "550f693708a34624a82d77afd4da0ab4bc01e09f42242a3a79e4532897681a51", + "summary": { + "admission_lag_max_ms": 0.11238199658691883, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.77756198309362, + "ttft_p50_ms": 92.07128698471934, + "ttft_p95_ms": 93.57301203999668 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 320.49866404850036, + 336.9630330707878 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r12p00.json", + "sha256": "edab8169453151de732923c076b8f7dca4082ee47ccd2597e65e7ba68516bc2e", + "summary": { + "admission_lag_max_ms": 0.09914662223309278, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 327.930542989634, + "ttft_p50_ms": 231.4880370395258, + "ttft_p95_ms": 320.49866404850036 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r12p00.json", + "sha256": "a6dde4363d362ca10254409e3dc734d0f0e640e63156426f3f4fd7115db49e16", + "summary": { + "admission_lag_max_ms": 0.17016706988215446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 428.6657409975305, + "ttft_p50_ms": 237.09688894450665, + "ttft_p95_ms": 336.9630330707878 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 911.7195709841326, + 910.7652489328757 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r14p00.json", + "sha256": "7eba81ee5c85c457a056f9aef28abd94d375fdaa116929cfe0a0c282a0738bd2", + "summary": { + "admission_lag_max_ms": 0.13163150288164616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 999.2953530745581, + "ttft_p50_ms": 580.023527960293, + "ttft_p95_ms": 911.7195709841326 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r14p00.json", + "sha256": "6a113d3d876ff01a9c1101e8b75efa2fc71f96683d6bb430e26d5976426b5d23", + "summary": { + "admission_lag_max_ms": 0.11634768452495337, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 988.8870939612389, + "ttft_p50_ms": 578.9327350212261, + "ttft_p95_ms": 910.7652489328757 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1389.8832038976252, + 1365.0342189939693 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r16p00.json", + "sha256": "d677251ee1a941c52469052a239c2e13be0a837285069aa8182a16e196445e48", + "summary": { + "admission_lag_max_ms": 0.11070503387600183, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1460.5093500576913, + "ttft_p50_ms": 864.3530959961936, + "ttft_p95_ms": 1389.8832038976252 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r16p00.json", + "sha256": "f5177f53aa0f87a4e7291f470978d86ad8bc46bd1a8362b79f48cd128646dc10", + "summary": { + "admission_lag_max_ms": 0.3672370221465826, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1491.050804965198, + "ttft_p50_ms": 841.7225609300658, + "ttft_p95_ms": 1365.0342189939693 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3258.010295103304, + 3258.664960041642 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r32p00.json", + "sha256": "02a6ac2b9c2b9ab97854d6fcc13b64bbb36a079e4afdae84231e4cb187fac4ef", + "summary": { + "admission_lag_max_ms": 0.1296499976888299, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.278529899195, + "ttft_p50_ms": 1755.0864539807662, + "ttft_p95_ms": 3258.010295103304 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r32p00.json", + "sha256": "9ca2f44fc7e3cb3d38ada548af858903b31bbee5cd50f1723c29255ebd4224e9", + "summary": { + "admission_lag_max_ms": 1.1343059595674276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.7101600775495, + "ttft_p50_ms": 1755.3453800501302, + "ttft_p95_ms": 3258.664960041642 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4190.255103982054, + 4189.1050330596045 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r64p00.json", + "sha256": "53908d3a254e19691839831b4337014b09591ced9189c2a9b29e0d92e80fcb56", + "summary": { + "admission_lag_max_ms": 0.11591101065278053, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4242.029857938178, + "ttft_p50_ms": 2340.5529640149325, + "ttft_p95_ms": 4190.255103982054 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r64p00.json", + "sha256": "1c8c85c873aa96c01dc0200f848474a53c1173ef72f3828e8c23aeae70e92b11", + "summary": { + "admission_lag_max_ms": 0.284243025816977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4240.770195960067, + "ttft_p50_ms": 2339.3583690049127, + "ttft_p95_ms": 4189.1050330596045 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp2_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 74.66826005838811, + 66.5705680148676 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r4p00.json", + "sha256": "fb354fc8a097caa21844d4ea5515b98b03459f400816e5e5590a6cda81cf4815", + "summary": { + "admission_lag_max_ms": 0.1035250024870038, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 93.38432096410543, + "ttft_p50_ms": 65.75745099689811, + "ttft_p95_ms": 74.66826005838811 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r4p00.json", + "sha256": "ac8e8412f79a8a32ebf085445a94aa548dd2d646fce0b9f56b4378081dcc8960", + "summary": { + "admission_lag_max_ms": 0.12108997907489538, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.89856600761414, + "ttft_p50_ms": 62.386377016082406, + "ttft_p95_ms": 66.5705680148676 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.44799195136875, + 72.29184894822538 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r8p00.json", + "sha256": "05ebbd68bbf24e9c8b25684474411c7db122b92701cc1617f9d786cf19592751", + "summary": { + "admission_lag_max_ms": 0.1825150102376938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.92095802165568, + "ttft_p50_ms": 60.598687967285514, + "ttft_p95_ms": 70.44799195136875 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r8p00.json", + "sha256": "d24f7daaf41073cc8aa030b74c06203783684c3e556770366883c37b826406f2", + "summary": { + "admission_lag_max_ms": 0.10628998279571533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 75.13270596973598, + "ttft_p50_ms": 60.161596979014575, + "ttft_p95_ms": 72.29184894822538 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.02151494007558, + 107.58775402791798 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r16p00.json", + "sha256": "a29b2128983af630e1da8f970cbda802849036bc522eb506821bf2fe1a34172f", + "summary": { + "admission_lag_max_ms": 0.10468997061252594, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.93136999942362, + "ttft_p50_ms": 59.5597909996286, + "ttft_p95_ms": 66.02151494007558 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r16p00.json", + "sha256": "275ae35651cbe2200d16c9b06a68568efda527564065255a10dee6fe93346010", + "summary": { + "admission_lag_max_ms": 0.3144120564684272, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 119.78317995090038, + "ttft_p50_ms": 59.12985489703715, + "ttft_p95_ms": 107.58775402791798 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 193.8455159543082, + 187.52167106140405 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r20p00.json", + "sha256": "4cb485f30b81c004bab379e6522d94b811c0bdc0be6d8e42c9d8ae842d9600dc", + "summary": { + "admission_lag_max_ms": 0.10407401714473963, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 197.7908629924059, + "ttft_p50_ms": 142.73560303263366, + "ttft_p95_ms": 193.8455159543082 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r20p00.json", + "sha256": "63aa9dc03a08b781c5d6e8805f2b61698c75e71673608ecd38bd1132c62e6987", + "summary": { + "admission_lag_max_ms": 0.09386602323502302, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 190.29632199089974, + "ttft_p50_ms": 136.1723990412429, + "ttft_p95_ms": 187.52167106140405 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 568.2198900030926, + 573.4539540717378 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r24p00.json", + "sha256": "5e4d5b44850090d07fe1d3abc44eddea3bfdf2b682f59b12a45a866621e38f98", + "summary": { + "admission_lag_max_ms": 0.7511250441893935, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 613.2891150191426, + "ttft_p50_ms": 375.27791305910796, + "ttft_p95_ms": 568.2198900030926 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r24p00.json", + "sha256": "239e7789aed2e9b2bcf1d127215a6b122071eb856be2db64ccc184fda153aa48", + "summary": { + "admission_lag_max_ms": 0.11180061846971512, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 614.9205249967054, + "ttft_p50_ms": 375.5133739905432, + "ttft_p95_ms": 573.4539540717378 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 860.0242879474536, + 886.1482660286129 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r28p00.json", + "sha256": "8b249b52779e5b3cd9028aaee30ee73ef34b0b0a64dcd75b9aad57266c29fbc2", + "summary": { + "admission_lag_max_ms": 0.11425174307078123, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.8854410191998, + "ttft_p50_ms": 534.7980660153553, + "ttft_p95_ms": 860.0242879474536 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r28p00.json", + "sha256": "c9b28c8f3187542a343ecf1067783569ec2abfa64631d6827a94ba6876aa7b63", + "summary": { + "admission_lag_max_ms": 0.312068616040051, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 922.4315020255744, + "ttft_p50_ms": 549.7413279954344, + "ttft_p95_ms": 886.1482660286129 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1147.1082780044526, + 1137.236590962857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r32p00.json", + "sha256": "d27ff9f940a6cd7fba339d9a176de39cf4b5bd164dcc500f2af36bcb07cdef30", + "summary": { + "admission_lag_max_ms": 0.16146607231348753, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.626160996966, + "ttft_p50_ms": 696.4708579471335, + "ttft_p95_ms": 1147.1082780044526 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r32p00.json", + "sha256": "093044e87fffdaa222e5bad3807cbf4900aa64891d5b4ba27f187c7a411252d7", + "summary": { + "admission_lag_max_ms": 0.10430300608277321, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.045956928283, + "ttft_p50_ms": 684.3201500596479, + "ttft_p95_ms": 1137.236590962857 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2064.763988018967, + 2039.6780440350994 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r64p00.json", + "sha256": "fad7bdd80d5ca603790a4e5d19dadd4f494540b106ce583e3e8c1c05d2374a58", + "summary": { + "admission_lag_max_ms": 6.394341005943716, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.53125, + "passed": 34, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2144.80376499705, + "ttft_p50_ms": 1158.5425049997866, + "ttft_p95_ms": 2064.763988018967 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r64p00.json", + "sha256": "24fdb945706dbbdd29916e0f08e0eff807334db212fded94c729da2e1b46efd3", + "summary": { + "admission_lag_max_ms": 0.10294502135366201, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.4275270039216, + "ttft_p50_ms": 1131.1213460285217, + "ttft_p95_ms": 2039.6780440350994 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.80304794665426, + 64.59337309934199 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r4p00.json", + "sha256": "2e82a30f02c2702668e59661930431cf82323746a111d42bf38b7fdc0d083b3d", + "summary": { + "admission_lag_max_ms": 0.1332589890807867, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.74050498381257, + "ttft_p50_ms": 64.10286005120724, + "ttft_p95_ms": 67.80304794665426 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r4p00.json", + "sha256": "8294196c7dc52765a6820f5e1d8439c480490daed9505263f707810f73c95783", + "summary": { + "admission_lag_max_ms": 0.1283179735764861, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.6796320155263, + "ttft_p50_ms": 60.962425894103944, + "ttft_p95_ms": 64.59337309934199 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.46496798284352, + 68.22343391831964 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r8p00.json", + "sha256": "38bbfeab47e5eb6296fcccf20f04ef25dc24db658be92300c492734bbfe333cf", + "summary": { + "admission_lag_max_ms": 0.10555307380855083, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 87.7126749837771, + "ttft_p50_ms": 62.65173899009824, + "ttft_p95_ms": 70.46496798284352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r8p00.json", + "sha256": "716f90177d6f203717c80a8e9b28697e53141f204af470493c299bc73a1ea3ee", + "summary": { + "admission_lag_max_ms": 0.24518708232790232, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 80.40577603969723, + "ttft_p50_ms": 60.29119901359081, + "ttft_p95_ms": 68.22343391831964 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 89.53388908412308, + 76.93439698778093 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r16p00.json", + "sha256": "063016efe822c31c2425e7d963eda4c054f3f7c0341df0a3cd73ab86a942e873", + "summary": { + "admission_lag_max_ms": 0.2996840048581362, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.41822991054505, + "ttft_p50_ms": 60.97789097111672, + "ttft_p95_ms": 89.53388908412308 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r16p00.json", + "sha256": "eecc6bdeff90a8bf195dfc5387c6bc2f4db20ed7c9aeb9b4ebe050efccc6051a", + "summary": { + "admission_lag_max_ms": 0.36267400719225407, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.42048599943519, + "ttft_p50_ms": 60.29245804529637, + "ttft_p95_ms": 76.93439698778093 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 206.2494510319084, + 201.13301498349756 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r20p00.json", + "sha256": "22a62729358fea7c38ca166a1a49f1235e98c7f07ecfe4948322110cda4b7178", + "summary": { + "admission_lag_max_ms": 0.11367700062692165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 255.14698098413646, + "ttft_p50_ms": 155.46583198010921, + "ttft_p95_ms": 206.2494510319084 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r20p00.json", + "sha256": "be103a759100bd2ab6dd31ee4ce41b6f0da40d232a60ebe137e3d933194a487c", + "summary": { + "admission_lag_max_ms": 0.17853605095297098, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 209.22932599205524, + "ttft_p50_ms": 155.12292610947043, + "ttft_p95_ms": 201.13301498349756 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 580.5047769099474, + 577.3971510352567 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r24p00.json", + "sha256": "890a733a97f399deb4c5efcca661d9a442ed76080cd5d8b6a2c09368aef701b7", + "summary": { + "admission_lag_max_ms": 0.1084543764591217, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 621.5670789824799, + "ttft_p50_ms": 374.6346529806033, + "ttft_p95_ms": 580.5047769099474 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r24p00.json", + "sha256": "87e8543652b1743b1b9b8c3de0fc680401b0bf9a33d61fa6132120c944b2a6eb", + "summary": { + "admission_lag_max_ms": 0.11801638174802065, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 623.2662099646404, + "ttft_p50_ms": 382.11427396163344, + "ttft_p95_ms": 577.3971510352567 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 873.0683700414374, + 889.4930009264499 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r28p00.json", + "sha256": "f877781730a5437b4cbdb76948915f1c871574b19a6df27076d21b65bb18e678", + "summary": { + "admission_lag_max_ms": 0.6807943573221564, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 943.9258739585057, + "ttft_p50_ms": 542.9852310335264, + "ttft_p95_ms": 873.0683700414374 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r28p00.json", + "sha256": "82151dc539b35302489c17976cfe89b8f4c64944298954607fad8f5c9cb9e7d8", + "summary": { + "admission_lag_max_ms": 6.750226486474276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 936.5667289821431, + "ttft_p50_ms": 546.0105559322983, + "ttft_p95_ms": 889.4930009264499 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1135.3467140579596, + 1146.743991994299 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r32p00.json", + "sha256": "082cc4c47b2eec6e49f0edd14cff5a51b880bd998d98da8b9830bbf7615483ef", + "summary": { + "admission_lag_max_ms": 0.10443700011819601, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1199.8225760180503, + "ttft_p50_ms": 680.4794340860099, + "ttft_p95_ms": 1135.3467140579596 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r32p00.json", + "sha256": "517d3cbc1104b5ea1bdb0e381c033cf978181a93449b6498d30b52cd3dd3e770", + "summary": { + "admission_lag_max_ms": 0.4415180301293731, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1210.0602959981188, + "ttft_p50_ms": 693.0603379150853, + "ttft_p95_ms": 1146.743991994299 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2084.9693010095507, + 2041.9401369290426 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r64p00.json", + "sha256": "7e11771ed598826fb3c5594c7e05d0989e01030be2d8b36fa21813aff462496d", + "summary": { + "admission_lag_max_ms": 0.10768393985927105, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.128676981665, + "ttft_p50_ms": 1113.0280290963128, + "ttft_p95_ms": 2084.9693010095507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r64p00.json", + "sha256": "cf550423832a92cba942b7e1dc14a42000e8c9507403de8cd49aa56af9db6706", + "summary": { + "admission_lag_max_ms": 0.11262902989983559, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2122.8132860269397, + "ttft_p50_ms": 1131.2310450011864, + "ttft_p95_ms": 2041.9401369290426 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.89354998525232, + 83.23234610725194 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r4p00.json", + "sha256": "266cba4088f8a9483ceb41fdc02d04a2181cb7ffb6b52372b6b0687e62f2863e", + "summary": { + "admission_lag_max_ms": 0.10568101424723864, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.09669100120664, + "ttft_p50_ms": 62.014322029426694, + "ttft_p95_ms": 66.89354998525232 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r4p00.json", + "sha256": "26f3926c40504b96a37cb7b9dae1af5040c085c1a730e20feaf1d3e9ddbf8e54", + "summary": { + "admission_lag_max_ms": 0.108166947029531, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.42452490516007, + "ttft_p50_ms": 61.261708033271134, + "ttft_p95_ms": 83.23234610725194 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.26283801253885, + 67.34374002553523 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r8p00.json", + "sha256": "e84e6ba2c8822e012aceea7cc6f376ea1b6265fe4a3bb68b7badfc36b8813545", + "summary": { + "admission_lag_max_ms": 0.15694997273385525, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54619695153087, + "ttft_p50_ms": 59.63437003083527, + "ttft_p95_ms": 67.26283801253885 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r8p00.json", + "sha256": "dae50073c9f79919181e8dcf274ca8929ba41c512f64363e40187de68839b87e", + "summary": { + "admission_lag_max_ms": 0.09962497279047966, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.00161597877741, + "ttft_p50_ms": 60.2562150452286, + "ttft_p95_ms": 67.34374002553523 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 71.37343694921583, + 64.9534990079701 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r16p00.json", + "sha256": "776c9956f65acf7c9b6c8df517a4f8c4d462289893a19e625a2a0ad7dcbf5506", + "summary": { + "admission_lag_max_ms": 0.1066719414666295, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 82.25136005785316, + "ttft_p50_ms": 58.52357891853899, + "ttft_p95_ms": 71.37343694921583 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r16p00.json", + "sha256": "90057204a5f6343ee4ee9cdd08323d87d83660d5bf8f02480882eb479d2e587e", + "summary": { + "admission_lag_max_ms": 0.10102998930960894, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 65.3582609957084, + "ttft_p50_ms": 58.98014397826046, + "ttft_p95_ms": 64.9534990079701 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 192.98715889453888, + 258.1990590551868 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r20p00.json", + "sha256": "6ce029388b206d0663acd41f4da1d67a6562389e1c84cf1826ce77c05d176d9a", + "summary": { + "admission_lag_max_ms": 0.12691703159362078, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 198.67739104665816, + "ttft_p50_ms": 142.75032398290932, + "ttft_p95_ms": 192.98715889453888 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r20p00.json", + "sha256": "0dbfb3f012574988ddebfb0af1532c735ac62d6d88863d24cd820aed50a92a59", + "summary": { + "admission_lag_max_ms": 0.3961039474233985, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 259.55891504418105, + "ttft_p50_ms": 162.6282810466364, + "ttft_p95_ms": 258.1990590551868 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 583.3089610096067, + 571.3486529421061 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r24p00.json", + "sha256": "0774ba2f285d81447c63343b32551d6fc33369b54010629381efa9a37df23a9c", + "summary": { + "admission_lag_max_ms": 0.09822729043662548, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 628.714332007803, + "ttft_p50_ms": 366.9840869260952, + "ttft_p95_ms": 583.3089610096067 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r24p00.json", + "sha256": "1986a05c366ebe73af0f89d2cd0e4efd3e577c800cf1e80997cd26ea0aaaec09", + "summary": { + "admission_lag_max_ms": 0.11414836626499891, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 617.9165309295058, + "ttft_p50_ms": 376.65714998729527, + "ttft_p95_ms": 571.3486529421061 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 869.708820944652, + 866.4831441128626 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r28p00.json", + "sha256": "ab987729fc833a59817eabaa18c320b0d0a2ecd402fd02bdb534b6c773c029d8", + "summary": { + "admission_lag_max_ms": 0.578614417463541, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 941.3277570856735, + "ttft_p50_ms": 540.4732850147411, + "ttft_p95_ms": 869.708820944652 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r28p00.json", + "sha256": "38f9c82c9ee30927cc0ba857e5e3fce2e79b8a24de03efaec77ce27258633a7e", + "summary": { + "admission_lag_max_ms": 0.12532551772892475, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 938.2033230504021, + "ttft_p50_ms": 541.139594046399, + "ttft_p95_ms": 866.4831441128626 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1132.6635719742626, + 1144.196804962121 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r32p00.json", + "sha256": "358f5b8e1b72bff5752c076f2966b966d98f6f5767ce52d5a255b7f68f061e54", + "summary": { + "admission_lag_max_ms": 0.32093306072056293, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1197.7118930080906, + "ttft_p50_ms": 678.4568330040202, + "ttft_p95_ms": 1132.6635719742626 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r32p00.json", + "sha256": "3f6dbb760582fdd7470422d8edf6b62949d158ad0d437f04015441a0a4509003", + "summary": { + "admission_lag_max_ms": 1.1304179206490517, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1193.1529619032517, + "ttft_p50_ms": 692.5535890040919, + "ttft_p95_ms": 1144.196804962121 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2042.27216495201, + 2043.943466967903 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r64p00.json", + "sha256": "a24eae7db6eb4cd105ed903917988e8fe6eef5d4cc2549d4e05e9a03dfafb107", + "summary": { + "admission_lag_max_ms": 0.134881935082376, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.23939598538, + "ttft_p50_ms": 1137.3268909519538, + "ttft_p95_ms": 2042.27216495201 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r64p00.json", + "sha256": "e85efaada0b64bb9dd686f59282171ccb256a1d8483c6f037589a68e3e1a104a", + "summary": { + "admission_lag_max_ms": 0.11892011389136314, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.3299129894003, + "ttft_p50_ms": 1134.724210947752, + "ttft_p95_ms": 2043.943466967903 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 82.33969006687403, + 69.57727600820363 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r4p00.json", + "sha256": "e69b89ad2e07e6560ebd870773532b33e4415e9ca1515e585cbe465616678569", + "summary": { + "admission_lag_max_ms": 0.14186999760568142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54833806212991, + "ttft_p50_ms": 62.749656033702195, + "ttft_p95_ms": 82.33969006687403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r4p00.json", + "sha256": "2c76dcf112a828a6720e079d29bb54a6a65d37d26abaaafdc6d5580751dbad65", + "summary": { + "admission_lag_max_ms": 0.11478399392217398, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 81.3983449479565, + "ttft_p50_ms": 62.220132909715176, + "ttft_p95_ms": 69.57727600820363 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 63.71171900536865, + 68.14328604377806 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r8p00.json", + "sha256": "0e5d687dfcdb4b6a0d1d060ceefe8a74faab94a5b241580d45b53c452f893b23", + "summary": { + "admission_lag_max_ms": 0.09818794205784798, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.63772000744939, + "ttft_p50_ms": 60.89008797425777, + "ttft_p95_ms": 63.71171900536865 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r8p00.json", + "sha256": "1bf016b603473d46fda4bc6d4daeaee5e1091ff4cda1cab7224e0c5edea512f8", + "summary": { + "admission_lag_max_ms": 0.11399004142731428, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 76.76690397784114, + "ttft_p50_ms": 60.9520060243085, + "ttft_p95_ms": 68.14328604377806 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 133.86176899075508, + 61.95738597307354 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r16p00.json", + "sha256": "197808b1b5fa8082e603a255f358826b5f8f48707223c7fd63abfe029674b817", + "summary": { + "admission_lag_max_ms": 0.10530801955610514, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 191.68931001331657, + "ttft_p50_ms": 61.92394997924566, + "ttft_p95_ms": 133.86176899075508 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r16p00.json", + "sha256": "53e809470daf9efcdabb254e85e8cc3b45d0ab8593c6c2f3aa3396e60d1656da", + "summary": { + "admission_lag_max_ms": 0.1288399798795581, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.40882098954171, + "ttft_p50_ms": 59.417470009066164, + "ttft_p95_ms": 61.95738597307354 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 201.29013701807708, + 200.0734859611839 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r20p00.json", + "sha256": "05b977824fb301517cdbe2d077b360b6053d57fee1c20acad3537321827d027a", + "summary": { + "admission_lag_max_ms": 0.3848400665447116, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.5107699232176, + "ttft_p50_ms": 150.0862929970026, + "ttft_p95_ms": 201.29013701807708 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r20p00.json", + "sha256": "baf422b7a3c95b0d3d376a31b0a30e075f69216c794ebc2ef51afafe1ec55229", + "summary": { + "admission_lag_max_ms": 0.24754798505455256, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.8192109214142, + "ttft_p50_ms": 148.99871195666492, + "ttft_p95_ms": 200.0734859611839 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 564.2021730309352, + 560.4996709153056 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r24p00.json", + "sha256": "029b2a9695322d536539c3dd7665725f933b21a8bb48fa94c0448c9f7f8bf0eb", + "summary": { + "admission_lag_max_ms": 0.14956865925341845, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 610.4218009859324, + "ttft_p50_ms": 368.3909500250593, + "ttft_p95_ms": 564.2021730309352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r24p00.json", + "sha256": "94b46583af3a7a4bbb9f7256c4c4a5ce804b3092459abe3984ea0ce67f591f3c", + "summary": { + "admission_lag_max_ms": 0.11110503692179918, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 606.6705230623484, + "ttft_p50_ms": 367.50321893487126, + "ttft_p95_ms": 560.4996709153056 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 870.7952990662307, + 894.6108610834926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r28p00.json", + "sha256": "07bfe263107bb3d2acaa356257343f90045e4295c799178bc0a0b5f1c1994284", + "summary": { + "admission_lag_max_ms": 0.6798935355618596, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 942.792383953929, + "ttft_p50_ms": 543.340316042304, + "ttft_p95_ms": 870.7952990662307 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r28p00.json", + "sha256": "861dd55076ef40a78c34824345c5279dff86262b604db120e9e38efa27b5be55", + "summary": { + "admission_lag_max_ms": 0.09912776295095682, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.2741150604561, + "ttft_p50_ms": 549.9696529004723, + "ttft_p95_ms": 894.6108610834926 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1136.603599996306, + 1134.0074760373682 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r32p00.json", + "sha256": "332b805322b571087b018ee459dd91c443dfe0b2bb539b5bdfb4de902bba6da2", + "summary": { + "admission_lag_max_ms": 0.19067595712840557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.8779760692269, + "ttft_p50_ms": 683.3873559953645, + "ttft_p95_ms": 1136.603599996306 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r32p00.json", + "sha256": "996a12a5ed962218dfbe6f30be688119030f10f6539647b6e4a9f25f032b22cb", + "summary": { + "admission_lag_max_ms": 0.11958007235080004, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1198.4891929896548, + "ttft_p50_ms": 682.7383439522237, + "ttft_p95_ms": 1134.0074760373682 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2040.8174369949847, + 2082.3236129945144 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r64p00.json", + "sha256": "c5f17f625071e905b5f1bf8045cd233b2fea0fe6ca00044935dee9c663d1526b", + "summary": { + "admission_lag_max_ms": 0.7995429914444685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2131.160691031255, + "ttft_p50_ms": 1134.8516750149429, + "ttft_p95_ms": 2040.8174369949847 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r64p00.json", + "sha256": "775db5b941347d9fd5d6df32d6ee037f8168bd47b9eae9a7bdb3dbd8c40aa990", + "summary": { + "admission_lag_max_ms": 0.17620902508497238, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2129.518053960055, + "ttft_p50_ms": 1112.3740070033818, + "ttft_p95_ms": 2082.3236129945144 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + }, + "real_campaign_resources": { + "fresh_server_anchors": 192, + "gpu_hours": 12.074444444444445, + "measured_requests": 12288, + "runs": [ + { + "duration_seconds": 908.0, + "gpu_hours": 0.25222222222222224, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive-20260716T164730291172Z", + "tp": 1 + }, + { + "duration_seconds": 540.0, + "gpu_hours": 0.15, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + { + "duration_seconds": 937.0, + "gpu_hours": 0.2602777777777778, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive-20260716T164731376588Z", + "tp": 1 + }, + { + "duration_seconds": 558.0, + "gpu_hours": 0.155, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + { + "duration_seconds": 972.0, + "gpu_hours": 0.27, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive-20260716T164732482666Z", + "tp": 1 + }, + { + "duration_seconds": 572.0, + "gpu_hours": 0.15888888888888889, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + { + "duration_seconds": 886.0, + "gpu_hours": 0.2461111111111111, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive-20260716T164729267199Z", + "tp": 1 + }, + { + "duration_seconds": 525.0, + "gpu_hours": 0.14583333333333334, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + { + "duration_seconds": 983.0, + "gpu_hours": 0.5461111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive-20260716T164734745039Z", + "tp": 2 + }, + { + "duration_seconds": 569.0, + "gpu_hours": 0.3161111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + { + "duration_seconds": 1033.0, + "gpu_hours": 0.5738888888888889, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive-20260716T170909930153Z", + "tp": 2 + }, + { + "duration_seconds": 584.0, + "gpu_hours": 0.3244444444444444, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + { + "duration_seconds": 1063.0, + "gpu_hours": 0.5905555555555555, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive-20260716T170910954319Z", + "tp": 2 + }, + { + "duration_seconds": 597.0, + "gpu_hours": 0.33166666666666667, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + { + "duration_seconds": 957.0, + "gpu_hours": 0.5316666666666666, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive-20260716T164733581881Z", + "tp": 2 + }, + { + "duration_seconds": 563.0, + "gpu_hours": 0.31277777777777777, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + { + "duration_seconds": 1008.0, + "gpu_hours": 1.12, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive-20260716T172922998830Z", + "tp": 4 + }, + { + "duration_seconds": 559.0, + "gpu_hours": 0.6211111111111111, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + { + "duration_seconds": 1024.0, + "gpu_hours": 1.1377777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive-20260716T172923955862Z", + "tp": 4 + }, + { + "duration_seconds": 556.0, + "gpu_hours": 0.6177777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + { + "duration_seconds": 959.0, + "gpu_hours": 1.0655555555555556, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive-20260716T175248149074Z", + "tp": 4 + }, + { + "duration_seconds": 567.0, + "gpu_hours": 0.63, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + { + "duration_seconds": 995.0, + "gpu_hours": 1.1055555555555556, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive-20260716T170911977980Z", + "tp": 4 + }, + { + "duration_seconds": 550.0, + "gpu_hours": 0.6111111111111112, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + ], + "successful_fleet_jobs": 24, + "warmup_requests": 4512 + }, + "schema": "qwen30-prefill-fidelity-comparison-v1", + "simulator": { + "tp1_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + } + }, + "simulator_sources": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json", + "sha256": "2e3c4711bf9a5046daed3517c58f23241880467ee8f8e70191dc379413ad027c" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json", + "sha256": "faffabd900cc38c17a349f13840257078543af9d5da39c363a5d77a87d2eef93" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json", + "sha256": "e54e08cb263c376daf92f418a90ad79ae1afaa2edbed69271513218ab104e93c" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json", + "sha256": "b4530266c764e2bb1b420a68052d70e24f63eb9c30f3b3e8f8c9e4865a25f280" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json", + "sha256": "cd06f00b5c6f514896842e9f4482bd6270fedb0bf003823f1b974d136acdeffd" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json", + "sha256": "6ec105c50563e669ef145aa2b9a431e695a6308abb6d5db951b8a1f9d4a9e2f4" + } + ] +} diff --git a/runs/frontier-phase-factorial-v0/run_frontier_qwen30_prefill_surface.py b/runs/frontier-phase-factorial-v0/run_frontier_qwen30_prefill_surface.py new file mode 100644 index 0000000..e7344fc --- /dev/null +++ b/runs/frontier-phase-factorial-v0/run_frontier_qwen30_prefill_surface.py @@ -0,0 +1,455 @@ +#!/usr/bin/env python3 +"""Freeze the Qwen30 fixed-shape prefill-only Frontier surface.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import importlib.util +import json +import math +import os +import subprocess +import sys +import time +from dataclasses import asdict, dataclass +from pathlib import Path +from typing import Any + + +MODEL = "qwen3-a3b-30b-moe" +RATES = (4.0, 8.0, 16.0, 32.0, 64.0) +TTFT_SLO_MS = 1256.0 +TARGET_PASS_RATE = 0.95 +NUM_BLOCKS = {1: 20080, 2: 76537, 4: 191727} + + +@dataclass(frozen=True) +class Config: + tp: int + mns: int + + @property + def name(self) -> str: + return f"tp{self.tp}_mns{self.mns}" + + +GRID = tuple(Config(tp, mns) for tp in (1, 2, 4) for mns in (8, 16, 32, 64)) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--frontier-source", type=Path, required=True) + parser.add_argument("--replayserve-root", type=Path, required=True) + parser.add_argument("--profile-root", type=Path, required=True) + parser.add_argument("--python-deps", type=Path, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument("--requests", type=int, default=64) + parser.add_argument("--rate", type=float, action="append") + parser.add_argument("--config", action="append") + parser.add_argument( + "--cc-backend", choices=("analytical", "vidur"), default="analytical" + ) + parser.add_argument("--allreduce-csv", type=Path) + parser.add_argument("--timeout-seconds", type=float, default=900.0) + parser.add_argument("--resume", action="store_true") + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def load_module(name: str, path: Path): + spec = importlib.util.spec_from_file_location(name, path) + if spec is None or spec.loader is None: + raise ImportError(path) + module = importlib.util.module_from_spec(spec) + sys.modules[name] = module + spec.loader.exec_module(module) + return module + + +def write_trace(path: Path, *, requests: int, rate: float) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + fields = [ + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", + ] + with path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=fields) + writer.writeheader() + for request_id in range(requests): + writer.writerow( + { + "arrived_at": f"{request_id / rate:.12f}", + "num_prefill_tokens": 2048, + "num_decode_tokens": 1, + "session_id": request_id, + "block_hash_ids": "|".join( + str(request_id * 128 + block + 1) for block in range(128) + ), + } + ) + + +def profile_paths(root: Path) -> dict[str, Path]: + paths = { + "linear": root / "linear_op.csv", + "attention": root / "attention.csv", + "moe": root / "moe.csv", + "manifest": root / "manifest.json", + } + missing = [str(path) for path in paths.values() if not path.is_file()] + if missing: + raise FileNotFoundError(missing) + return paths + + +def validate_profile(paths: dict[str, Path]) -> dict[str, Any]: + manifest = json.loads(paths["manifest"].read_text()) + expected = manifest["outputs"] + for filename in ("linear_op.csv", "attention.csv", "moe.csv"): + actual = sha256(paths[{"linear_op.csv": "linear", "attention.csv": "attention", "moe.csv": "moe"}[filename]]) + if actual != expected[filename]: + raise ValueError(f"profile hash mismatch for {filename}") + with paths["attention"].open(newline="") as source: + rows = list(csv.DictReader(source)) + coverage = {} + for tp in (1, 2, 4): + exact = [ + row + for row in rows + if int(row["num_tensor_parallel_workers"]) == tp + and row["is_prefill"].lower() == "true" + and row.get("is_true_mixed_batch", "").lower() != "true" + and int(float(row["batch_size"])) == 1 + and int(float(row["total_tokens"])) == 2048 + ] + if len(exact) != 1: + raise ValueError(f"expected one exact TP{tp} 2048-token prefill row, got {len(exact)}") + coverage[str(tp)] = {"exact_prefill_2048_rows": len(exact), "profile_batch_size": int(exact[0]["batch_size"])} + return {"manifest": manifest, "attention": coverage} + + +def knobs(config: Config, paths: dict[str, Path], cache: Path) -> dict[str, Any]: + return { + "simulation_mode": "online", + "sys_arch": "co-location", + "num_replicas": 1, + "cluster_scheduler": "sticky_round_robin", + "model_name": MODEL, + "device": "h20", + "network_device": "h20_dgx", + "attn_tensor_parallel_size": config.tp, + "attn_data_parallel_size": 1, + "moe_tensor_parallel_size": config.tp, + "moe_expert_parallel_size": 1, + "num_pipeline_stages": 1, + "replica_scheduler": "vllm_v1", + "decode_cuda_graph_mode": "none", + "batch_size_cap": config.mns, + "max_tokens_in_batch": 8192, + "long_prefill_token_threshold": 0, + "block_size": 16, + "num_blocks_mode": "explicit", + "num_blocks": NUM_BLOCKS[config.tp], + "gpu_memory_utilization": 0.92, + "non_kv_cache_overhead_bytes": 0, + "trace_max_tokens": 40960, + "cache_dir": str(cache), + "enable_dummy_mode": False, + "linear_op_input_file": str(paths["linear"]), + "atten_input_file": str(paths["attention"]), + "moe_input_file": str(paths["moe"]), + "prediction_max_prefill_chunk_size": 18000, + "prediction_max_batch_size": 128, + "prediction_max_tokens_per_request": 32768, + "no_cache": False, + "skip_cpu_overhead_modeling": True, + "enable_prefix_caching": False, + "enable_chunked_prefill": True, + } + + +def configure_cc_command( + command: list[str], *, backend: str, allreduce_csv: Path | None, cache: Path +) -> list[str]: + configured = list(command) + option = "--cc_backend_config_type" + try: + index = configured.index(option) + except ValueError as error: + raise ValueError(f"Frontier command is missing {option}") from error + configured[index + 1] = backend + if backend == "analytical": + if allreduce_csv is not None: + raise ValueError("--allreduce-csv requires --cc-backend vidur") + return configured + if allreduce_csv is None: + raise ValueError("--cc-backend vidur requires --allreduce-csv") + configured.extend( + [ + "--vidur_cc_backend_config_all_reduce_input_file", + str(allreduce_csv), + "--vidur_cc_backend_config_cache_dir", + str(cache), + "--vidur_cc_backend_config_k_fold_cv_splits", + "6", + "--vidur_cc_backend_config_num_training_job_threads", + "1", + ] + ) + return configured + + +def find_metrics(run_dir: Path) -> tuple[Path, Path]: + systems = list((run_dir / "metrics").rglob("system_metrics.json")) + requests = list((run_dir / "metrics").rglob("request_metrics.csv")) + if len(systems) != 1 or len(requests) != 1: + raise RuntimeError(f"metric pair mismatch: {len(systems)}/{len(requests)}") + return systems[0], requests[0] + + +def score(system_path: Path, request_path: Path, expected_requests: int) -> dict[str, Any]: + system = json.loads(system_path.read_text()) + metadata = system["simulation_metadata"] + if int(metadata["completed_requests"]) != expected_requests: + raise ValueError("Frontier completion count mismatch") + with request_path.open(newline="") as source: + rows = list(csv.DictReader(source)) + if len(rows) != expected_requests: + raise ValueError("request metric count mismatch") + values = [] + passed = 0 + for row in rows: + if int(float(row["request_num_prefill_tokens"])) != 2048: + raise ValueError("prefill shape drift") + if int(float(row["request_num_decode_tokens"])) != 1: + raise ValueError("decode shape drift") + ttft = float(row["ttft"]) + if not math.isfinite(ttft) or ttft < 0: + raise ValueError("invalid TTFT") + values.append(ttft) + passed += int(ttft <= TTFT_SLO_MS) + ordered = sorted(values) + pass_rate = passed / expected_requests + return { + "ttft_p50_ms": ordered[math.ceil(0.50 * len(ordered)) - 1], + "ttft_p95_ms": ordered[math.ceil(0.95 * len(ordered)) - 1], + "ttft_max_ms": max(ordered), + "passed": passed, + "pass_rate": pass_rate, + "feasible": pass_rate >= TARGET_PASS_RATE, + "throughput_requests_per_second": float(system["throughput_metrics"]["requests_per_second"]), + } + + +def main() -> None: + args = parse_args() + args.frontier_source = args.frontier_source.resolve() + args.replayserve_root = args.replayserve_root.resolve() + args.profile_root = args.profile_root.resolve() + args.python_deps = args.python_deps.resolve() + args.output_root = args.output_root.resolve() + if args.allreduce_csv is not None: + args.allreduce_csv = args.allreduce_csv.resolve() + if not args.allreduce_csv.is_file(): + raise FileNotFoundError(args.allreduce_csv) + rates = tuple(args.rate or RATES) + selected = list(GRID) + if args.config: + wanted = set(args.config) + selected = [config for config in GRID if config.name in wanted] + if {config.name for config in selected} != wanted: + raise ValueError(f"unknown configs: {wanted - {config.name for config in selected}}") + paths = profile_paths(args.profile_root) + coverage = validate_profile(paths) + builder = load_module( + "qwen30_prefill_frontier_builder", + args.replayserve_root / "tools/run_frontier_sweep.py", + ) + frontier_head = subprocess.run( + ["git", "-C", str(args.frontier_source), "rev-parse", "HEAD"], + check=True, + text=True, + stdout=subprocess.PIPE, + ).stdout.strip() + traces = {} + for rate in rates: + trace = args.output_root / "traces" / f"r{rate:g}.csv" + write_trace(trace, requests=args.requests, rate=rate) + traces[rate] = trace + + config_results = [] + for config in selected: + loads = [] + config_knobs = knobs(config, paths, args.output_root / "cache") + for rate in rates: + run_dir = args.output_root / "runs" / config.name / f"r{rate:g}" + result_path = run_dir / "result.json" + if args.resume and result_path.is_file(): + loads.append(json.loads(result_path.read_text())) + continue + run_dir.mkdir(parents=True, exist_ok=True) + command = builder.build_frontier_command( + python_bin="/usr/bin/python3", + trace_file=traces[rate], + metrics_root=run_dir / "metrics", + run_id=f"qwen30_prefill_{config.name}_r{rate:g}", + knobs=config_knobs, + ) + command = configure_cc_command( + command, + backend=args.cc_backend, + allreduce_csv=args.allreduce_csv, + cache=args.output_root / "cc-cache", + ) + write_json(run_dir / "command.json", command) + environment = os.environ.copy() + pythonpath = [str(args.python_deps), str(args.frontier_source)] + if environment.get("PYTHONPATH"): + pythonpath.append(environment["PYTHONPATH"]) + environment.update( + { + "PYTHONPATH": ":".join(pythonpath), + "CUDA_VISIBLE_DEVICES": "", + "NVIDIA_VISIBLE_DEVICES": "void", + "WANDB_DISABLED": "true", + "VIDUR_DISABLE_WANDB": "1", + "FRONTIER_LOG_LEVEL": "WARNING", + "PYTHONDONTWRITEBYTECODE": "1", + } + ) + started = time.time() + with (run_dir / "stdout.log").open("w") as stdout, ( + run_dir / "stderr.log" + ).open("w") as stderr: + completed = subprocess.run( + command, + cwd=args.frontier_source, + env=environment, + stdout=stdout, + stderr=stderr, + timeout=args.timeout_seconds, + check=False, + ) + if completed.returncode != 0: + raise RuntimeError( + f"Frontier failed for {config.name} rate={rate}: {completed.returncode}" + ) + system_path, request_path = find_metrics(run_dir) + result = { + "status": "completed", + "config": asdict(config) | {"name": config.name}, + "offered_request_rate": rate, + "offered_request_rate_per_gpu": rate / config.tp, + "request_count": args.requests, + "elapsed_seconds": time.time() - started, + "trace_sha256": sha256(traces[rate]), + "request_metrics_sha256": sha256(request_path), + "score": score(system_path, request_path, args.requests), + } + write_json(result_path, result) + loads.append(result) + print( + json.dumps( + { + "config": config.name, + "rate": rate, + "pass_rate": result["score"]["pass_rate"], + "feasible": result["score"]["feasible"], + }, + sort_keys=True, + ), + flush=True, + ) + config_results.append({"config": asdict(config) | {"name": config.name}, "loads": loads}) + + capacities = [] + for item in config_results: + feasible = [ + load["offered_request_rate"] + for load in item["loads"] + if load["score"]["feasible"] + ] + capacity = max(feasible) if feasible else None + capacities.append( + { + "config": item["config"], + "maximum_tested_feasible_request_rate": capacity, + "maximum_tested_feasible_request_rate_per_gpu": ( + capacity / item["config"]["tp"] if capacity is not None else None + ), + "lower_censored": capacity is None, + "upper_censored": capacity == max(rates) if capacity is not None else False, + } + ) + capacities.sort( + key=lambda row: ( + -(row["maximum_tested_feasible_request_rate_per_gpu"] or -1), + row["config"]["name"], + ) + ) + full = selected == list(GRID) and rates == RATES and args.requests == 64 + manifest = { + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "frozen_before_real" if full else "partial_not_decision_bearing", + "contract": { + "rates": rates, + "requests_per_anchor": args.requests, + "input_tokens": 2048, + "output_tokens": 1, + "ttft_slo_ms": TTFT_SLO_MS, + "target_pass_rate": TARGET_PASS_RATE, + "prefix_caching": False, + "arrival": "open_loop_uniform", + }, + "frontier": { + "source": str(args.frontier_source), + "git_head": frontier_head, + "git_status_short": subprocess.run( + ["git", "-C", str(args.frontier_source), "status", "--short"], + check=True, + text=True, + stdout=subprocess.PIPE, + ).stdout, + }, + "profiles": { + "root": str(args.profile_root), + "coverage": coverage, + "sha256": {name: sha256(path) for name, path in paths.items()}, + }, + "collective": { + "backend": args.cc_backend, + "allreduce_csv": ( + str(args.allreduce_csv) if args.allreduce_csv is not None else None + ), + "allreduce_csv_sha256": ( + sha256(args.allreduce_csv) if args.allreduce_csv is not None else None + ), + }, + "config_results": config_results, + "capacity": capacities, + } + write_json(args.output_root / "frontier_surface_frozen.json", manifest) + print(args.output_root / "frontier_surface_frozen.json") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-phase-factorial-v0/run_qwen30_prefill_real_config.sh b/runs/frontier-phase-factorial-v0/run_qwen30_prefill_real_config.sh new file mode 100644 index 0000000..2861c21 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/run_qwen30_prefill_real_config.sh @@ -0,0 +1,144 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}" +TP="${TP:?TP is required}" +MNS="${MNS:?MNS is required}" +RATES="${RATES:-4 8 16 32 64}" +INPUT_TOKENS="${INPUT_TOKENS:-2048}" +OUTPUT_TOKENS="${OUTPUT_TOKENS:-1}" +TPOT_SLO_MS="${TPOT_SLO_MS:-150}" +WARMUP_SECONDS="${WARMUP_SECONDS:-2}" +SERVER_PORT="${SERVER_PORT:?SERVER_PORT is required}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +SERVED_MODEL="qwen3-30b-prefill-only" +SERVER_PID="" + +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" +exec > >(tee -a "${OUTPUT_ROOT}/logs/controller.log") 2>&1 + +cleanup() { + if [[ -n "${SERVER_PID}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then + kill -TERM -- "-${SERVER_PID}" 2>/dev/null || true + for _ in $(seq 1 30); do + kill -0 "${SERVER_PID}" 2>/dev/null || break + sleep 1 + done + kill -KILL -- "-${SERVER_PID}" 2>/dev/null || true + fi + SERVER_PID="" +} +trap cleanup EXIT INT TERM + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:-}" +if [[ "${#GPU_IDS[@]}" -ne "${TP}" ]]; then + echo "ERROR: expected ${TP} allocated GPUs, got ${CUDA_VISIBLE_DEVICES:-unset}" >&2 + exit 1 +fi +read -r -a RATE_ARRAY <<< "${RATES}" + +echo "QWEN30_FIXED_REAL_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 config=TP${TP}_MNS${MNS}_MBT8192 rates=${RATES// /,} rounds=2 requests=64 shape=ISL${INPUT_TOKENS}_OSL${OUTPUT_TOKENS} ttft_slo=1000+1000*ISL/8000ms tpot_slo=${TPOT_SLO_MS}ms warmup_seconds=${WARMUP_SECONDS} arrivals=uniform prefix=off cuda_graph=runtime_default isolation=fresh_server_per_anchor output=${OUTPUT_ROOT}" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +sha256sum run_qwen30_prefill_real_config.sh qwen30_prefill_client.py \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +"${VENV_ROOT}/bin/python" - "${TP}" "${MNS}" "${RATES}" "${INPUT_TOKENS}" "${OUTPUT_TOKENS}" "${TPOT_SLO_MS}" "${WARMUP_SECONDS}" \ + > "${OUTPUT_ROOT}/provenance/contract.json" <<'PY' +import importlib.metadata as metadata +import json +import platform +import sys + +tp, mns, rates, input_tokens, output_tokens, tpot_slo_ms, warmup_seconds = sys.argv[1:] +input_tokens = int(input_tokens) +print(json.dumps({ + "python": platform.python_version(), + "torch": metadata.version("torch"), + "transformers": metadata.version("transformers"), + "vllm": metadata.version("vllm"), + "config": {"tp": int(tp), "mns": int(mns), "mbt": 8192}, + "rates": [float(value) for value in rates.split()], + "rounds": 2, + "requests_per_anchor": 64, + "anchor_isolation": "fresh_server_per_rate_per_round", + "target_rate_warmup_requests": f"min(32, max(4, ceil(rate * {warmup_seconds})))", + "input_tokens": input_tokens, + "output_tokens": int(output_tokens), + "ttft_slo_ms": 1000.0 + 1000.0 * input_tokens / 8000.0, + "tpot_slo_ms": float(tpot_slo_ms) if int(output_tokens) > 1 else None, + "target_pass_rate": 0.95, +}, indent=2, sort_keys=True)) +PY +nvidia-smi --query-gpu=index,name,uuid,driver_version --format=csv,noheader \ + > "${OUTPUT_ROOT}/provenance/gpus.csv" +sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model.sha256" + +export TOKENIZERS_PARALLELISM=false +export VLLM_USE_V1=1 +export TORCH_CUDA_ARCH_LIST=9.0 +export HF_HUB_OFFLINE=1 +export TRANSFORMERS_OFFLINE=1 + +for ROUND in 1 2; do + ROUND_ROOT="${OUTPUT_ROOT}/round${ROUND}" + mkdir -p "${ROUND_ROOT}/logs" "${ROUND_ROOT}/results" + ORDERED_RATES=("${RATE_ARRAY[@]}") + if [[ "${ROUND}" -eq 2 ]]; then + ORDERED_RATES=() + for ((index=${#RATE_ARRAY[@]}-1; index>=0; index--)); do + ORDERED_RATES+=("${RATE_ARRAY[index]}") + done + fi + for RATE in "${ORDERED_RATES[@]}"; do + KEY="$(printf 'r%.2f' "${RATE}" | tr '.' 'p')" + SERVER_LOG="${ROUND_ROOT}/logs/server_${KEY}.log" + setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \ + --host 127.0.0.1 --port "${SERVER_PORT}" --served-model-name "${SERVED_MODEL}" \ + --tensor-parallel-size "${TP}" --gpu-memory-utilization 0.92 \ + --max-model-len 40960 --max-num-batched-tokens 8192 --max-num-seqs "${MNS}" \ + --no-enable-prefix-caching --enable-chunked-prefill --no-enable-log-requests \ + > "${SERVER_LOG}" 2>&1 & + SERVER_PID=$! + READY=0 + for _ in $(seq 1 120); do + if curl -fsS --max-time 2 "http://127.0.0.1:${SERVER_PORT}/v1/models" \ + > "${ROUND_ROOT}/results/models_${KEY}.json" 2>/dev/null; then + READY=1 + break + fi + if ! kill -0 "${SERVER_PID}" 2>/dev/null; then + tail -200 "${SERVER_LOG}" + exit 1 + fi + sleep 3 + done + if [[ "${READY}" -ne 1 ]]; then + tail -200 "${SERVER_LOG}" + exit 1 + fi + + WARMUP_REQUESTS="$("${VENV_ROOT}/bin/python" - "${RATE}" "${WARMUP_SECONDS}" <<'PY' +import math +import sys +print(min(32, max(4, math.ceil(float(sys.argv[1]) * float(sys.argv[2]))))) +PY +)" + "${VENV_ROOT}/bin/python" qwen30_prefill_client.py --port "${SERVER_PORT}" \ + --served-model "${SERVED_MODEL}" --model-path "${MODEL_ROOT}" --rate "${RATE}" \ + --requests "${WARMUP_REQUESTS}" --input-tokens "${INPUT_TOKENS}" \ + --output-tokens "${OUTPUT_TOKENS}" --tpot-slo-ms "${TPOT_SLO_MS}" \ + --output "${ROUND_ROOT}/results/warmup_${KEY}.json" + "${VENV_ROOT}/bin/python" qwen30_prefill_client.py --port "${SERVER_PORT}" \ + --served-model "${SERVED_MODEL}" --model-path "${MODEL_ROOT}" --rate "${RATE}" \ + --requests 64 --input-tokens "${INPUT_TOKENS}" \ + --output-tokens "${OUTPUT_TOKENS}" --tpot-slo-ms "${TPOT_SLO_MS}" \ + --output "${ROUND_ROOT}/results/${KEY}.json" + cleanup + done +done + +find "${OUTPUT_ROOT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 \ + | sort -z | xargs -0 sha256sum > "${OUTPUT_ROOT}/provenance/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "QWEN30_PREFILL_REAL_CONFIG_COMPLETE" diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json new file mode 100644 index 0000000..48c05b5 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.505136728286743, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.398303508758545, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.264978170394897, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.103909015655518, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.05237078666687, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.048887729644775, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.058743476867676, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.106712818145752, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.1020348072052, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.897645235061646, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.957297325134277, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.799486875534058, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 5.0, + 6.0, + 7.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json new file mode 100644 index 0000000..da9191c --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.598209381103516, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.660333633422852, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.409451007843018, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.360321521759033, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.402549028396606, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.40966272354126, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.155991315841675, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.103453874588013, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.255317687988281, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 10.106109380722046, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 10.112579107284546, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.847446918487549, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 10.0, + 12.0, + 14.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json new file mode 100644 index 0000000..0190387 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 10.649231672286987, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 10.245610237121582, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.953459978103638, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 10.059623956680298, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 10.000702381134033, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.852763652801514, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.849465370178223, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.80930495262146, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.798201560974121, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.800058603286743, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.804483413696289, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.80262541770935, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 20.0, + 24.0, + 28.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json new file mode 100644 index 0000000..0dafcdd --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json @@ -0,0 +1,647 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 8.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 20.451046466827393, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 22.0544171333313, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 18.859471321105957, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 16.640390872955322, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 18.142091274261475, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 19.96323037147522, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 18.554351568222046, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 25.70284128189087, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 20.056548833847046, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 11.15293574333191, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.411258935928345, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.355234384536743, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.098335266113281, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.8981294631958, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 9.754717826843262, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.84917163848877, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "f0cfb072afcc2fc82a92c59ddb4e8cfc1e3201e15848f37785eeb9fdd975e779", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.0197002243768605, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 10.197595357894897, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "cfebfbd8aa5d9603158d8ff4e786119433b4c0d705f733ba242da46a6a6fdfb0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 7.742175153637853, + "ttft_max_ms": 570.3133207139617, + "ttft_p50_ms": 391.4107605377559, + "ttft_p95_ms": 516.4107605377559 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 10.00463604927063, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "03578ba406b3299e70ba7975e6b9872d381a5f4bd4fbc3c8f6a8ba49e9b53f2e", + "score": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "throughput_requests_per_second": 8.916224898348922, + "ttft_max_ms": 3318.8404181099845, + "ttft_p50_ms": 1814.2305264490028, + "ttft_p95_ms": 3193.8404181099845 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 10.153925895690918, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "0b761b5f42e65b694eb34735433d43b473ed67415fb4e17b89894fc840b645fc", + "score": { + "feasible": false, + "pass_rate": 0.203125, + "passed": 13, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 5147.666503402199, + "ttft_p50_ms": 2739.3537469047046, + "ttft_p95_ms": 4962.713638565687 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 10.055675268173218, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 64.0, + "request_count": 64, + "request_metrics_sha256": "078209f2f56f21243384bdc70add105bf4942ee44ec8cc9573cad399a4a10e03", + "score": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "throughput_requests_per_second": 9.07297385348011, + "ttft_max_ms": 6100.791503402199, + "ttft_p50_ms": 3208.1037469047046, + "ttft_p95_ms": 5853.338638565687 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 4.0, + 8.0, + 16.0, + 32.0, + 64.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json new file mode 100644 index 0000000..34e2063 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json @@ -0,0 +1,647 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 8.0, + "upper_censored": false + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 264.1678559780121, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e5f6527532682ee5aca3f423367d14eba063f30e63f46d2ba0ed602db4f6eb17", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032196934357252, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.15904188156128, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "8dd3bedcbf76d30bf4285e10b886f95bef9ae8e5d41f4bb50a7650b322b8eb83", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00276032449276, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.447325944900513, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "38c1b13c1f5fbdabd7f79b82ccb1fb37fce6840bcf03614c988d7e4882814506", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.174042777348715, + "ttft_max_ms": 1048.2973400621365, + "ttft_p50_ms": 643.9195307403329, + "ttft_p95_ms": 962.8278877316862 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.053350687026978, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "cb3b7b5a2fcf63cd01304b3b530bb21c0a9bb1c8f5e06d489e3863afd0b4d2f3", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 2851.9297930156335, + "ttft_p50_ms": 1510.0519836938302, + "ttft_p95_ms": 2758.1797930156335 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 9.744416952133179, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a4474c98a68171ada0fb007c36fcf69e9e2dc13f06c74889591e23f93418bd5a", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 3773.8047930156335, + "ttft_p50_ms": 1995.3684369244447, + "ttft_p95_ms": 3622.636520081023 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.402071952819824, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e5f6527532682ee5aca3f423367d14eba063f30e63f46d2ba0ed602db4f6eb17", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032196934357252, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.398229360580444, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "8dd3bedcbf76d30bf4285e10b886f95bef9ae8e5d41f4bb50a7650b322b8eb83", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00276032449276, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.70562195777893, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "38c1b13c1f5fbdabd7f79b82ccb1fb37fce6840bcf03614c988d7e4882814506", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.174042777348715, + "ttft_max_ms": 1048.2973400621365, + "ttft_p50_ms": 643.9195307403329, + "ttft_p95_ms": 962.8278877316862 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.598989009857178, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "cb3b7b5a2fcf63cd01304b3b530bb21c0a9bb1c8f5e06d489e3863afd0b4d2f3", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 2851.9297930156335, + "ttft_p50_ms": 1510.0519836938302, + "ttft_p95_ms": 2758.1797930156335 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 9.697861671447754, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a4474c98a68171ada0fb007c36fcf69e9e2dc13f06c74889591e23f93418bd5a", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 3773.8047930156335, + "ttft_p50_ms": 1995.3684369244447, + "ttft_p95_ms": 3622.636520081023 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.397430419921875, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e5f6527532682ee5aca3f423367d14eba063f30e63f46d2ba0ed602db4f6eb17", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032196934357252, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.3956880569458, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "8dd3bedcbf76d30bf4285e10b886f95bef9ae8e5d41f4bb50a7650b322b8eb83", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00276032449276, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.647802591323853, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "38c1b13c1f5fbdabd7f79b82ccb1fb37fce6840bcf03614c988d7e4882814506", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.174042777348715, + "ttft_max_ms": 1048.2973400621365, + "ttft_p50_ms": 643.9195307403329, + "ttft_p95_ms": 962.8278877316862 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.694743156433105, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "cb3b7b5a2fcf63cd01304b3b530bb21c0a9bb1c8f5e06d489e3863afd0b4d2f3", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 2851.9297930156335, + "ttft_p50_ms": 1510.0519836938302, + "ttft_p95_ms": 2758.1797930156335 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 9.594672203063965, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a4474c98a68171ada0fb007c36fcf69e9e2dc13f06c74889591e23f93418bd5a", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 3773.8047930156335, + "ttft_p50_ms": 1995.3684369244447, + "ttft_p95_ms": 3622.636520081023 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.398636817932129, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "e5f6527532682ee5aca3f423367d14eba063f30e63f46d2ba0ed602db4f6eb17", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.032196934357252, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.495105981826782, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "8dd3bedcbf76d30bf4285e10b886f95bef9ae8e5d41f4bb50a7650b322b8eb83", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.00276032449276, + "ttft_max_ms": 122.24062760264154, + "ttft_p50_ms": 122.24062760264154, + "ttft_p95_ms": 122.24062760264154 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.54387378692627, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "38c1b13c1f5fbdabd7f79b82ccb1fb37fce6840bcf03614c988d7e4882814506", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.174042777348715, + "ttft_max_ms": 1048.2973400621365, + "ttft_p50_ms": 643.9195307403329, + "ttft_p95_ms": 962.8278877316862 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.192227840423584, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "cb3b7b5a2fcf63cd01304b3b530bb21c0a9bb1c8f5e06d489e3863afd0b4d2f3", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 2851.9297930156335, + "ttft_p50_ms": 1510.0519836938302, + "ttft_p95_ms": 2758.1797930156335 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.294084072113037, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 32.0, + "request_count": 64, + "request_metrics_sha256": "a4474c98a68171ada0fb007c36fcf69e9e2dc13f06c74889591e23f93418bd5a", + "score": { + "feasible": false, + "pass_rate": 0.265625, + "passed": 17, + "throughput_requests_per_second": 13.539443740959392, + "ttft_max_ms": 3773.8047930156335, + "ttft_p50_ms": 1995.3684369244447, + "ttft_p95_ms": 3622.636520081023 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 4.0, + 8.0, + 16.0, + 32.0, + 64.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json new file mode 100644 index 0000000..82a1227 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json @@ -0,0 +1,647 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 4.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 4.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 4.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 16.0, + "maximum_tested_feasible_request_rate_per_gpu": 4.0, + "upper_censored": false + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 244.95408463478088, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "c0c3c34bea43dae2477b10152b07632e5d03968f47cdfd70ad5b676e2502af4c", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.039607508408214, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685405, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.85124945640564, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "b5c6ba0cd3b2a81a12fb93b2cad6f80341f9352c110385046490f59d7933a5d5", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.032004183533672, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685495, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.70260500907898, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "c40f10c895f2fc45671f7ff7c6c08f30c20854f0ff4fe3e6719934a7b08b0112", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.412041282697286, + "ttft_max_ms": 314.4210260695206, + "ttft_p50_ms": 239.37774488651664, + "ttft_p95_ms": 309.403713596319 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.855336904525757, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "9f77ee40d102bdf14d213132ef10b4a7f4175908ffee411ef30ad3a5f1e4b16f", + "score": { + "feasible": false, + "pass_rate": 0.625, + "passed": 40, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1850.42236218991, + "ttft_p50_ms": 1017.6783445024347, + "ttft_p95_ms": 1787.92236218991 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.947782754898071, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "efb43cc32f2f7129082e47d89239f08b10c6e7e6dd4a4f0b299d2364cb781af0", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 17.184897883039092, + "ttft_max_ms": 2771.0751922610093, + "ttft_p50_ms": 1468.937931419684, + "ttft_p95_ms": 2661.056949107159 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.648890733718872, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "c0c3c34bea43dae2477b10152b07632e5d03968f47cdfd70ad5b676e2502af4c", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.039607508408214, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685405, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.353798151016235, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "b5c6ba0cd3b2a81a12fb93b2cad6f80341f9352c110385046490f59d7933a5d5", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.032004183533672, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685495, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.39468502998352, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "c40f10c895f2fc45671f7ff7c6c08f30c20854f0ff4fe3e6719934a7b08b0112", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.412041282697286, + "ttft_max_ms": 314.4210260695206, + "ttft_p50_ms": 239.37774488651664, + "ttft_p95_ms": 309.403713596319 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.29756498336792, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "9f77ee40d102bdf14d213132ef10b4a7f4175908ffee411ef30ad3a5f1e4b16f", + "score": { + "feasible": false, + "pass_rate": 0.625, + "passed": 40, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1850.42236218991, + "ttft_p50_ms": 1017.6783445024347, + "ttft_p95_ms": 1787.92236218991 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.34343934059143, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "efb43cc32f2f7129082e47d89239f08b10c6e7e6dd4a4f0b299d2364cb781af0", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 17.184897883039092, + "ttft_max_ms": 2771.0751922610093, + "ttft_p50_ms": 1468.937931419684, + "ttft_p95_ms": 2661.056949107159 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.296925783157349, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "c0c3c34bea43dae2477b10152b07632e5d03968f47cdfd70ad5b676e2502af4c", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.039607508408214, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685405, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.345365285873413, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "b5c6ba0cd3b2a81a12fb93b2cad6f80341f9352c110385046490f59d7933a5d5", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.032004183533672, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685495, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.350512504577637, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "c40f10c895f2fc45671f7ff7c6c08f30c20854f0ff4fe3e6719934a7b08b0112", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.412041282697286, + "ttft_max_ms": 314.4210260695206, + "ttft_p50_ms": 239.37774488651664, + "ttft_p95_ms": 309.403713596319 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.300485372543335, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "9f77ee40d102bdf14d213132ef10b4a7f4175908ffee411ef30ad3a5f1e4b16f", + "score": { + "feasible": false, + "pass_rate": 0.625, + "passed": 40, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1850.42236218991, + "ttft_p50_ms": 1017.6783445024347, + "ttft_p95_ms": 1787.92236218991 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.292456150054932, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "efb43cc32f2f7129082e47d89239f08b10c6e7e6dd4a4f0b299d2364cb781af0", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 17.184897883039092, + "ttft_max_ms": 2771.0751922610093, + "ttft_p50_ms": 1468.937931419684, + "ttft_p95_ms": 2661.056949107159 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.294773817062378, + "offered_request_rate": 4.0, + "offered_request_rate_per_gpu": 1.0, + "request_count": 64, + "request_metrics_sha256": "c0c3c34bea43dae2477b10152b07632e5d03968f47cdfd70ad5b676e2502af4c", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 4.039607508408214, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685405, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "0cef631f351f5afd880172d0467d62d9813de5ba36fd3795f91afe3256450c05" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.34990382194519, + "offered_request_rate": 8.0, + "offered_request_rate_per_gpu": 2.0, + "request_count": 64, + "request_metrics_sha256": "b5c6ba0cd3b2a81a12fb93b2cad6f80341f9352c110385046490f59d7933a5d5", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 8.032004183533672, + "ttft_max_ms": 93.12333977685495, + "ttft_p50_ms": 93.12333977685495, + "ttft_p95_ms": 93.12333977685495 + }, + "status": "completed", + "trace_sha256": "c49314cbe4e4b07a9335b57e861ec5b6541ea4cc061ca9abf923453276c419fd" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.346213579177856, + "offered_request_rate": 16.0, + "offered_request_rate_per_gpu": 4.0, + "request_count": 64, + "request_metrics_sha256": "c40f10c895f2fc45671f7ff7c6c08f30c20854f0ff4fe3e6719934a7b08b0112", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 15.412041282697286, + "ttft_max_ms": 314.4210260695206, + "ttft_p50_ms": 239.37774488651664, + "ttft_p95_ms": 309.403713596319 + }, + "status": "completed", + "trace_sha256": "2504b6340f764e726110e48251f8017a2f9ba67b44b8af7d3a838bf674823d33" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.498731136322021, + "offered_request_rate": 32.0, + "offered_request_rate_per_gpu": 8.0, + "request_count": 64, + "request_metrics_sha256": "9f77ee40d102bdf14d213132ef10b4a7f4175908ffee411ef30ad3a5f1e4b16f", + "score": { + "feasible": false, + "pass_rate": 0.625, + "passed": 40, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1850.42236218991, + "ttft_p50_ms": 1017.6783445024347, + "ttft_p95_ms": 1787.92236218991 + }, + "status": "completed", + "trace_sha256": "5508a3ce590f7d1b0b3c211a2d4f2131886a7ae2d85cafa0c436a8e3029675f1" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.300755977630615, + "offered_request_rate": 64.0, + "offered_request_rate_per_gpu": 16.0, + "request_count": 64, + "request_metrics_sha256": "efb43cc32f2f7129082e47d89239f08b10c6e7e6dd4a4f0b299d2364cb781af0", + "score": { + "feasible": false, + "pass_rate": 0.390625, + "passed": 25, + "throughput_requests_per_second": 17.184897883039092, + "ttft_max_ms": 2771.0751922610093, + "ttft_p50_ms": 1468.937931419684, + "ttft_p95_ms": 2661.056949107159 + }, + "status": "completed", + "trace_sha256": "8987c3ebd4a7db983dc0e99e44710cd44a9d5d21cdb8e6b58878b160918593d9" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 4.0, + 8.0, + 16.0, + 32.0, + 64.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/test_phase_factorial.py b/runs/frontier-phase-factorial-v0/test_phase_factorial.py new file mode 100644 index 0000000..07293da --- /dev/null +++ b/runs/frontier-phase-factorial-v0/test_phase_factorial.py @@ -0,0 +1,64 @@ +#!/usr/bin/env python3 + +from __future__ import annotations + +import importlib.util +import sys +from pathlib import Path + + +ROOT = Path(__file__).parent + + +def load(name: str): + path = ROOT / name + spec = importlib.util.spec_from_file_location(path.stem, path) + assert spec and spec.loader + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def test_percentile() -> None: + client = load("qwen30_prefill_client.py") + assert client.percentile([4.0, 1.0, 3.0, 2.0], 0.50) == 2.0 + assert client.percentile([4.0, 1.0, 3.0, 2.0], 0.95) == 4.0 + assert client.percentile([], 0.95) is None + assert client.ttft_slo_ms(512) == 1064.0 + assert client.ttft_slo_ms(2048) == 1256.0 + + +def test_grid_and_trace(tmp_path: Path) -> None: + surface = load("run_frontier_qwen30_prefill_surface.py") + assert len(surface.GRID) == 12 + assert {config.tp for config in surface.GRID} == {1, 2, 4} + trace = tmp_path / "trace.csv" + surface.write_trace(trace, requests=3, rate=4.0) + lines = trace.read_text().splitlines() + assert len(lines) == 4 + assert lines[1].split(",")[:3] == ["0.000000000000", "2048", "1"] + assert lines[3].split(",")[:3] == ["0.500000000000", "2048", "1"] + + +def test_kendall_tau_b() -> None: + analysis = load("analyze_qwen30_prefill_fidelity.py") + assert analysis.kendall_tau_b([1, 2, 3], [1, 2, 3])["kendall_tau_b"] == 1 + assert analysis.kendall_tau_b([1, 2, 3], [3, 2, 1])["kendall_tau_b"] == -1 + + +def test_configure_cc_command(tmp_path: Path) -> None: + surface = load("run_frontier_qwen30_prefill_surface.py") + base = ["python", "--cc_backend_config_type", "analytical", "--other", "x"] + analytical = surface.configure_cc_command( + base, backend="analytical", allreduce_csv=None, cache=tmp_path + ) + assert analytical == base + profile = tmp_path / "all_reduce.csv" + profile.write_text("header\n") + vidur = surface.configure_cc_command( + base, backend="vidur", allreduce_csv=profile, cache=tmp_path / "cache" + ) + assert vidur[2] == "vidur" + assert "--vidur_cc_backend_config_all_reduce_input_file" in vidur + assert str(profile) in vidur diff --git a/runs/frontier-qwen30-vllm020-profile-v1/analyze_op_trace_delta.py b/runs/frontier-qwen30-vllm020-profile-v1/analyze_op_trace_delta.py new file mode 100644 index 0000000..07883e4 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/analyze_op_trace_delta.py @@ -0,0 +1,96 @@ +#!/usr/bin/env python3 +"""Compare Frontier operator predictions at the identical initial scheduler state.""" + +from __future__ import annotations + +import argparse +import json +import re +from pathlib import Path + + +PATTERN = re.compile( + r"\[OP-TRACE\]\[MONOLITHIC\]\[(?:ATTENTION|MOE)\]\[([^]]+)\] " + r"batch_id=(\d+), layer_id=(\d+), predicted_time_ms=([0-9.eE+-]+)" +) +COMPONENTS = ( + "input_layernorm", + "attn_pre_proj", + "attn_rope", + "attn_kv_cache_save", + "attn_prefill", + "attn_decode", + "attn_post_proj", + "post_attention_layernorm", + "moe_gating_linear", + "moe_gating_routing_topk", + "moe_shuffling", + "moe_grouped_gemm", +) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--old", type=Path, required=True) + parser.add_argument("--new", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def first_state(path: Path) -> dict[str, float]: + values: dict[str, set[float]] = {} + with path.open(errors="replace") as handle: + for line in handle: + match = PATTERN.search(line) + if match and int(match[2]) == 0 and int(match[3]) == 0: + values.setdefault(match[1], set()).add(float(match[4])) + ambiguous = {name: sorted(items) for name, items in values.items() if len(items) != 1} + if ambiguous: + raise ValueError(f"ambiguous initial predictions in {path}: {ambiguous}") + result = {name: next(iter(items)) for name, items in values.items()} + result.setdefault("attn_decode", 0.0) + missing = sorted(set(COMPONENTS) - set(result)) + if missing: + raise ValueError(f"missing initial predictions in {path}: {missing}") + return result + + +def main() -> None: + args = parse_args() + old = first_state(args.old) + new = first_state(args.new) + rows = [] + for component in COMPONENTS: + rows.append( + { + "component": component, + "historical_profile_ms": old[component], + "vllm020_profile_ms": new[component], + "new_over_old": new[component] / old[component] + if old[component] != 0 + else None, + } + ) + old_total = sum(old[name] for name in COMPONENTS) + new_total = sum(new[name] for name in COMPONENTS) + output = { + "schema": "frontier-initial-op-trace-delta.v1", + "comparison_contract": { + "fixture": "fidelity_p1_tp1_mns64_low1", + "scheduler_state": "batch_id=0, layer_id=0 before profile-dependent trajectories diverge", + "calibration_a_tp": 1.0, + }, + "layer_component_sum_ms": { + "historical_profile": old_total, + "vllm020_profile": new_total, + "new_over_old": new_total / old_total, + }, + "rows": rows, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(output, indent=2, sort_keys=True) + "\n") + print(args.output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/analyze_p1_profile_ablation.py b/runs/frontier-qwen30-vllm020-profile-v1/analyze_p1_profile_ablation.py new file mode 100644 index 0000000..67ec218 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/analyze_p1_profile_ablation.py @@ -0,0 +1,142 @@ +#!/usr/bin/env python3 +"""Analyze real, calibrated, old-profile, and new-profile P1 probe outcomes.""" + +from __future__ import annotations + +import argparse +import csv +import json +import statistics +from pathlib import Path +from typing import Any + + +MODES = ("historical-calibrated", "historical-profile-only", "vllm020-profile-only") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--controller-state", type=Path, required=True) + parser.add_argument("--calibrated", type=Path, required=True) + parser.add_argument("--old-profile-only", type=Path, required=True) + parser.add_argument("--new-profile-only", type=Path, required=True) + parser.add_argument("--output-json", type=Path, required=True) + parser.add_argument("--output-csv", type=Path, required=True) + return parser.parse_args() + + +def load_results(path: Path) -> dict[tuple[str, str], dict[str, Any]]: + payload = json.loads(path.read_text()) + if payload["status"] != "PASS" or len(payload["results"]) != 12: + raise ValueError(f"expected 12 passing simulator probes in {path}") + return {(row["cell"], row["role"]): row for row in payload["results"]} + + +def main() -> None: + args = parse_args() + controller = json.loads(args.controller_state.read_text()) + real: dict[tuple[str, str], dict[str, Any]] = {} + for cell, value in controller["cells"].items(): + tp = int(value["tp"]) + for run in value["runs"]: + if run["role"] in ("low1", "high1"): + real[(cell, run["role"])] = { + **run, + "tp": tp, + "offered_req_s_per_gpu": int(run["selected_count"]) / 60 / tp, + } + if len(real) != 12: + raise ValueError(f"expected 12 real P1 probes, found {len(real)}") + + modes = { + "historical-calibrated": load_results(args.calibrated), + "historical-profile-only": load_results(args.old_profile_only), + "vllm020-profile-only": load_results(args.new_profile_only), + } + rows: list[dict[str, Any]] = [] + summaries: dict[str, Any] = {} + for mode in MODES: + predicted = modes[mode] + false_feasible = 0 + false_infeasible = 0 + pass_errors: list[float] = [] + capacity_lower_bounds: dict[str, float] = {} + for key in sorted(real): + real_row = real[key] + sim_row = predicted[key] + scorer = sim_row["scorer"] + sim_feasible = bool(scorer["slo"]["feasible"]) + real_feasible = bool(real_row["feasible"]) + false_feasible += int(sim_feasible and not real_feasible) + false_infeasible += int(real_feasible and not sim_feasible) + pass_error = abs(float(scorer["slo"]["pass_rate"]) - float(real_row["pass_rate"])) + pass_errors.append(pass_error) + rows.append( + { + "mode": mode, + "cell": key[0], + "role": key[1], + "real_feasible": real_feasible, + "sim_feasible": sim_feasible, + "real_pass_rate": float(real_row["pass_rate"]), + "sim_pass_rate": float(scorer["slo"]["pass_rate"]), + "pass_rate_absolute_error": pass_error, + "offered_req_s_per_gpu": float(real_row["offered_req_s_per_gpu"]), + "sim_throughput_req_s_per_gpu": float( + scorer["throughput_requests_per_second_per_gpu"] + ), + } + ) + if sim_feasible: + capacity_lower_bounds[key[0]] = max( + capacity_lower_bounds.get(key[0], 0.0), + float(real_row["offered_req_s_per_gpu"]), + ) + agreement = 12 - false_feasible - false_infeasible + summaries[mode] = { + "probe_classification": { + "agreement": agreement, + "accuracy": agreement / 12, + "false_feasible": false_feasible, + "false_infeasible": false_infeasible, + }, + "pass_rate_mae": statistics.mean(pass_errors), + "feasible_probe_count": sum( + bool(row["scorer"]["slo"]["feasible"]) for row in predicted.values() + ), + "p1_capacity_lower_bounds_req_s_per_gpu": { + cell: capacity_lower_bounds.get(cell, 0.0) + for cell in sorted(controller["cells"]) + }, + "rank_identifiable": bool(capacity_lower_bounds), + } + + output = { + "schema": "frontier-qwen30-p1-profile-ablation.v1", + "scope": { + "cells": 6, + "probes_per_cell": 2, + "roles": ["low1", "high1"], + "reading": "held-out boundary classification, not a complete capacity sweep", + }, + "sources": { + "controller_state": str(args.controller_state.resolve()), + "historical_calibrated": str(args.calibrated.resolve()), + "historical_profile_only": str(args.old_profile_only.resolve()), + "vllm020_profile_only": str(args.new_profile_only.resolve()), + }, + "summaries": summaries, + "rows": rows, + } + args.output_json.parent.mkdir(parents=True, exist_ok=True) + args.output_json.write_text(json.dumps(output, indent=2, sort_keys=True) + "\n") + args.output_csv.parent.mkdir(parents=True, exist_ok=True) + with args.output_csv.open("w", newline="") as handle: + writer = csv.DictWriter(handle, fieldnames=list(rows[0])) + writer.writeheader() + writer.writerows(rows) + print(args.output_json) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/analyze_routing_mismatch.py b/runs/frontier-qwen30-vllm020-profile-v1/analyze_routing_mismatch.py new file mode 100644 index 0000000..56d8492 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/analyze_routing_mismatch.py @@ -0,0 +1,152 @@ +#!/usr/bin/env python3 +"""Compare captured vLLM expert loads with Frontier's fixed routing prior.""" + +from __future__ import annotations + +import argparse +import json +import math +import statistics +from pathlib import Path +from typing import Any + +import numpy as np + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--routing", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--seed", type=int, default=42) + return parser.parse_args() + + +def gini(values: np.ndarray) -> float: + values = np.asarray(values, dtype=np.float64) + if values.sum() == 0: + return 0.0 + ordered = np.sort(values) + n = len(ordered) + indices = np.arange(1, n + 1, dtype=np.float64) + return float((2 * np.sum(indices * ordered) / np.sum(ordered) - (n + 1)) / n) + + +def stats(values: np.ndarray) -> dict[str, float]: + values = np.asarray(values, dtype=np.float64) + mean = float(np.mean(values)) + return { + "load_cv": float(np.std(values) / mean), + "load_gini": gini(values), + "max_load_ratio": float(np.max(values) / mean), + "expert_utilization": float(np.count_nonzero(values) / len(values)), + } + + +def proportional_counts(total: int, ratios: np.ndarray) -> np.ndarray: + exact = total * ratios / ratios.sum() + counts = np.floor(exact).astype(np.int64) + remainder = total - int(counts.sum()) + order = sorted( + range(len(ratios)), + key=lambda i: (-(exact[i] - counts[i]), -(ratios[i] / ratios.sum()), i), + ) + for index in range(remainder): + counts[order[index % len(order)]] += 1 + assert int(counts.sum()) == total + return counts + + +def correlation(left: np.ndarray, right: np.ndarray) -> float: + value = float(np.corrcoef(left, right)[0, 1]) + return value if math.isfinite(value) else 0.0 + + +def distribution(values: list[float]) -> dict[str, float]: + return { + "min": min(values), + "median": statistics.median(values), + "max": max(values), + } + + +def main() -> None: + args = parse_args() + routing = json.loads(args.routing.read_text()) + phases = routing["phases"] + layer_count = len(phases["prefill"]["per_layer"]) + expert_count = len(phases["prefill"]["per_layer"][0]["counts"]) + + rows: list[dict[str, Any]] = [] + for phase in ("prefill", "decode"): + for layer, actual in enumerate(phases[phase]["per_layer"]): + rng = np.random.RandomState(args.seed + layer) + ratios = rng.uniform(0.1, 1.0, expert_count) + synthetic = proportional_counts(int(actual["total_routed_tokens"]), ratios) + synthetic_stats = stats(synthetic) + for name in synthetic_stats: + if not math.isclose( + stats(np.asarray(actual["counts"]))[name], + float(actual[name]), + rel_tol=0, + abs_tol=1e-12, + ): + raise ValueError(f"captured metric mismatch: {phase} layer {layer} {name}") + rows.append( + { + "phase": phase, + "layer": layer, + "total_routed_tokens": int(actual["total_routed_tokens"]), + "actual": {name: float(actual[name]) for name in synthetic_stats}, + "frontier_simulation": synthetic_stats, + "actual_vs_frontier_pearson": correlation( + np.asarray(actual["counts"], dtype=np.float64), synthetic + ), + } + ) + + phase_summary: dict[str, Any] = {} + for phase in ("prefill", "decode"): + selected = [row for row in rows if row["phase"] == phase] + phase_summary[phase] = { + "token_count": int(phases[phase]["token_count"]), + "actual": { + name: distribution([row["actual"][name] for row in selected]) + for name in selected[0]["actual"] + }, + "frontier_simulation": { + name: distribution([row["frontier_simulation"][name] for row in selected]) + for name in selected[0]["frontier_simulation"] + }, + "actual_vs_frontier_pearson": distribution( + [row["actual_vs_frontier_pearson"] for row in selected] + ), + } + + phase_correlations = [] + for layer in range(layer_count): + prefill = np.asarray(phases["prefill"]["per_layer"][layer]["counts"]) + decode = np.asarray(phases["decode"]["per_layer"][layer]["counts"]) + phase_correlations.append(correlation(prefill, decode)) + + output = { + "schema": "frontier-routing-mismatch.v1", + "source": str(args.routing.resolve()), + "frontier_contract": { + "mode": "simulation", + "seed": args.seed, + "allocation": "per-layer fixed Uniform(0.1, 1.0), normalized once and reused for every batch and phase", + "layer_count": layer_count, + "expert_count": expert_count, + }, + "phase_summary": phase_summary, + "actual_prefill_vs_decode_pearson": distribution(phase_correlations), + "frontier_prefill_vs_decode_pearson": 1.0, + "rows": rows, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(output, indent=2, sort_keys=True) + "\n") + print(args.output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/analyze_s2_profile_ablation.py b/runs/frontier-qwen30-vllm020-profile-v1/analyze_s2_profile_ablation.py new file mode 100644 index 0000000..ac4dfec --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/analyze_s2_profile_ablation.py @@ -0,0 +1,229 @@ +#!/usr/bin/env python3 +"""Score a replacement-profile S2 sweep against the frozen real oracle.""" + +from __future__ import annotations + +import argparse +import importlib.util +import json +import math +import sys +from pathlib import Path +from typing import Any + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--shard-metrics", type=Path, action="append", required=True) + parser.add_argument("--ground-truth", type=Path, required=True) + parser.add_argument("--historical-metrics", type=Path, required=True) + parser.add_argument("--historical-analyzer", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def load_module(path: Path): + spec = importlib.util.spec_from_file_location("simfid_s2_analyzer", path) + if spec is None or spec.loader is None: + raise ImportError(path) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def ranking(scores: dict[str, float]) -> list[dict[str, Any]]: + return [ + {"rank": index + 1, "cell": cell, "score": score} + for index, (cell, score) in enumerate( + sorted(scores.items(), key=lambda item: (-item[1], item[0])) + ) + ] + + +def robust_loao( + analyzer: Any, + runs: list[dict[str, Any]], + mode: str, + reading: str, + real_scores: dict[str, float], +) -> dict[str, Any]: + """Historical LOAO with an explicit null range for all-tied concordance.""" + anchors = sorted( + {(row["cell_id"], int(row["probe_index"])) for row in runs}, + key=lambda item: (analyzer.CELL_ORDER.index(item[0]), item[1]), + ) + replicates = [] + undefined = [] + for cell, probe in anchors: + scores, detail = analyzer.cell_scores( + runs, mode, reading, removed=(cell, probe) + ) + if scores is None: + undefined.append( + {"removed_cell_id": cell, "removed_probe_index": probe, **detail} + ) + continue + metric = analyzer.rank_metrics(real_scores, scores) + replicates.append( + { + "removed_cell_id": cell, + "removed_probe_index": probe, + "top1_optimistic_regret": metric["top1"]["optimistic_regret"], + "top1_worst_case_regret": metric["top1"]["worst_case_regret"], + "top5_minimum_exact_five_overlap": metric["top5"]["minimum_exact_five_overlap"], + "top5_maximum_exact_five_overlap": metric["top5"]["maximum_exact_five_overlap"], + "top5_optimistic_regret": metric["top5"]["optimistic_regret"], + "top5_worst_case_regret": metric["top5"]["worst_case_regret"], + "tau_b": metric["kendall_tau_b"]["tau_b"], + "pairwise_exact_sign_accuracy": metric["pairwise_direction"]["exact_sign_accuracy"], + "pairwise_non_tied_concordance": metric["pairwise_direction"]["non_tied_concordance"], + "trap_reproduced": metric["named_interactions"]["trap_reproduced"], + "tp2_mns32_unique_global_best": metric["named_interactions"]["tp2_mns32_unique_global_best"], + } + ) + scalar_keys = ( + "top1_optimistic_regret", + "top1_worst_case_regret", + "top5_minimum_exact_five_overlap", + "top5_maximum_exact_five_overlap", + "top5_optimistic_regret", + "top5_worst_case_regret", + "tau_b", + "pairwise_exact_sign_accuracy", + "pairwise_non_tied_concordance", + ) + ranges = {} + for key in scalar_keys: + values = [row[key] for row in replicates if row[key] is not None] + ranges[key] = { + "min": min(values) if values else None, + "max": max(values) if values else None, + } + return { + "replicate_count": len(anchors), + "defined_replicates": len(replicates), + "undefined_replicates": len(undefined), + "undefined": undefined, + "ranges": ranges, + "trap_reproduced_count": sum(row["trap_reproduced"] for row in replicates), + "tp2_mns32_unique_global_best_count": sum( + row["tp2_mns32_unique_global_best"] for row in replicates + ), + "replicates": replicates, + "range_semantics": "deterministic LOAO sensitivity ranges, not confidence intervals", + } + + +def main() -> None: + args = parse_args() + analyzer = load_module(args.historical_analyzer) + ground = json.loads(args.ground_truth.read_text()) + cells = {str(cell["cell_id"]): cell for cell in ground["cells"]} + if set(cells) != set(analyzer.CELL_ORDER): + raise ValueError("ground-truth cells differ from the frozen 3x4 surface") + + result_rows: list[dict[str, Any]] = [] + sources = [] + for path in args.shard_metrics: + payload = json.loads(path.read_text()) + if payload["status"] != "PASS": + raise ValueError(f"shard did not pass: {path}") + sources.append({"path": str(path.resolve()), "runs": len(payload["results"])}) + result_rows.extend(payload["results"]) + if len(result_rows) != 92: + raise ValueError(f"expected 92 replacement-profile probes, found {len(result_rows)}") + + seen: set[tuple[str, int]] = set() + runs = [] + for row in result_rows: + cell_id = str(row["cell"]) + probe_index = int(row["probe_index"]) + key = (cell_id, probe_index) + if key in seen: + raise ValueError(f"duplicate probe {key}") + seen.add(key) + real_probe = cells[cell_id]["probe_history"][probe_index] + if not math.isclose( + float(real_probe["sampling_u"]), float(row["sampling_u"]), rel_tol=0, abs_tol=1e-15 + ): + raise ValueError(f"sampling_u mismatch for {key}") + if int(real_probe["request_count"]) != int(row["selected_count"]): + raise ValueError(f"request count mismatch for {key}") + runs.append( + { + **row, + "cell_id": cell_id, + "mode": "vllm020-profile-only", + "probe_index": probe_index, + "sampling_u": float(row["sampling_u"]), + "request_count": int(row["selected_count"]), + "tensor_parallel_size": int(row["tensor_parallel_size"]), + "real_anchor": { + "feasible": bool(real_probe["feasible"]), + "pass_rate": float(real_probe["pass_rate"]), + "request_count": int(real_probe["request_count"]), + }, + } + ) + + historical = json.loads(args.historical_metrics.read_text()) + real_scores = {cell: float(value) for cell, value in historical["real_scores"].items()} + analyses = {} + for reading in analyzer.READINGS: + scores, detail = analyzer.cell_scores(runs, "vllm020-profile-only", reading) + if scores is None: + raise ValueError(f"undefined {reading} scores: {detail}") + analysis = { + "reading": reading, + "simulated_scores": scores, + "ranking": ranking(scores), + "cell_score_details": detail["cells"], + "metrics": analyzer.rank_metrics(real_scores, scores), + "loao": robust_loao( + analyzer, runs, "vllm020-profile-only", reading, real_scores + ), + } + if reading == "SLO-gated": + analysis["false_feasibility"] = analyzer.false_feasibility( + runs, "vllm020-profile-only" + ) + analyses[reading] = analysis + + historical_modes = {} + for label, key in ( + ("historical-profile-only", "uncalibrated/SLO-gated"), + ("historical-per-tp-calibration", "frozen-calibrated/SLO-gated"), + ): + value = historical["analyses"][key] + historical_modes[label] = { + "simulated_scores": value["simulated_scores"], + "ranking": ranking(value["simulated_scores"]), + "metrics": value["metrics"], + "false_feasibility": value["false_feasibility"], + } + + output = { + "schema": "frontier-qwen30-s2-profile-ablation.v1", + "scope": { + "cells": len(analyzer.CELL_ORDER), + "probes": len(runs), + "trace_horizon_seconds": 60, + "calibration_a_tp": 1.0, + }, + "sources": { + "replacement_profile_shards": sources, + "ground_truth": str(args.ground_truth.resolve()), + "historical_metrics": str(args.historical_metrics.resolve()), + }, + "real_scores": real_scores, + "historical_modes": historical_modes, + "vllm020_profile_only": analyses, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(output, indent=2, sort_keys=True) + "\n") + print(args.output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-initial-op-trace-delta.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-initial-op-trace-delta.json new file mode 100644 index 0000000..58346dc --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-initial-op-trace-delta.json @@ -0,0 +1,87 @@ +{ + "comparison_contract": { + "calibration_a_tp": 1.0, + "fixture": "fidelity_p1_tp1_mns64_low1", + "scheduler_state": "batch_id=0, layer_id=0 before profile-dependent trajectories diverge" + }, + "layer_component_sum_ms": { + "historical_profile": 1.148334, + "new_over_old": 2.700194368537377, + "vllm020_profile": 3.1007249999999997 + }, + "rows": [ + { + "component": "input_layernorm", + "historical_profile_ms": 0.010619, + "new_over_old": 1.955833882663151, + "vllm020_profile_ms": 0.020769 + }, + { + "component": "attn_pre_proj", + "historical_profile_ms": 0.204389, + "new_over_old": 1.0199276869107439, + "vllm020_profile_ms": 0.208462 + }, + { + "component": "attn_rope", + "historical_profile_ms": 0.018908, + "new_over_old": 7.394171779141103, + "vllm020_profile_ms": 0.139809 + }, + { + "component": "attn_kv_cache_save", + "historical_profile_ms": 0.014226, + "new_over_old": 1.427386475467454, + "vllm020_profile_ms": 0.020306 + }, + { + "component": "attn_prefill", + "historical_profile_ms": 0.107759, + "new_over_old": 2.670709639102071, + "vllm020_profile_ms": 0.287793 + }, + { + "component": "attn_decode", + "historical_profile_ms": 0.0, + "new_over_old": null, + "vllm020_profile_ms": 0.0 + }, + { + "component": "attn_post_proj", + "historical_profile_ms": 0.134153, + "new_over_old": 0.967559428413826, + "vllm020_profile_ms": 0.129801 + }, + { + "component": "post_attention_layernorm", + "historical_profile_ms": 0.011481, + "new_over_old": 0.9000087100426792, + "vllm020_profile_ms": 0.010333 + }, + { + "component": "moe_gating_linear", + "historical_profile_ms": 0.011802, + "new_over_old": 2.2642772411455687, + "vllm020_profile_ms": 0.026723 + }, + { + "component": "moe_gating_routing_topk", + "historical_profile_ms": 0.009662, + "new_over_old": 2.2922790312564687, + "vllm020_profile_ms": 0.022148 + }, + { + "component": "moe_shuffling", + "historical_profile_ms": 0.030431, + "new_over_old": 0.0, + "vllm020_profile_ms": 0.0 + }, + { + "component": "moe_grouped_gemm", + "historical_profile_ms": 0.594904, + "new_over_old": 3.756204362384519, + "vllm020_profile_ms": 2.234581 + } + ], + "schema": "frontier-initial-op-trace-delta.v1" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.csv b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.csv new file mode 100644 index 0000000..2750398 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.csv @@ -0,0 +1,37 @@ +mode,cell,role,real_feasible,sim_feasible,real_pass_rate,sim_pass_rate,pass_rate_absolute_error,offered_req_s_per_gpu,sim_throughput_req_s_per_gpu +historical-calibrated,tp1_mns64,high1,False,True,0.24581005586592178,0.9888268156424581,0.7430167597765364,2.9833333333333334,2.8823345959582403 +historical-calibrated,tp1_mns64,low1,True,True,1.0,1.0,0.0,2.033333333333333,1.9806287931485522 +historical-calibrated,tp1_mns8,high1,False,False,0.20670391061452514,0.16759776536312848,0.03910614525139666,2.9833333333333334,2.162661799726284 +historical-calibrated,tp1_mns8,low1,True,True,1.0,0.9754098360655737,0.024590163934426257,2.033333333333333,1.9806313066623409 +historical-calibrated,tp2_mns64,high1,True,True,1.0,1.0,0.0,2.875,2.809583898291986 +historical-calibrated,tp2_mns64,low1,True,True,1.0,1.0,0.0,1.9583333333333333,1.911740196570019 +historical-calibrated,tp2_mns8,high1,False,False,0.20238095238095238,0.15476190476190477,0.047619047619047616,2.8,2.083217332903509 +historical-calibrated,tp2_mns8,low1,True,True,1.0,0.9868995633187773,0.013100436681222738,1.9083333333333334,1.8637394408880168 +historical-calibrated,tp4_mns16,high1,False,False,0.10666666666666667,0.2693333333333333,0.16266666666666663,3.125,2.6829294600763345 +historical-calibrated,tp4_mns16,low1,False,True,0.6196078431372549,1.0,0.3803921568627451,2.125,2.0865667413849516 +historical-calibrated,tp4_mns64,high1,True,True,1.0,1.0,0.0,3.125,3.0718260341342285 +historical-calibrated,tp4_mns64,low1,True,True,1.0,1.0,0.0,2.125,2.0967062495193307 +historical-profile-only,tp1_mns64,high1,False,False,0.24581005586592178,0.24581005586592178,0.0,2.9833333333333334,2.82546810606982 +historical-profile-only,tp1_mns64,low1,True,False,1.0,0.7377049180327869,0.2622950819672131,2.033333333333333,1.9427723213024963 +historical-profile-only,tp1_mns8,high1,False,False,0.20670391061452514,0.09497206703910614,0.111731843575419,2.9833333333333334,1.5728783582441022 +historical-profile-only,tp1_mns8,low1,True,False,1.0,0.1721311475409836,0.8278688524590164,2.033333333333333,1.5629164685645112 +historical-profile-only,tp2_mns64,high1,True,False,1.0,0.2608695652173913,0.7391304347826086,2.875,2.7121507155597584 +historical-profile-only,tp2_mns64,low1,True,False,1.0,0.8382978723404255,0.16170212765957448,1.9583333333333333,1.849101894293262 +historical-profile-only,tp2_mns8,high1,False,False,0.20238095238095238,0.041666666666666664,0.16071428571428573,2.8,0.9859710483293953 +historical-profile-only,tp2_mns8,low1,True,False,1.0,0.0611353711790393,0.9388646288209607,1.9083333333333334,0.9771135206177947 +historical-profile-only,tp4_mns16,high1,False,False,0.10666666666666667,0.028,0.07866666666666668,3.125,0.9604809617478992 +historical-profile-only,tp4_mns16,low1,False,False,0.6196078431372549,0.058823529411764705,0.5607843137254902,2.125,0.9495221159963836 +historical-profile-only,tp4_mns64,high1,True,False,1.0,0.0026666666666666666,0.9973333333333333,3.125,2.281094127221267 +historical-profile-only,tp4_mns64,low1,True,False,1.0,0.4627450980392157,0.5372549019607843,2.125,1.9793960278798877 +vllm020-profile-only,tp1_mns64,high1,False,False,0.24581005586592178,0.0,0.24581005586592178,2.9833333333333334,2.074240230678028 +vllm020-profile-only,tp1_mns64,low1,True,False,1.0,0.0,1.0,2.033333333333333,1.6189525827713862 +vllm020-profile-only,tp1_mns8,high1,False,False,0.20670391061452514,0.0,0.20670391061452514,2.9833333333333334,0.5271192714936628 +vllm020-profile-only,tp1_mns8,low1,True,False,1.0,0.0,1.0,2.033333333333333,0.5225514701978402 +vllm020-profile-only,tp2_mns64,high1,True,False,1.0,0.0,1.0,2.875,1.7237810220598582 +vllm020-profile-only,tp2_mns64,low1,True,False,1.0,0.0,1.0,1.9583333333333333,1.608919184557297 +vllm020-profile-only,tp2_mns8,high1,False,False,0.20238095238095238,0.0,0.20238095238095238,2.8,0.39315152956277477 +vllm020-profile-only,tp2_mns8,low1,True,False,1.0,0.0,1.0,1.9083333333333334,0.3894201655965407 +vllm020-profile-only,tp4_mns16,high1,False,False,0.10666666666666667,0.0,0.10666666666666667,3.125,0.43700447014112276 +vllm020-profile-only,tp4_mns16,low1,False,False,0.6196078431372549,0.0,0.6196078431372549,2.125,0.43475717185302865 +vllm020-profile-only,tp4_mns64,high1,True,False,1.0,0.0,1.0,3.125,1.2598449015844073 +vllm020-profile-only,tp4_mns64,low1,True,False,1.0,0.0,1.0,2.125,1.2025831680037373 diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.json new file mode 100644 index 0000000..0bf3077 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1-profile-ablation.json @@ -0,0 +1,511 @@ +{ + "rows": [ + { + "cell": "tp1_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.7430167597765364, + "real_feasible": false, + "real_pass_rate": 0.24581005586592178, + "role": "high1", + "sim_feasible": true, + "sim_pass_rate": 0.9888268156424581, + "sim_throughput_req_s_per_gpu": 2.8823345959582403 + }, + { + "cell": "tp1_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 0.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 1.9806287931485522 + }, + { + "cell": "tp1_mns8", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.03910614525139666, + "real_feasible": false, + "real_pass_rate": 0.20670391061452514, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.16759776536312848, + "sim_throughput_req_s_per_gpu": 2.162661799726284 + }, + { + "cell": "tp1_mns8", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 0.024590163934426257, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 0.9754098360655737, + "sim_throughput_req_s_per_gpu": 1.9806313066623409 + }, + { + "cell": "tp2_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.875, + "pass_rate_absolute_error": 0.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 2.809583898291986 + }, + { + "cell": "tp2_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 1.9583333333333333, + "pass_rate_absolute_error": 0.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 1.911740196570019 + }, + { + "cell": "tp2_mns8", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.8, + "pass_rate_absolute_error": 0.047619047619047616, + "real_feasible": false, + "real_pass_rate": 0.20238095238095238, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.15476190476190477, + "sim_throughput_req_s_per_gpu": 2.083217332903509 + }, + { + "cell": "tp2_mns8", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 1.9083333333333334, + "pass_rate_absolute_error": 0.013100436681222738, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 0.9868995633187773, + "sim_throughput_req_s_per_gpu": 1.8637394408880168 + }, + { + "cell": "tp4_mns16", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 0.16266666666666663, + "real_feasible": false, + "real_pass_rate": 0.10666666666666667, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.2693333333333333, + "sim_throughput_req_s_per_gpu": 2.6829294600763345 + }, + { + "cell": "tp4_mns16", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 0.3803921568627451, + "real_feasible": false, + "real_pass_rate": 0.6196078431372549, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 2.0865667413849516 + }, + { + "cell": "tp4_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 0.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 3.0718260341342285 + }, + { + "cell": "tp4_mns64", + "mode": "historical-calibrated", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 0.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": true, + "sim_pass_rate": 1.0, + "sim_throughput_req_s_per_gpu": 2.0967062495193307 + }, + { + "cell": "tp1_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.0, + "real_feasible": false, + "real_pass_rate": 0.24581005586592178, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.24581005586592178, + "sim_throughput_req_s_per_gpu": 2.82546810606982 + }, + { + "cell": "tp1_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 0.2622950819672131, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.7377049180327869, + "sim_throughput_req_s_per_gpu": 1.9427723213024963 + }, + { + "cell": "tp1_mns8", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.111731843575419, + "real_feasible": false, + "real_pass_rate": 0.20670391061452514, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.09497206703910614, + "sim_throughput_req_s_per_gpu": 1.5728783582441022 + }, + { + "cell": "tp1_mns8", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 0.8278688524590164, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.1721311475409836, + "sim_throughput_req_s_per_gpu": 1.5629164685645112 + }, + { + "cell": "tp2_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.875, + "pass_rate_absolute_error": 0.7391304347826086, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.2608695652173913, + "sim_throughput_req_s_per_gpu": 2.7121507155597584 + }, + { + "cell": "tp2_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 1.9583333333333333, + "pass_rate_absolute_error": 0.16170212765957448, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.8382978723404255, + "sim_throughput_req_s_per_gpu": 1.849101894293262 + }, + { + "cell": "tp2_mns8", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.8, + "pass_rate_absolute_error": 0.16071428571428573, + "real_feasible": false, + "real_pass_rate": 0.20238095238095238, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.041666666666666664, + "sim_throughput_req_s_per_gpu": 0.9859710483293953 + }, + { + "cell": "tp2_mns8", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 1.9083333333333334, + "pass_rate_absolute_error": 0.9388646288209607, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0611353711790393, + "sim_throughput_req_s_per_gpu": 0.9771135206177947 + }, + { + "cell": "tp4_mns16", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 0.07866666666666668, + "real_feasible": false, + "real_pass_rate": 0.10666666666666667, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.028, + "sim_throughput_req_s_per_gpu": 0.9604809617478992 + }, + { + "cell": "tp4_mns16", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 0.5607843137254902, + "real_feasible": false, + "real_pass_rate": 0.6196078431372549, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.058823529411764705, + "sim_throughput_req_s_per_gpu": 0.9495221159963836 + }, + { + "cell": "tp4_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 0.9973333333333333, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0026666666666666666, + "sim_throughput_req_s_per_gpu": 2.281094127221267 + }, + { + "cell": "tp4_mns64", + "mode": "historical-profile-only", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 0.5372549019607843, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.4627450980392157, + "sim_throughput_req_s_per_gpu": 1.9793960278798877 + }, + { + "cell": "tp1_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.24581005586592178, + "real_feasible": false, + "real_pass_rate": 0.24581005586592178, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 2.074240230678028 + }, + { + "cell": "tp1_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 1.6189525827713862 + }, + { + "cell": "tp1_mns8", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.9833333333333334, + "pass_rate_absolute_error": 0.20670391061452514, + "real_feasible": false, + "real_pass_rate": 0.20670391061452514, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.5271192714936628 + }, + { + "cell": "tp1_mns8", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.033333333333333, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.5225514701978402 + }, + { + "cell": "tp2_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.875, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 1.7237810220598582 + }, + { + "cell": "tp2_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 1.9583333333333333, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 1.608919184557297 + }, + { + "cell": "tp2_mns8", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.8, + "pass_rate_absolute_error": 0.20238095238095238, + "real_feasible": false, + "real_pass_rate": 0.20238095238095238, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.39315152956277477 + }, + { + "cell": "tp2_mns8", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 1.9083333333333334, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.3894201655965407 + }, + { + "cell": "tp4_mns16", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 0.10666666666666667, + "real_feasible": false, + "real_pass_rate": 0.10666666666666667, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.43700447014112276 + }, + { + "cell": "tp4_mns16", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 0.6196078431372549, + "real_feasible": false, + "real_pass_rate": 0.6196078431372549, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 0.43475717185302865 + }, + { + "cell": "tp4_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 3.125, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "high1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 1.2598449015844073 + }, + { + "cell": "tp4_mns64", + "mode": "vllm020-profile-only", + "offered_req_s_per_gpu": 2.125, + "pass_rate_absolute_error": 1.0, + "real_feasible": true, + "real_pass_rate": 1.0, + "role": "low1", + "sim_feasible": false, + "sim_pass_rate": 0.0, + "sim_throughput_req_s_per_gpu": 1.2025831680037373 + } + ], + "schema": "frontier-qwen30-p1-profile-ablation.v1", + "scope": { + "cells": 6, + "probes_per_cell": 2, + "reading": "held-out boundary classification, not a complete capacity sweep", + "roles": [ + "low1", + "high1" + ] + }, + "sources": { + "controller_state": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/controller-state.json", + "historical_calibrated": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/results/metrics.json", + "historical_profile_only": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/comparison/old-profile-only-v2/results/metrics.json", + "vllm020_profile_only": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/comparison/new-profile-only-v2/results/metrics.json" + }, + "summaries": { + "historical-calibrated": { + "feasible_probe_count": 9, + "p1_capacity_lower_bounds_req_s_per_gpu": { + "tp1_mns64": 2.9833333333333334, + "tp1_mns8": 2.033333333333333, + "tp2_mns64": 2.875, + "tp2_mns8": 1.9083333333333334, + "tp4_mns16": 2.125, + "tp4_mns64": 3.125 + }, + "pass_rate_mae": 0.11754094806600345, + "probe_classification": { + "accuracy": 0.8333333333333334, + "agreement": 10, + "false_feasible": 2, + "false_infeasible": 0 + }, + "rank_identifiable": true + }, + "historical-profile-only": { + "feasible_probe_count": 0, + "p1_capacity_lower_bounds_req_s_per_gpu": { + "tp1_mns64": 0.0, + "tp1_mns8": 0.0, + "tp2_mns64": 0.0, + "tp2_mns8": 0.0, + "tp4_mns16": 0.0, + "tp4_mns64": 0.0 + }, + "pass_rate_mae": 0.44802887255544604, + "probe_classification": { + "accuracy": 0.4166666666666667, + "agreement": 5, + "false_feasible": 0, + "false_infeasible": 7 + }, + "rank_identifiable": false + }, + "vllm020-profile-only": { + "feasible_probe_count": 0, + "p1_capacity_lower_bounds_req_s_per_gpu": { + "tp1_mns64": 0.0, + "tp1_mns8": 0.0, + "tp2_mns64": 0.0, + "tp2_mns8": 0.0, + "tp4_mns16": 0.0, + "tp4_mns64": 0.0 + }, + "pass_rate_mae": 0.69843078572211, + "probe_classification": { + "accuracy": 0.4166666666666667, + "agreement": 5, + "false_feasible": 0, + "false_infeasible": 7 + }, + "rank_identifiable": false + } + } +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1b-opprof-summary.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1b-opprof-summary.json new file mode 100644 index 0000000..34efb09 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-p1b-opprof-summary.json @@ -0,0 +1,2171 @@ +{ + "cell_root": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells", + "cells": [ + { + "cell": "tp1_mns64", + "cudagraph_runtime_mode_counts": { + "FULL": 26305, + "NONE": 698, + "PIECEWISE": 141 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 33, + "mean": 4.150661496350365, + "min": 1, + "p50": 3, + "p95": 13, + "p99": 21 + }, + "kv_usage": { + "max": 0.35797781996203415, + "mean": 0.03727774678883895, + "min": 0.00029973024278151517, + "p50": 0.023728644220201822, + "p95": 0.13018283544809672, + "p99": 0.21860325706863826 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 33, + "mean": 4.150661496350365, + "min": 1, + "p50": 3, + "p95": 13, + "p99": 21 + }, + "steps": 26304, + "submit_to_complete_ms": { + "max": 771.351779, + "mean": 20.43450573646594, + "min": 5.345856, + "p50": 15.161645, + "p95": 35.769253, + "p99": 179.487345 + }, + "total_tokens": { + "max": 33, + "mean": 4.150661496350365, + "min": 1, + "p50": 3, + "p95": 13, + "p99": 21 + } + }, + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.02532720551503642, + "mean": 0.02532720551503642, + "min": 0.02532720551503642, + "p50": 0.02532720551503642, + "p95": 0.02532720551503642, + "p99": 0.02532720551503642 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 1, + "submit_to_complete_ms": { + "max": 7.226884, + "mean": 7.226884, + "min": 7.226884, + "p50": 7.226884, + "p95": 7.226884, + "p99": 7.226884 + }, + "total_tokens": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.022329903087221492, + "mean": 0.00819839067915799, + "min": 0.000549505445099463, + "p50": 0.007043660705365218, + "p95": 0.020731341792386893, + "p99": 0.022329903087221492 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 7089, + "mean": 2337.153846153846, + "min": 163, + "p50": 1828, + "p95": 6569, + "p99": 7089 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 52, + "submit_to_complete_ms": { + "max": 462.299574, + "mean": 154.1463595769231, + "min": 41.988631, + "p50": 121.390432, + "p95": 426.234755, + "p99": 462.299574 + }, + "total_tokens": { + "max": 7089, + "mean": 2337.153846153846, + "min": 163, + "p50": 1828, + "p95": 6569, + "p99": 7089 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.011539614347087612, + "mean": 0.002140381349093522, + "min": 0.00029973024278151517, + "p50": 0.0007992806474172998, + "p95": 0.011539614347087612, + "p99": 0.011539614347087612 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 118, + "mean": 43.07692307692308, + "min": 6, + "p50": 27, + "p95": 118, + "p99": 118 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 13, + "submit_to_complete_ms": { + "max": 20.969793, + "mean": 15.511612692307692, + "min": 10.185672, + "p50": 15.302969, + "p95": 20.969793, + "p99": 20.969793 + }, + "total_tokens": { + "max": 118, + "mean": 43.07692307692308, + "min": 6, + "p50": 27, + "p95": 118, + "p99": 118 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 32, + "mean": 7.1671826625387, + "min": 1, + "p50": 5, + "p95": 20, + "p99": 27 + }, + "kv_usage": { + "max": 0.35622939354580874, + "mean": 0.08236302498030411, + "min": 0.003446897791987258, + "p50": 0.06049555400139872, + "p95": 0.2183534818663203, + "p99": 0.3386452193026276 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8189, + "mean": 3407.3668730650156, + "min": 119, + "p50": 2823, + "p95": 7968, + "p99": 8188 + }, + "queue_waiting": { + "max": 1, + "mean": 0.009287925696594427, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 33, + "mean": 8.292569659442725, + "min": 2, + "p50": 6, + "p95": 21, + "p99": 28 + }, + "steps": 646, + "submit_to_complete_ms": { + "max": 1031.099195, + "mean": 262.67329438235294, + "min": 41.593929, + "p50": 208.770469, + "p95": 560.110521, + "p99": 836.430251 + }, + "total_tokens": { + "max": 8192, + "mean": 3414.534055727554, + "min": 131, + "p50": 2838, + "p95": 7973, + "p99": 8192 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 26, + "mean": 5.8203125, + "min": 1, + "p50": 4, + "p95": 15, + "p99": 25 + }, + "kv_usage": { + "max": 0.31431711459686285, + "mean": 0.05598086721950245, + "min": 0.0004995504046357846, + "p50": 0.03536816864821657, + "p95": 0.13977420321710465, + "p99": 0.29343590768308525 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 122, + "mean": 38.9140625, + "min": 2, + "p50": 27, + "p95": 94, + "p99": 119 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 27, + "mean": 6.8203125, + "min": 2, + "p50": 5, + "p95": 16, + "p99": 26 + }, + "steps": 128, + "submit_to_complete_ms": { + "max": 549.993756, + "mean": 38.9609057265625, + "min": 13.155768, + "p50": 24.716054, + "p95": 86.104443, + "p99": 407.701862 + }, + "total_tokens": { + "max": 123, + "mean": 44.734375, + "min": 3, + "p50": 35, + "p95": 105, + "p99": 120 + } + } + ], + "max_num_seqs": 64, + "phase_counts": { + "pure_decode": 26304, + "pure_prefill": 66, + "true_mixed": 774 + }, + "prefix": { + "hit_rate": 0.1199154023961761, + "hits": 317232, + "queries": 2645465 + }, + "records": 27144, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp1_mns64/opprof/opprof-v1-dp0-pid234511-1784005631788937266.jsonl", + "tensor_parallel_size": 1 + }, + { + "cell": "tp1_mns8", + "cudagraph_runtime_mode_counts": { + "FULL": 27496, + "NONE": 800, + "PIECEWISE": 13 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 8, + "mean": 3.827968721585743, + "min": 1, + "p50": 3, + "p95": 8, + "p99": 8 + }, + "kv_usage": { + "max": 0.13572814567888947, + "mean": 0.03436450743578334, + "min": 0.0002985223145429705, + "p50": 0.02666799343250903, + "p95": 0.09512911090103982, + "p99": 0.11562764316632668 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 13, + "mean": 0.7083106019276232, + "min": 0, + "p50": 0, + "p95": 5, + "p99": 9 + }, + "scheduled_requests": { + "max": 8, + "mean": 3.827968721585743, + "min": 1, + "p50": 3, + "p95": 8, + "p99": 8 + }, + "steps": 27495, + "submit_to_complete_ms": { + "max": 561.758474, + "mean": 19.407119342862337, + "min": 5.267886, + "p50": 15.55276, + "p95": 24.123839, + "p99": 151.668642 + }, + "total_tokens": { + "max": 8, + "mean": 3.827968721585743, + "min": 1, + "p50": 3, + "p95": 8, + "p99": 8 + } + }, + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.025225135578884506, + "mean": 0.025225135578884506, + "min": 0.025225135578884506, + "p50": 0.025225135578884506, + "p95": 0.025225135578884506, + "p99": 0.025225135578884506 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 1, + "submit_to_complete_ms": { + "max": 7.195164, + "mean": 7.195164, + "min": 7.195164, + "p50": 7.195164, + "p95": 7.195164, + "p99": 7.195164 + }, + "total_tokens": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.02318523309617393, + "mean": 0.007211759417999546, + "min": 0.0002985223145429705, + "p50": 0.005373401661774246, + "p95": 0.02064779342255829, + "p99": 0.02318523309617393 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 7089, + "mean": 2017.593220338983, + "min": 18, + "p50": 1501, + "p95": 6569, + "p99": 7089 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 59, + "submit_to_complete_ms": { + "max": 463.900987, + "mean": 140.730381559322, + "min": 41.733004, + "p50": 101.529539, + "p95": 425.237592, + "p99": 463.900987 + }, + "total_tokens": { + "max": 7089, + "mean": 2017.593220338983, + "min": 18, + "p50": 1501, + "p95": 6569, + "p99": 7089 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.011493109109905975, + "mean": 0.004925618189959698, + "min": 0.0005472909099955015, + "p50": 0.0027364545499776183, + "p95": 0.011493109109905975, + "p99": 0.011493109109905975 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 10, + "mean": 8.0, + "min": 6, + "p50": 8, + "p95": 10, + "p99": 10 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 3, + "submit_to_complete_ms": { + "max": 11.597111, + "mean": 10.493325333333333, + "min": 9.464417, + "p50": 10.418448, + "p95": 11.597111, + "p99": 11.597111 + }, + "total_tokens": { + "max": 10, + "mean": 8.0, + "min": 6, + "p50": 8, + "p95": 10, + "p99": 10 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 7, + "mean": 4.6396761133603235, + "min": 1, + "p50": 5, + "p95": 7, + "p99": 7 + }, + "kv_usage": { + "max": 0.13562863824070848, + "mean": 0.05433522417609084, + "min": 0.0004477834718145113, + "p50": 0.05353500174138015, + "p95": 0.10617443653913128, + "p99": 0.12702124483805166 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8189, + "mean": 2800.905533063428, + "min": 13, + "p50": 1881, + "p95": 7847, + "p99": 8186 + }, + "queue_waiting": { + "max": 12, + "mean": 1.2199730094466936, + "min": 0, + "p50": 0, + "p95": 7, + "p99": 10 + }, + "scheduled_requests": { + "max": 8, + "mean": 5.711201079622132, + "min": 2, + "p50": 6, + "p95": 8, + "p99": 8 + }, + "steps": 741, + "submit_to_complete_ms": { + "max": 1024.27287, + "mean": 224.85357696356274, + "min": 40.32457, + "p50": 144.763835, + "p95": 549.585886, + "p99": 785.810047 + }, + "total_tokens": { + "max": 8192, + "mean": 2805.545209176788, + "min": 17, + "p50": 1884, + "p95": 7854, + "p99": 8192 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 5, + "mean": 2.4, + "min": 1, + "p50": 2, + "p95": 5, + "p99": 5 + }, + "kv_usage": { + "max": 0.06567490919946262, + "mean": 0.033126026170456235, + "min": 0.008557639683566376, + "p50": 0.02617045624160408, + "p95": 0.06567490919946262, + "p99": 0.06567490919946262 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 13, + "mean": 7.5, + "min": 2, + "p50": 7, + "p95": 13, + "p99": 13 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 6, + "mean": 3.4, + "min": 2, + "p50": 3, + "p95": 6, + "p99": 6 + }, + "steps": 10, + "submit_to_complete_ms": { + "max": 22.147232, + "mean": 17.1428369, + "min": 13.210445, + "p50": 15.229545, + "p95": 22.147232, + "p99": 22.147232 + }, + "total_tokens": { + "max": 15, + "mean": 9.9, + "min": 3, + "p50": 9, + "p95": 15, + "p99": 15 + } + } + ], + "max_num_seqs": 8, + "phase_counts": { + "pure_decode": 27495, + "pure_prefill": 63, + "true_mixed": 751 + }, + "prefix": { + "hit_rate": 0.12292016112369349, + "hits": 307568, + "queries": 2502177 + }, + "records": 28309, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp1_mns8/opprof/opprof-v1-dp0-pid227734-1784005177936445360.jsonl", + "tensor_parallel_size": 1 + }, + { + "cell": "tp2_mns64", + "cudagraph_runtime_mode_counts": { + "FULL": 30746, + "NONE": 1555, + "PIECEWISE": 298 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 39, + "mean": 7.531678917582775, + "min": 1, + "p50": 6, + "p95": 21, + "p99": 30 + }, + "kv_usage": { + "max": 0.10309634347542496, + "mean": 0.016984430152386328, + "min": 7.673713693745476e-05, + "p50": 0.011983782885060523, + "p95": 0.05189988361534226, + "p99": 0.07517681515302665 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 39, + "mean": 7.531678917582775, + "min": 1, + "p50": 6, + "p95": 21, + "p99": 30 + }, + "steps": 30746, + "submit_to_complete_ms": { + "max": 327.486803, + "mean": 19.866698748357507, + "min": 6.465612, + "p50": 13.818435, + "p95": 29.678643, + "p99": 212.284891 + }, + "total_tokens": { + "max": 39, + "mean": 7.531678917582775, + "min": 1, + "p50": 6, + "p95": 21, + "p99": 30 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.0061261814321708075, + "mean": 0.0026679719731172026, + "min": 0.00014068475105188938, + "p50": 0.0020719026973103905, + "p95": 0.005857601452889827, + "p99": 0.0061261814321708075 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 7612, + "mean": 2967.242424242424, + "min": 129, + "p50": 2331, + "p95": 7273, + "p99": 7612 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 33, + "submit_to_complete_ms": { + "max": 285.377954, + "mean": 125.60165333333333, + "min": 58.085137, + "p50": 87.591746, + "p95": 269.63345, + "p99": 285.377954 + }, + "total_tokens": { + "max": 7612, + "mean": 2967.242424242424, + "min": 129, + "p50": 2331, + "p95": 7273, + "p99": 7612 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.0007545818465513054, + "mean": 0.00022381664940082646, + "min": 7.673713693745476e-05, + "p50": 8.952665976036389e-05, + "p95": 0.0007545818465513054, + "p99": 0.0007545818465513054 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 93, + "mean": 46.25, + "min": 10, + "p50": 32, + "p95": 93, + "p99": 93 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 8, + "submit_to_complete_ms": { + "max": 17.211201, + "mean": 12.989633, + "min": 10.521338, + "p50": 13.034104, + "p95": 17.211201, + "p99": 17.211201 + }, + "total_tokens": { + "max": 93, + "mean": 46.25, + "min": 10, + "p50": 32, + "p95": 93, + "p99": 93 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 37, + "mean": 11.961892247043364, + "min": 1, + "p50": 11, + "p95": 28, + "p99": 32 + }, + "kv_usage": { + "max": 0.10278939492767525, + "mean": 0.03137185077117546, + "min": 0.0009336351660719222, + "p50": 0.02681962935962856, + "p95": 0.07204338206141525, + "p99": 0.08896392075611659 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8191, + "mean": 3282.0696452036796, + "min": 114, + "p50": 2770, + "p95": 7855, + "p99": 8176 + }, + "queue_waiting": { + "max": 1, + "mean": 0.0032851511169513796, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 39, + "mean": 13.06898817345598, + "min": 2, + "p50": 12, + "p95": 29, + "p99": 34 + }, + "steps": 1522, + "submit_to_complete_ms": { + "max": 560.812857, + "mean": 146.33081431603154, + "min": 56.009573, + "p50": 112.614117, + "p95": 307.930472, + "p99": 403.669994 + }, + "total_tokens": { + "max": 8192, + "mean": 3294.0315374507227, + "min": 130, + "p50": 2779, + "p95": 7865, + "p99": 8192 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 34, + "mean": 10.627586206896552, + "min": 1, + "p50": 9, + "p95": 25, + "p99": 33 + }, + "kv_usage": { + "max": 0.10290450063308143, + "mean": 0.025070992877117823, + "min": 0.0008313189834887602, + "p50": 0.021217818363196916, + "p95": 0.06711941577459746, + "p99": 0.08507590581795388 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 122, + "mean": 38.53793103448276, + "min": 2, + "p50": 27, + "p95": 96, + "p99": 119 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 35, + "mean": 11.63448275862069, + "min": 2, + "p50": 10, + "p95": 26, + "p99": 34 + }, + "steps": 290, + "submit_to_complete_ms": { + "max": 310.419123, + "mean": 25.218022886206896, + "min": 10.997474, + "p50": 20.06169, + "p95": 35.895051, + "p99": 196.114372 + }, + "total_tokens": { + "max": 127, + "mean": 49.16551724137931, + "min": 5, + "p50": 41, + "p95": 104, + "p99": 124 + } + } + ], + "max_num_seqs": 64, + "phase_counts": { + "pure_decode": 30746, + "pure_prefill": 41, + "true_mixed": 1812 + }, + "prefix": { + "hit_rate": 0.1232028946400688, + "hits": 717296, + "queries": 5822071 + }, + "records": 32599, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp2_mns64/opprof/opprof-v1-dp0-pid247791-1784006495467408663.jsonl", + "tensor_parallel_size": 2 + }, + { + "cell": "tp2_mns8", + "cudagraph_runtime_mode_counts": { + "FULL": 27697, + "NONE": 1201, + "PIECEWISE": 13 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 8, + "mean": 5.417373722785861, + "min": 1, + "p50": 6, + "p95": 8, + "p99": 8 + }, + "kv_usage": { + "max": 0.03395735232188657, + "mean": 0.012474261139679475, + "min": 7.656674706180144e-05, + "p50": 0.011918890292612616, + "p95": 0.026530377856896714, + "p99": 0.03005244822173725 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 24, + "mean": 2.201646387695418, + "min": 0, + "p50": 0, + "p95": 13, + "p99": 20 + }, + "scheduled_requests": { + "max": 8, + "mean": 5.417373722785861, + "min": 1, + "p50": 6, + "p95": 8, + "p99": 8 + }, + "steps": 27697, + "submit_to_complete_ms": { + "max": 314.563834, + "mean": 16.962280882044986, + "min": 6.575488, + "p50": 13.532942, + "p95": 15.503119, + "p99": 174.775316 + }, + "total_tokens": { + "max": 8, + "mean": 5.417373722785861, + "min": 1, + "p50": 6, + "p95": 8, + "p99": 8 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.005844595025713661, + "mean": 0.0019588326123298008, + "min": 7.656674706180144e-05, + "p50": 0.001365440322601219, + "p95": 0.005704222656100488, + "p99": 0.005844595025713661 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 7273, + "mean": 2182.5666666666666, + "min": 24, + "p50": 1060, + "p95": 7073, + "p99": 7273 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 30, + "submit_to_complete_ms": { + "max": 269.442406, + "mean": 103.43858230000001, + "min": 55.279405, + "p50": 66.596689, + "p95": 260.579, + "p99": 269.442406 + }, + "total_tokens": { + "max": 7273, + "mean": 2182.5666666666666, + "min": 24, + "p50": 1060, + "p95": 7073, + "p99": 7273 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.00014037236961317312, + "mean": 0.00014037236961317312, + "min": 0.00014037236961317312, + "p50": 0.00014037236961317312, + "p95": 0.00014037236961317312, + "p99": 0.00014037236961317312 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 10, + "mean": 10.0, + "min": 10, + "p50": 10, + "p95": 10, + "p99": 10 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 1, + "submit_to_complete_ms": { + "max": 10.336689, + "mean": 10.336689, + "min": 10.336689, + "p50": 10.336689, + "p95": 10.336689, + "p99": 10.336689 + }, + "total_tokens": { + "max": 10, + "mean": 10.0, + "min": 10, + "p50": 10, + "p95": 10, + "p99": 10 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 7, + "mean": 5.772843723313407, + "min": 1, + "p50": 7, + "p95": 7, + "p99": 7 + }, + "kv_usage": { + "max": 0.03394459119737636, + "mean": 0.0159228647453862, + "min": 0.0008167119686586233, + "p50": 0.016079016882967756, + "p95": 0.02733432870104513, + "p99": 0.0316731110345444 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8191, + "mean": 2652.5926558497013, + "min": 13, + "p50": 1837, + "p95": 7634, + "p99": 8066 + }, + "queue_waiting": { + "max": 23, + "mean": 2.9812126387702818, + "min": 0, + "p50": 1, + "p95": 13, + "p99": 21 + }, + "scheduled_requests": { + "max": 8, + "mean": 6.7933390264731, + "min": 2, + "p50": 8, + "p95": 8, + "p99": 8 + }, + "steps": 1171, + "submit_to_complete_ms": { + "max": 506.291757, + "mean": 121.45648444577284, + "min": 53.060612, + "p50": 81.564408, + "p95": 292.93805, + "p99": 310.013083 + }, + "total_tokens": { + "max": 8192, + "mean": 2658.3654995730144, + "min": 17, + "p50": 1840, + "p95": 7641, + "p99": 8073 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 7, + "mean": 3.3333333333333335, + "min": 1, + "p50": 3, + "p95": 7, + "p99": 7 + }, + "kv_usage": { + "max": 0.01494327680155172, + "mean": 0.009233737010238682, + "min": 0.0034837869913096897, + "p50": 0.009315620892513032, + "p95": 0.01494327680155172, + "p99": 0.01494327680155172 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 13, + "mean": 6.916666666666667, + "min": 2, + "p50": 6, + "p95": 13, + "p99": 13 + }, + "queue_waiting": { + "max": 11, + "mean": 0.9166666666666666, + "min": 0, + "p50": 0, + "p95": 11, + "p99": 11 + }, + "scheduled_requests": { + "max": 8, + "mean": 4.333333333333333, + "min": 2, + "p50": 4, + "p95": 8, + "p99": 8 + }, + "steps": 12, + "submit_to_complete_ms": { + "max": 15.924106, + "mean": 13.055620500000002, + "min": 10.64629, + "p50": 12.616157, + "p95": 15.924106, + "p99": 15.924106 + }, + "total_tokens": { + "max": 16, + "mean": 10.25, + "min": 5, + "p50": 9, + "p95": 16, + "p99": 16 + } + } + ], + "max_num_seqs": 8, + "phase_counts": { + "pure_decode": 27697, + "pure_prefill": 31, + "true_mixed": 1183 + }, + "prefix": { + "hit_rate": 0.12439998012356517, + "hits": 450624, + "queries": 3622380 + }, + "records": 28911, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp2_mns8/opprof/opprof-v1-dp0-pid241594-1784006096655191299.jsonl", + "tensor_parallel_size": 2 + }, + { + "cell": "tp4_mns16", + "cudagraph_runtime_mode_counts": { + "FULL": 17418, + "NONE": 1752, + "PIECEWISE": 53 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 16, + "mean": 13.292530286501695, + "min": 1, + "p50": 16, + "p95": 16, + "p99": 16 + }, + "kv_usage": { + "max": 0.02384243790712981, + "mean": 0.011910732119220589, + "min": 0.00021424527002555926, + "p50": 0.012732290332947316, + "p95": 0.01879237082795593, + "p99": 0.02112866448680606 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 48, + "mean": 11.14313601653557, + "min": 0, + "p50": 5, + "p95": 48, + "p99": 48 + }, + "scheduled_requests": { + "max": 16, + "mean": 13.292530286501695, + "min": 1, + "p50": 16, + "p95": 16, + "p99": 16 + }, + "steps": 17417, + "submit_to_complete_ms": { + "max": 185.583096, + "mean": 18.308918529310446, + "min": 6.71116, + "p50": 12.899641, + "p95": 64.79132, + "p99": 141.517546 + }, + "total_tokens": { + "max": 16, + "mean": 13.292530286501695, + "min": 1, + "p50": 16, + "p95": 16, + "p99": 16 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.0014436050337436335, + "mean": 0.0005439024265827402, + "min": 0.0003213679050383389, + "p50": 0.0003570754500425988, + "p95": 0.0014436050337436335, + "p99": 0.0014436050337436335 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 4459, + "mean": 1152.125, + "min": 201, + "p50": 699, + "p95": 4459, + "p99": 4459 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 8, + "submit_to_complete_ms": { + "max": 169.859824, + "mean": 79.969349875, + "min": 59.109261, + "p50": 63.446202, + "p95": 169.859824, + "p99": 169.859824 + }, + "total_tokens": { + "max": 4459, + "mean": 1152.125, + "min": 201, + "p50": 699, + "p95": 4459, + "p99": 4459 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.0014334028780281782, + "mean": 0.0006869451515105393, + "min": 0.0002805592821762959, + "p50": 0.00034687329432714353, + "p95": 0.0014334028780281782, + "p99": 0.0014334028780281782 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 23, + "mean": 13.0, + "min": 3, + "p50": 13, + "p95": 23, + "p99": 23 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 3, + "submit_to_complete_ms": { + "max": 13.064208, + "mean": 12.249279000000001, + "min": 10.993105, + "p50": 12.690524, + "p95": 13.064208, + "p99": 13.064208 + }, + "total_tokens": { + "max": 23, + "mean": 13.0, + "min": 3, + "p50": 13, + "p95": 23, + "p99": 23 + } + }, + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 15, + "mean": 15.0, + "min": 15, + "p50": 15, + "p95": 15, + "p99": 15 + }, + "kv_usage": { + "max": 0.014119783510255668, + "mean": 0.014119783510255668, + "min": 0.014119783510255668, + "p50": 0.014119783510255668, + "p95": 0.014119783510255668, + "p99": 0.014119783510255668 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "queue_waiting": { + "max": 3, + "mean": 3.0, + "min": 3, + "p50": 3, + "p95": 3, + "p99": 3 + }, + "scheduled_requests": { + "max": 16, + "mean": 16.0, + "min": 16, + "p50": 16, + "p95": 16, + "p99": 16 + }, + "steps": 1, + "submit_to_complete_ms": { + "max": 31.93902, + "mean": 31.93902, + "min": 31.93902, + "p50": 31.93902, + "p95": 31.93902, + "p99": 31.93902 + }, + "total_tokens": { + "max": 16, + "mean": 16.0, + "min": 16, + "p50": 16, + "p95": 16, + "p99": 16 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 15, + "mean": 13.41112385321101, + "min": 1, + "p50": 15, + "p95": 15, + "p99": 15 + }, + "kv_usage": { + "max": 0.02384243790712981, + "mean": 0.013269559018371043, + "min": 0.0002397506593143639, + "p50": 0.013645383269484834, + "p95": 0.019067829032274553, + "p99": 0.021705086284732 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8191, + "mean": 2915.523509174312, + "min": 18, + "p50": 2078, + "p95": 7887, + "p99": 8180 + }, + "queue_waiting": { + "max": 48, + "mean": 13.461582568807339, + "min": 0, + "p50": 10, + "p95": 47, + "p99": 47 + }, + "scheduled_requests": { + "max": 16, + "mean": 14.555045871559633, + "min": 2, + "p50": 16, + "p95": 16, + "p99": 16 + }, + "steps": 1744, + "submit_to_complete_ms": { + "max": 379.907708, + "mean": 108.41418502752293, + "min": 56.308147, + "p50": 83.525673, + "p95": 217.228507, + "p99": 297.197569 + }, + "total_tokens": { + "max": 8192, + "mean": 2928.934633027523, + "min": 33, + "p50": 2093, + "p95": 7900, + "p99": 8192 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 15, + "mean": 9.54, + "min": 1, + "p50": 9, + "p95": 15, + "p99": 15 + }, + "kv_usage": { + "max": 0.022806919102006273, + "mean": 0.009802027168340671, + "min": 0.0015405255130409579, + "p50": 0.008386171998143177, + "p95": 0.018516912623637416, + "p99": 0.022806919102006273 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 25, + "mean": 13.94, + "min": 2, + "p50": 13, + "p95": 23, + "p99": 25 + }, + "queue_waiting": { + "max": 47, + "mean": 5.82, + "min": 0, + "p50": 0, + "p95": 47, + "p99": 47 + }, + "scheduled_requests": { + "max": 16, + "mean": 10.54, + "min": 2, + "p50": 10, + "p95": 16, + "p99": 16 + }, + "steps": 50, + "submit_to_complete_ms": { + "max": 91.408766, + "mean": 24.5079404, + "min": 11.206266, + "p50": 14.216713, + "p95": 83.023614, + "p99": 91.408766 + }, + "total_tokens": { + "max": 32, + "mean": 23.48, + "min": 4, + "p50": 25, + "p95": 32, + "p99": 32 + } + } + ], + "max_num_seqs": 16, + "phase_counts": { + "pure_decode": 17417, + "pure_prefill": 11, + "true_mixed": 1795 + }, + "prefix": { + "hit_rate": 0.13543755577854244, + "hits": 798096, + "queries": 5892723 + }, + "records": 19223, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp4_mns16/opprof/opprof-v1-dp0-pid256090-1784007028237607411.jsonl", + "tensor_parallel_size": 4 + }, + { + "cell": "tp4_mns64", + "cudagraph_runtime_mode_counts": { + "FULL": 21812, + "NONE": 2809, + "PIECEWISE": 354 + }, + "dropped_records_before_max": 0, + "groups": [ + { + "cudagraph_runtime_mode": "FULL", + "decode_tokens": { + "max": 64, + "mean": 20.639510361269025, + "min": 1, + "p50": 15, + "p95": 62, + "p99": 64 + }, + "kv_usage": { + "max": 0.07507819047813735, + "mean": 0.01804654466537005, + "min": 4.088391013712389e-05, + "p50": 0.013655225985813257, + "p95": 0.05230074204296897, + "p99": 0.062281526605204474 + }, + "phase": "pure_decode", + "prefill_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 64, + "mean": 20.639510361269025, + "min": 1, + "p50": 15, + "p95": 62, + "p99": 64 + }, + "steps": 21812, + "submit_to_complete_ms": { + "max": 198.715632, + "mean": 21.35240870649184, + "min": 6.440609, + "p50": 13.906458, + "p95": 68.920454, + "p99": 155.977421 + }, + "total_tokens": { + "max": 64, + "mean": 20.639510361269025, + "min": 1, + "p50": 15, + "p95": 62, + "p99": 64 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.0025552443835727967, + "mean": 0.0008350538645516159, + "min": 0.00032196079233015595, + "p50": 0.00038839714630312105, + "p95": 0.0025552443835727967, + "p99": 0.0025552443835727967 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 7950, + "mean": 2158.0, + "min": 201, + "p50": 959, + "p95": 7950, + "p99": 7950 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 10, + "submit_to_complete_ms": { + "max": 173.889993, + "mean": 88.2164954, + "min": 60.32347, + "p50": 64.173068, + "p95": 173.889993, + "p99": 173.889993 + }, + "total_tokens": { + "max": 7950, + "mean": 2158.0, + "min": 201, + "p50": 959, + "p95": 7950, + "p99": 7950 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "kv_usage": { + "max": 0.00034751323616588614, + "mean": 0.0003142950591794591, + "min": 0.00028107688219303206, + "p50": 0.00028107688219303206, + "p95": 0.00034751323616588614, + "p99": 0.00034751323616588614 + }, + "phase": "pure_prefill", + "prefill_tokens": { + "max": 23, + "mean": 13.0, + "min": 3, + "p50": 3, + "p95": 23, + "p99": 23 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 1, + "mean": 1.0, + "min": 1, + "p50": 1, + "p95": 1, + "p99": 1 + }, + "steps": 2, + "submit_to_complete_ms": { + "max": 14.426163, + "mean": 12.918343, + "min": 11.410523, + "p50": 11.410523, + "p95": 14.426163, + "p99": 14.426163 + }, + "total_tokens": { + "max": 23, + "mean": 13.0, + "min": 3, + "p50": 3, + "p95": 23, + "p99": 23 + } + }, + { + "cudagraph_runtime_mode": "NONE", + "decode_tokens": { + "max": 63, + "mean": 31.312254376563057, + "min": 1, + "p50": 27, + "p95": 62, + "p99": 63 + }, + "kv_usage": { + "max": 0.07563012326498908, + "mean": 0.02885461746554652, + "min": 0.00024019297205579715, + "p50": 0.025511559925591287, + "p95": 0.05791716919806211, + "p99": 0.06878206831701383 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 8191, + "mean": 3841.2243658449447, + "min": 73, + "p50": 3432, + "p95": 8154, + "p99": 8181 + }, + "queue_waiting": { + "max": 11, + "mean": 0.1907824222936763, + "min": 0, + "p50": 0, + "p95": 1, + "p99": 5 + }, + "scheduled_requests": { + "max": 64, + "mean": 32.843515541264736, + "min": 2, + "p50": 28, + "p95": 63, + "p99": 64 + }, + "steps": 2799, + "submit_to_complete_ms": { + "max": 373.577501, + "mean": 134.91752637549124, + "min": 55.209363, + "p50": 114.740321, + "p95": 314.91308, + "p99": 343.860853 + }, + "total_tokens": { + "max": 8192, + "mean": 3872.5366202215077, + "min": 129, + "p50": 3456, + "p95": 8192, + "p99": 8192 + } + }, + { + "cudagraph_runtime_mode": "PIECEWISE", + "decode_tokens": { + "max": 63, + "mean": 25.65340909090909, + "min": 1, + "p50": 20, + "p95": 60, + "p99": 63 + }, + "kv_usage": { + "max": 0.07241562583045447, + "mean": 0.022061428307276035, + "min": 0.00047016496657736884, + "p50": 0.017590302336515484, + "p95": 0.05158016312680147, + "p99": 0.05824424047915944 + }, + "phase": "true_mixed", + "prefill_tokens": { + "max": 114, + "mean": 35.32670454545455, + "min": 2, + "p50": 25, + "p95": 87, + "p99": 106 + }, + "queue_waiting": { + "max": 0, + "mean": 0.0, + "min": 0, + "p50": 0, + "p95": 0, + "p99": 0 + }, + "scheduled_requests": { + "max": 64, + "mean": 26.678977272727273, + "min": 2, + "p50": 21, + "p95": 61, + "p99": 64 + }, + "steps": 352, + "submit_to_complete_ms": { + "max": 196.960993, + "mean": 27.942972144886365, + "min": 11.037722, + "p50": 17.402246, + "p95": 97.041082, + "p99": 187.504426 + }, + "total_tokens": { + "max": 128, + "mean": 60.98011363636363, + "min": 4, + "p50": 60, + "p95": 114, + "p99": 126 + } + } + ], + "max_num_seqs": 64, + "phase_counts": { + "pure_decode": 21812, + "pure_prefill": 12, + "true_mixed": 3151 + }, + "prefix": { + "hit_rate": 0.13386788146702433, + "hits": 1667008, + "queries": 12452636 + }, + "records": 24975, + "stream": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/cells/tp4_mns64/opprof/opprof-v1-dp0-pid262293-1784007376383569515.jsonl", + "tensor_parallel_size": 4 + } + ], + "contains_prompt_text": false, + "measurement_semantics": "complete_mono_ns - submit_mono_ns from the existing vLLM opprof record; this includes the instrumented submit-to-completion interval and is not claimed to be a single-kernel CUDA-event duration", + "schema_version": "qwen30_vllm020_opprof_summary.v1" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-routing-mismatch.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-routing-mismatch.json new file mode 100644 index 0000000..e2a4fb0 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-routing-mismatch.json @@ -0,0 +1,1853 @@ +{ + "actual_prefill_vs_decode_pearson": { + "max": 0.9436904928897797, + "median": 0.6667880423793499, + "min": 0.48720283009127985 + }, + "frontier_contract": { + "allocation": "per-layer fixed Uniform(0.1, 1.0), normalized once and reused for every batch and phase", + "expert_count": 128, + "layer_count": 48, + "mode": "simulation", + "seed": 42 + }, + "frontier_prefill_vs_decode_pearson": 1.0, + "phase_summary": { + "decode": { + "actual": { + "expert_utilization": { + "max": 1.0, + "median": 0.96875, + "min": 0.921875 + }, + "load_cv": { + "max": 1.5122257286912568, + "median": 1.106820300518243, + "min": 0.7532423759874015 + }, + "load_gini": { + "max": 0.647222736853249, + "median": 0.5427848362430705, + "min": 0.3623759672389899 + }, + "max_load_ratio": { + "max": 8.630736064059132, + "median": 5.642131198028951, + "min": 3.392670157068063 + } + }, + "actual_vs_frontier_pearson": { + "max": 0.17269736915133127, + "median": -0.024141635934415796, + "min": -0.20201754939516248 + }, + "frontier_simulation": { + "expert_utilization": { + "max": 1.0, + "median": 1.0, + "min": 1.0 + }, + "load_cv": { + "max": 0.5319387026702922, + "median": 0.4852775824549782, + "min": 0.4143052194562559 + }, + "load_gini": { + "max": 0.3057696286957192, + "median": 0.2796991393497844, + "min": 0.2373612300777641 + }, + "max_load_ratio": { + "max": 2.0104712041884816, + "median": 1.8392362180474282, + "min": 1.6482907299045273 + } + }, + "token_count": 6494 + }, + "prefill": { + "actual": { + "expert_utilization": { + "max": 1.0, + "median": 0.984375, + "min": 0.9453125 + }, + "load_cv": { + "max": 1.636444303545179, + "median": 1.0764214888570005, + "min": 0.802893981590087 + }, + "load_gini": { + "max": 0.6967902002938451, + "median": 0.5262265677660404, + "min": 0.42613218222079463 + }, + "max_load_ratio": { + "max": 10.381092808695286, + "median": 6.121413826515567, + "min": 3.8360365673842525 + } + }, + "actual_vs_frontier_pearson": { + "max": 0.1797880356509229, + "median": -0.01729201163458405, + "min": -0.20008636491506096 + }, + "frontier_simulation": { + "expert_utilization": { + "max": 1.0, + "median": 1.0, + "min": 1.0 + }, + "load_cv": { + "max": 0.5319950183538887, + "median": 0.48529878894960354, + "min": 0.41421985083210333 + }, + "load_gini": { + "max": 0.3057985522338542, + "median": 0.2797113699735645, + "min": 0.23730966555377675 + }, + "max_load_ratio": { + "max": 2.0093524876774658, + "median": 1.838817036693769, + "min": 1.649408097063656 + } + }, + "token_count": 23737 + } + }, + "rows": [ + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 1.0209765758908849, + "load_gini": 0.46969517196254795, + "max_load_ratio": 5.4827484517841345 + }, + "actual_vs_frontier_pearson": -0.09214641606187762, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5013611454809463, + "load_gini": 0.28879565986118716, + "max_load_ratio": 1.8509499936807516 + }, + "layer": 0, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 0.9656310061810012, + "load_gini": 0.4405106117295783, + "max_load_ratio": 6.154105405063825 + }, + "actual_vs_frontier_pearson": -0.14455565868271067, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4633214169333404, + "load_gini": 0.2672016031143363, + "max_load_ratio": 1.7579306567805535 + }, + "layer": 1, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 1.4100541859296152, + "load_gini": 0.5967419337426803, + "max_load_ratio": 8.47284829590934 + }, + "actual_vs_frontier_pearson": 0.10746117235272135, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4915202242607809, + "load_gini": 0.28343845117853994, + "max_load_ratio": 1.8556683658423558 + }, + "layer": 2, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 1.1750115312430194, + "load_gini": 0.49178514160382525, + "max_load_ratio": 7.965286261953912 + }, + "actual_vs_frontier_pearson": 0.07854011611932744, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5153981688841472, + "load_gini": 0.29577933447781946, + "max_load_ratio": 1.899481821628681 + }, + "layer": 3, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 1.2753111721870192, + "load_gini": 0.5858468253675697, + "max_load_ratio": 7.488056620465939 + }, + "actual_vs_frontier_pearson": 0.047570378008613835, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5319950183538887, + "load_gini": 0.3057985522338542, + "max_load_ratio": 2.0093524876774658 + }, + "layer": 4, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.2689973799039043, + "load_gini": 0.5545393103330243, + "max_load_ratio": 9.569532796899356 + }, + "actual_vs_frontier_pearson": -0.06358262657671977, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.46397269621367815, + "load_gini": 0.2675475161404137, + "max_load_ratio": 1.7828706239204617 + }, + "layer": 5, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.3617880106985483, + "load_gini": 0.5929317066973501, + "max_load_ratio": 10.381092808695286 + }, + "actual_vs_frontier_pearson": -0.01786861982721008, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5065069021962776, + "load_gini": 0.29121334835910173, + "max_load_ratio": 1.9075704596200025 + }, + "layer": 6, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.4508622399317908, + "load_gini": 0.633230738799132, + "max_load_ratio": 8.563845473311707 + }, + "actual_vs_frontier_pearson": -0.018533706086765106, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4174667228740326, + "load_gini": 0.23938283981758435, + "max_load_ratio": 1.649408097063656 + }, + "layer": 7, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0194348783142033, + "load_gini": 0.5101870727924758, + "max_load_ratio": 4.746008341407928 + }, + "actual_vs_frontier_pearson": -0.07019410060413582, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.45521406973427264, + "load_gini": 0.2624579704153853, + "max_load_ratio": 1.7660192947718751 + }, + "layer": 8, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0499203578258598, + "load_gini": 0.5348460473101067, + "max_load_ratio": 4.87475249610313 + }, + "actual_vs_frontier_pearson": 0.055984358629187964, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4981686541927147, + "load_gini": 0.28707712050280154, + "max_load_ratio": 1.935880692589628 + }, + "layer": 9, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.9182859260805778, + "load_gini": 0.4667757285566836, + "max_load_ratio": 6.065130387159287 + }, + "actual_vs_frontier_pearson": 0.04566572558971987, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4410838523022549, + "load_gini": 0.2539255039600623, + "max_load_ratio": 1.7262501579812108 + }, + "layer": 10, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.8938335625889317, + "load_gini": 0.46192282552028474, + "max_load_ratio": 4.6934321944643385 + }, + "actual_vs_frontier_pearson": 0.05102058572481217, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4344763177468921, + "load_gini": 0.250220186312508, + "max_load_ratio": 1.7592787631124405 + }, + "layer": 11, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.9842127914363633, + "load_gini": 0.48561951344420096, + "max_load_ratio": 6.4446223195854575 + }, + "actual_vs_frontier_pearson": -0.117474412760527, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4881943608268652, + "load_gini": 0.28113027986371497, + "max_load_ratio": 1.86038673800396 + }, + "layer": 12, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 1.073206640427139, + "load_gini": 0.5187799341481654, + "max_load_ratio": 6.354973248514977 + }, + "actual_vs_frontier_pearson": -0.020599525349797518, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4551370538191022, + "load_gini": 0.2622560506803724, + "max_load_ratio": 1.7660192947718751 + }, + "layer": 13, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0501051337772571, + "load_gini": 0.5237924640329443, + "max_load_ratio": 4.951594557020685 + }, + "actual_vs_frontier_pearson": -0.1404264118915223, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4843196647309149, + "load_gini": 0.2793572699003666, + "max_load_ratio": 1.8071365378944264 + }, + "layer": 14, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 1.1589582379096603, + "load_gini": 0.5428402257024898, + "max_load_ratio": 6.021990984538905 + }, + "actual_vs_frontier_pearson": 0.010817413166755824, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4523916462775862, + "load_gini": 0.2604694642067238, + "max_load_ratio": 1.8084846442263134 + }, + "layer": 15, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.1496287137851553, + "load_gini": 0.5617946797194253, + "max_load_ratio": 6.50933142351603 + }, + "actual_vs_frontier_pearson": 0.1797880356509229, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48641030098466176, + "load_gini": 0.2805311032091249, + "max_load_ratio": 1.8334246113662216 + }, + "layer": 16, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.2394125145202786, + "load_gini": 0.5729841103814719, + "max_load_ratio": 7.101150103214391 + }, + "actual_vs_frontier_pearson": -0.20008636491506096, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.443419246367453, + "load_gini": 0.25561211794877203, + "max_load_ratio": 1.7754560390950835 + }, + "layer": 17, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.2168439400279965, + "load_gini": 0.575970613520032, + "max_load_ratio": 8.145258457260816 + }, + "actual_vs_frontier_pearson": -0.06530349438725558, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4516257974939683, + "load_gini": 0.25957785774845177, + "max_load_ratio": 1.8010700594009352 + }, + "layer": 18, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.2427942152510565, + "load_gini": 0.5980218283955427, + "max_load_ratio": 6.7688418924042635 + }, + "actual_vs_frontier_pearson": -0.04180797903095555, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4681222179773095, + "load_gini": 0.26999935161878086, + "max_load_ratio": 1.8111808568900871 + }, + "layer": 19, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 0.9706255724368628, + "load_gini": 0.5072024621371698, + "max_load_ratio": 4.138012385726924 + }, + "actual_vs_frontier_pearson": 0.007542212699855623, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4876159778421919, + "load_gini": 0.2809273727461348, + "max_load_ratio": 1.7734338795972533 + }, + "layer": 20, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.9454143633213303, + "load_gini": 0.5064155182310317, + "max_load_ratio": 3.8980494586510512 + }, + "actual_vs_frontier_pearson": -0.005313631721923769, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4407612899555989, + "load_gini": 0.25342531240784427, + "max_load_ratio": 1.7006361376753591 + }, + "layer": 21, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.8037705358927761, + "load_gini": 0.42613218222079463, + "max_load_ratio": 5.447697687155074 + }, + "actual_vs_frontier_pearson": 0.006893096788933473, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5096984088007344, + "load_gini": 0.292796698587648, + "max_load_ratio": 1.9850865737035008 + }, + "layer": 22, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.802893981590087, + "load_gini": 0.43984914769136796, + "max_load_ratio": 3.953321818258415 + }, + "actual_vs_frontier_pearson": 0.055518113087740455, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5020829253028674, + "load_gini": 0.28819319193137294, + "max_load_ratio": 1.9695833508868013 + }, + "layer": 23, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.8487498379405487, + "load_gini": 0.44496173234296665, + "max_load_ratio": 4.682647343809243 + }, + "actual_vs_frontier_pearson": -0.11069406175185324, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4937827979946325, + "load_gini": 0.28478219657075443, + "max_load_ratio": 1.8812823861482075 + }, + "layer": 24, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.8842495554569121, + "load_gini": 0.4609973189272023, + "max_load_ratio": 5.068205754728904 + }, + "actual_vs_frontier_pearson": 0.04041373985743766, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5094957270929035, + "load_gini": 0.29290695630766317, + "max_load_ratio": 1.9264439482664195 + }, + "layer": 25, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.910991140491345, + "load_gini": 0.47597056415090355, + "max_load_ratio": 4.189240426338627 + }, + "actual_vs_frontier_pearson": -0.001853343235349625, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5141430671669248, + "load_gini": 0.2956995210536293, + "max_load_ratio": 1.870497535493112 + }, + "layer": 26, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.0416423824930179, + "load_gini": 0.5048694416944011, + "max_load_ratio": 5.7442810801701985 + }, + "actual_vs_frontier_pearson": 0.021178350576558223, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4563109042451374, + "load_gini": 0.2625302961884399, + "max_load_ratio": 1.8846526519779248 + }, + "layer": 27, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.953125, + "load_cv": 1.0513699281689028, + "load_gini": 0.5286606714991364, + "max_load_ratio": 5.484096558116021 + }, + "actual_vs_frontier_pearson": -0.195005081355267, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4931389294946141, + "load_gini": 0.28378707950667725, + "max_load_ratio": 1.7983738467371615 + }, + "layer": 28, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.953125, + "load_cv": 1.079636337286862, + "load_gini": 0.5304158262943928, + "max_load_ratio": 6.738509499936807 + }, + "actual_vs_frontier_pearson": -0.016715403441958022, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4500354635096657, + "load_gini": 0.2587639253854741, + "max_load_ratio": 1.7316425833087585 + }, + "layer": 29, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.1836591453389935, + "load_gini": 0.5595659927012682, + "max_load_ratio": 7.912710115010322 + }, + "actual_vs_frontier_pearson": 0.009080640196552386, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4501844468319078, + "load_gini": 0.25952453908981754, + "max_load_ratio": 1.845557568353204 + }, + "layer": 30, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.1837797460931923, + "load_gini": 0.5779810067089355, + "max_load_ratio": 6.173652946876185 + }, + "actual_vs_frontier_pearson": -0.11789359921128652, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5029280507742879, + "load_gini": 0.2893184789305725, + "max_load_ratio": 1.92846610776425 + }, + "layer": 31, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 0.9485322853469729, + "load_gini": 0.4766967017472723, + "max_load_ratio": 5.981547794582298 + }, + "actual_vs_frontier_pearson": -0.11341887598082027, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4862779131682921, + "load_gini": 0.2800654700467624, + "max_load_ratio": 1.8307283987024476 + }, + "layer": 32, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.9147261414047101, + "load_gini": 0.48626378056936437, + "max_load_ratio": 4.4292033534145006 + }, + "actual_vs_frontier_pearson": -0.0734554681462204, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4826868638135814, + "load_gini": 0.278262015787589, + "max_load_ratio": 1.8347727176981083 + }, + "layer": 33, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.822096011945377, + "load_gini": 0.4335047210051819, + "max_load_ratio": 5.387032902220162 + }, + "actual_vs_frontier_pearson": 0.041109767501958036, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5205338188216663, + "load_gini": 0.2990128478219658, + "max_load_ratio": 1.84286135568943 + }, + "layer": 34, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.8350377321986099, + "load_gini": 0.4458607441704934, + "max_load_ratio": 3.8360365673842525 + }, + "actual_vs_frontier_pearson": -0.002498639457316309, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4931172415832444, + "load_gini": 0.28332819345852456, + "max_load_ratio": 1.9008299279605678 + }, + "layer": 35, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.936559272961729, + "load_gini": 0.4615142959830645, + "max_load_ratio": 5.575767788684333 + }, + "actual_vs_frontier_pearson": -0.04463576992397696, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.47788893428675927, + "load_gini": 0.27538099801996885, + "max_load_ratio": 1.7997219530690483 + }, + "layer": 36, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0424990294361898, + "load_gini": 0.5039767655716814, + "max_load_ratio": 6.859839069806631 + }, + "actual_vs_frontier_pearson": -0.05047148070882558, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.41421985083210333, + "load_gini": 0.23730966555377675, + "max_load_ratio": 1.6965918186796984 + }, + "layer": 37, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 1.081067450691553, + "load_gini": 0.515087617037747, + "max_load_ratio": 6.495850360197161 + }, + "actual_vs_frontier_pearson": -0.03189531872213542, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48198382691720465, + "load_gini": 0.27773105081202343, + "max_load_ratio": 1.8037662720647092 + }, + "layer": 38, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.140261877881891, + "load_gini": 0.5301265232021739, + "max_load_ratio": 6.088722247967309 + }, + "actual_vs_frontier_pearson": -0.1047856089108959, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.49823287732736005, + "load_gini": 0.28601938692758133, + "max_load_ratio": 1.9102666722837764 + }, + "layer": 39, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.16210598515629, + "load_gini": 0.5750575314909214, + "max_load_ratio": 5.264355226018452 + }, + "actual_vs_frontier_pearson": 0.07101910025349224, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5231673221525147, + "load_gini": 0.30064416838690655, + "max_load_ratio": 1.9163331507772676 + }, + "layer": 40, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.335821482050506, + "load_gini": 0.6172422174506045, + "max_load_ratio": 8.129755234444117 + }, + "actual_vs_frontier_pearson": 0.04271119117719951, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4592680576699305, + "load_gini": 0.26492856616253113, + "max_load_ratio": 1.8145511227198046 + }, + "layer": 41, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.5340944192835082, + "load_gini": 0.6802103190693853, + "max_load_ratio": 9.211610565783376 + }, + "actual_vs_frontier_pearson": 0.012121305370270835, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.493325400417425, + "load_gini": 0.28234122230062764, + "max_load_ratio": 1.9055483001221722 + }, + "layer": 42, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.5076855954950434, + "load_gini": 0.6597892728124868, + "max_load_ratio": 7.8324977882630495 + }, + "actual_vs_frontier_pearson": 0.1787454639272607, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4651216460072469, + "load_gini": 0.267678755739984, + "max_load_ratio": 1.870497535493112 + }, + "layer": 43, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 1.4676145461991041, + "load_gini": 0.666736825683532, + "max_load_ratio": 6.086700088469478 + }, + "actual_vs_frontier_pearson": -0.1318454052741815, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.47511235321400247, + "load_gini": 0.2737368414816532, + "max_load_ratio": 1.84286135568943 + }, + "layer": 44, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.5817687300685335, + "load_gini": 0.6967902002938451, + "max_load_ratio": 6.617853983232927 + }, + "actual_vs_frontier_pearson": -0.042931525341602716, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.47921880192242294, + "load_gini": 0.2763617157549396, + "max_load_ratio": 1.7754560390950835 + }, + "layer": 45, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.5934457162349587, + "load_gini": 0.6958883908823776, + "max_load_ratio": 7.270337447866201 + }, + "actual_vs_frontier_pearson": 0.02211603536084719, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5297987184237168, + "load_gini": 0.30463451046888834, + "max_load_ratio": 1.8799342798163206 + }, + "layer": 46, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.636444303545179, + "load_gini": 0.6918075387317268, + "max_load_ratio": 8.385895437502633 + }, + "actual_vs_frontier_pearson": -0.01870726023327083, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4952385440074709, + "load_gini": 0.28391033776382857, + "max_load_ratio": 1.9170072039432111 + }, + "layer": 47, + "phase": "prefill", + "total_routed_tokens": 189896 + }, + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 0.7564860612847418, + "load_gini": 0.3623759672389899, + "max_load_ratio": 4.043116723129042 + }, + "actual_vs_frontier_pearson": -0.11076278775511662, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5013193007157182, + "load_gini": 0.288770162842624, + "max_load_ratio": 1.8503233754234678 + }, + "layer": 0, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 1.0, + "load_cv": 0.7532423759874015, + "load_gini": 0.36718870303356943, + "max_load_ratio": 4.479211579919926 + }, + "actual_vs_frontier_pearson": -0.20201754939516248, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4633429809491391, + "load_gini": 0.2672133050700647, + "max_load_ratio": 1.7566984909146905 + }, + "layer": 1, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.9982932903554329, + "load_gini": 0.48608902207807203, + "max_load_ratio": 5.430243301509085 + }, + "actual_vs_frontier_pearson": 0.12507453392144424, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4914341503043837, + "load_gini": 0.2833871891361257, + "max_load_ratio": 1.8552510009239298 + }, + "layer": 2, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.887995113903465, + "load_gini": 0.4407307466507546, + "max_load_ratio": 4.843855866954112 + }, + "actual_vs_frontier_pearson": 0.09916207755901395, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5153762623410879, + "load_gini": 0.2957582840891593, + "max_load_ratio": 1.8995996304280875 + }, + "layer": 3, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0374295217881528, + "load_gini": 0.5126796730635972, + "max_load_ratio": 5.198644902987373 + }, + "actual_vs_frontier_pearson": 0.1087630450097359, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5319387026702922, + "load_gini": 0.3057696286957192, + "max_load_ratio": 2.0104712041884816 + }, + "layer": 4, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.9864025871023987, + "load_gini": 0.5000595501616878, + "max_load_ratio": 6.716353557129658 + }, + "actual_vs_frontier_pearson": -0.060788024591760324, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.46389740539367647, + "load_gini": 0.26750173236833996, + "max_load_ratio": 1.7838004311672313 + }, + "layer": 5, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.2953358972433757, + "load_gini": 0.629114675565907, + "max_load_ratio": 5.74314752078842 + }, + "actual_vs_frontier_pearson": -0.07165433962410654, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5065225414304516, + "load_gini": 0.2912228475323375, + "max_load_ratio": 1.9069910686787803 + }, + "layer": 6, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.2699902455381382, + "load_gini": 0.6262021312942716, + "max_load_ratio": 6.66954111487527 + }, + "actual_vs_frontier_pearson": 0.01205569571775735, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.41747315017256753, + "load_gini": 0.2393889431590699, + "max_load_ratio": 1.6482907299045273 + }, + "layer": 7, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.0541527297772746, + "load_gini": 0.5379860857137357, + "max_load_ratio": 5.171542962734832 + }, + "actual_vs_frontier_pearson": -0.04958458083380716, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4553076424255581, + "load_gini": 0.2625118498806591, + "max_load_ratio": 1.7665537419156143 + }, + "layer": 8, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.192957622384299, + "load_gini": 0.5506200435016939, + "max_load_ratio": 8.320295657530028 + }, + "actual_vs_frontier_pearson": -0.05509740986705497, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4981311718832551, + "load_gini": 0.28705553924776717, + "max_load_ratio": 1.9365568216815523 + }, + "layer": 9, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.8076732412246237, + "load_gini": 0.4410110534724361, + "max_load_ratio": 3.392670157068063 + }, + "actual_vs_frontier_pearson": -0.09000721184282558, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4410207617483597, + "load_gini": 0.2538882044964583, + "max_load_ratio": 1.7271327379119188 + }, + "layer": 10, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.946297012928892, + "load_gini": 0.4790344332268248, + "max_load_ratio": 4.728056667693255 + }, + "actual_vs_frontier_pearson": 0.034097483022671266, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4344893792424141, + "load_gini": 0.25022677182784103, + "max_load_ratio": 1.7591623036649215 + }, + "layer": 11, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.9479122820728094, + "load_gini": 0.48800966517169697, + "max_load_ratio": 5.097628580227902 + }, + "actual_vs_frontier_pearson": -0.02153501737741947, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48821731504828314, + "load_gini": 0.28114623835463504, + "max_load_ratio": 1.8601786264243918 + }, + "layer": 12, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.9047451307311554, + "load_gini": 0.4740120688327687, + "max_load_ratio": 4.555589775177086 + }, + "actual_vs_frontier_pearson": -0.05894159287248561, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4550660315489719, + "load_gini": 0.26222101651524476, + "max_load_ratio": 1.7665537419156143 + }, + "layer": 13, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.1102662993215426, + "load_gini": 0.550685007314444, + "max_load_ratio": 5.5214043732676314 + }, + "actual_vs_frontier_pearson": -0.07473125180903177, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4843214689622731, + "load_gini": 0.2793576281952572, + "max_load_ratio": 1.80597474591931 + }, + "layer": 14, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.0424321211996845, + "load_gini": 0.5389009927433015, + "max_load_ratio": 4.885740683708038 + }, + "actual_vs_frontier_pearson": -0.122334501258848, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.45239738453493195, + "load_gini": 0.26047060267169697, + "max_load_ratio": 1.808438558669541 + }, + "layer": 15, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.1000158272634826, + "load_gini": 0.5458891139898367, + "max_load_ratio": 5.321835540498922 + }, + "actual_vs_frontier_pearson": 0.06285259077675373, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48638660674407463, + "load_gini": 0.2805170517978133, + "max_load_ratio": 1.833076686171851 + }, + "layer": 16, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.1000359607690584, + "load_gini": 0.5519229288574068, + "max_load_ratio": 5.201108715737604 + }, + "actual_vs_frontier_pearson": -0.14790969026732273, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.44351655698688397, + "load_gini": 0.25566538583692644, + "max_load_ratio": 1.7764089929165383 + }, + "layer": 17, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.921875, + "load_cv": 1.275279013013624, + "load_gini": 0.6363662611641516, + "max_load_ratio": 5.637203572528488 + }, + "actual_vs_frontier_pearson": -0.017099613299867974, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4514828672543027, + "load_gini": 0.259494942446874, + "max_load_ratio": 1.801047120418848 + }, + "layer": 18, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.921875, + "load_cv": 1.2773056983164457, + "load_gini": 0.6308308029527256, + "max_load_ratio": 5.979673544810594 + }, + "actual_vs_frontier_pearson": -0.08479638082455722, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4680898752103892, + "load_gini": 0.2699814853133662, + "max_load_ratio": 1.810902371419772 + }, + "layer": 19, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.142557882037356, + "load_gini": 0.573428958461657, + "max_load_ratio": 5.856482907299045 + }, + "actual_vs_frontier_pearson": -0.16520423265299047, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48760009420143696, + "load_gini": 0.2809182634932246, + "max_load_ratio": 1.7739451801663073 + }, + "layer": 20, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.2250892374067721, + "load_gini": 0.5839970068524791, + "max_load_ratio": 6.6547582383738835 + }, + "actual_vs_frontier_pearson": 0.030713168827124045, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.44079057860501963, + "load_gini": 0.25344789421004, + "max_load_ratio": 1.7000307976593778 + }, + "layer": 21, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.9505437008130998, + "load_gini": 0.49268164603865117, + "max_load_ratio": 5.647058823529412 + }, + "actual_vs_frontier_pearson": -0.06756357177648568, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5096150033498628, + "load_gini": 0.2927473918232215, + "max_load_ratio": 1.9858330766861718 + }, + "layer": 22, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.0080029689110093, + "load_gini": 0.521078351170311, + "max_load_ratio": 5.109947643979058 + }, + "actual_vs_frontier_pearson": 0.051969177479756205, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.502148823760943, + "load_gini": 0.288229700011549, + "max_load_ratio": 1.968586387434555 + }, + "layer": 23, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.0336730047933789, + "load_gini": 0.5219367156221126, + "max_load_ratio": 6.2309824453341545 + }, + "actual_vs_frontier_pearson": 0.046624159557719196, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4937452666192137, + "load_gini": 0.2847580458885124, + "max_load_ratio": 1.8798891284262396 + }, + "layer": 24, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.9370799623414752, + "load_gini": 0.49417972359100704, + "max_load_ratio": 4.348629504157684 + }, + "actual_vs_frontier_pearson": 0.003630591230348502, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5094100439322153, + "load_gini": 0.29286198077071135, + "max_load_ratio": 1.9267015706806283 + }, + "layer": 25, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.1307693127116234, + "load_gini": 0.562033222975054, + "max_load_ratio": 5.900831536803203 + }, + "actual_vs_frontier_pearson": 0.12889829096423783, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5140701353957097, + "load_gini": 0.2956587330612874, + "max_load_ratio": 1.8700338774253156 + }, + "layer": 26, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.1066234149756957, + "load_gini": 0.5586361569333231, + "max_load_ratio": 5.615029257776409 + }, + "actual_vs_frontier_pearson": 0.027435906393056993, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4562891247455973, + "load_gini": 0.26251997035725294, + "max_load_ratio": 1.8848167539267016 + }, + "layer": 27, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.15194374689878, + "load_gini": 0.5561218167731752, + "max_load_ratio": 6.691715429627348 + }, + "actual_vs_frontier_pearson": -0.0851905104644269, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4931885285350055, + "load_gini": 0.28381426605327986, + "max_load_ratio": 1.7985833076686173 + }, + "layer": 28, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9296875, + "load_cv": 1.1356431750011955, + "load_gini": 0.5496513007198953, + "max_load_ratio": 5.863874345549738 + }, + "actual_vs_frontier_pearson": -0.016979392910092525, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.45012219999295283, + "load_gini": 0.25880680724514926, + "max_load_ratio": 1.7320603634123806 + }, + "layer": 29, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9296875, + "load_cv": 1.3264510778008716, + "load_gini": 0.647222736853249, + "max_load_ratio": 5.84416384354789 + }, + "actual_vs_frontier_pearson": -0.08251025635608537, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.45023597421285283, + "load_gini": 0.25955389109177696, + "max_load_ratio": 1.8453957499230058 + }, + "layer": 30, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.921875, + "load_cv": 1.2804663472936682, + "load_gini": 0.6305014725130891, + "max_load_ratio": 5.723437018786572 + }, + "actual_vs_frontier_pearson": -0.07964706640411469, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5029240802777508, + "load_gini": 0.2893154378079765, + "max_load_ratio": 1.9291653834308593 + }, + "layer": 31, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.1507881607782275, + "load_gini": 0.5729281957383738, + "max_load_ratio": 6.021558361564521 + }, + "actual_vs_frontier_pearson": -0.10452258144829583, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48623369594768334, + "load_gini": 0.2800406505043116, + "max_load_ratio": 1.83061287342162 + }, + "layer": 32, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.2686811219716572, + "load_gini": 0.583570832210502, + "max_load_ratio": 7.844779796735448 + }, + "actual_vs_frontier_pearson": -0.01888683352026748, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48261072157815704, + "load_gini": 0.2782189569217739, + "max_load_ratio": 1.8355404989220818 + }, + "layer": 33, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9921875, + "load_cv": 0.8839459404636472, + "load_gini": 0.4794891799160763, + "max_load_ratio": 3.9618109023714196 + }, + "actual_vs_frontier_pearson": -0.028493286082986517, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5205697308164966, + "load_gini": 0.2990305353980598, + "max_load_ratio": 1.8429319371727748 + }, + "layer": 34, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 0.9362312679691519, + "load_gini": 0.4916611728133662, + "max_load_ratio": 4.747767169695103 + }, + "actual_vs_frontier_pearson": -0.13290690148552678, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.49297481569205803, + "load_gini": 0.2832476372420696, + "max_load_ratio": 1.8995996304280875 + }, + "layer": 35, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 0.9574253286736439, + "load_gini": 0.49605465140899296, + "max_load_ratio": 4.767477671696951 + }, + "actual_vs_frontier_pearson": -0.02212359104446765, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.47796935463002527, + "load_gini": 0.27542581373190633, + "max_load_ratio": 1.7985833076686173 + }, + "layer": 36, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 0.897443530050085, + "load_gini": 0.46608167154296276, + "max_load_ratio": 4.484139205420388 + }, + "actual_vs_frontier_pearson": -0.019587227325278997, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4143052194562559, + "load_gini": 0.2373612300777641, + "max_load_ratio": 1.6975669849091468 + }, + "layer": 37, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.0030386322057563, + "load_gini": 0.5019200415768401, + "max_load_ratio": 5.198644902987373 + }, + "actual_vs_frontier_pearson": 0.028921565758140203, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.48203207236683104, + "load_gini": 0.27775819506467503, + "max_load_ratio": 1.803510933169079 + }, + "layer": 38, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9765625, + "load_cv": 1.1070171860607907, + "load_gini": 0.5396805584963043, + "max_load_ratio": 5.164151524484139 + }, + "actual_vs_frontier_pearson": -0.042234189776402496, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.49820199968589846, + "load_gini": 0.2860022833577147, + "max_load_ratio": 1.9094548814290113 + }, + "layer": 39, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.1143666092181908, + "load_gini": 0.5215875351285801, + "max_load_ratio": 6.679396365876193 + }, + "actual_vs_frontier_pearson": -0.0013066971352570613, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5231859746946843, + "load_gini": 0.3006501193409301, + "max_load_ratio": 1.9168463196797043 + }, + "layer": 40, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9375, + "load_cv": 1.3163699264356221, + "load_gini": 0.576891890591315, + "max_load_ratio": 6.450261780104712 + }, + "actual_vs_frontier_pearson": 0.038349159249901725, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.45930246027155186, + "load_gini": 0.2649497974091468, + "max_load_ratio": 1.815829996920234 + }, + "layer": 41, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9609375, + "load_cv": 1.4367616763803446, + "load_gini": 0.6064525307014168, + "max_load_ratio": 8.630736064059132 + }, + "actual_vs_frontier_pearson": -0.005385768830336229, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.49323689442136087, + "load_gini": 0.2822840073529411, + "max_load_ratio": 1.9069910686787803 + }, + "layer": 42, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.9453125, + "load_cv": 1.5122257286912568, + "load_gini": 0.6255329438712658, + "max_load_ratio": 8.19956883276871 + }, + "actual_vs_frontier_pearson": 0.17269736915133127, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4650946236339636, + "load_gini": 0.2676644426586079, + "max_load_ratio": 1.8700338774253156 + }, + "layer": 43, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.4453751038205662, + "load_gini": 0.6368104813096704, + "max_load_ratio": 6.94548814290114 + }, + "actual_vs_frontier_pearson": -0.18470982231222025, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4751830338518044, + "load_gini": 0.27377735698336925, + "max_load_ratio": 1.8429319371727748 + }, + "layer": 44, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.96875, + "load_cv": 1.464195386895173, + "load_gini": 0.6367434121881737, + "max_load_ratio": 7.145056975669849 + }, + "actual_vs_frontier_pearson": -0.026159680824363943, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4792291952239595, + "load_gini": 0.2763707966007083, + "max_load_ratio": 1.7764089929165383 + }, + "layer": 45, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.3411915327005308, + "load_gini": 0.5801415850207885, + "max_load_ratio": 6.0437326763165995 + }, + "actual_vs_frontier_pearson": 0.02225870145809893, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.5297396161604814, + "load_gini": 0.3045975732406838, + "max_load_ratio": 1.8798891284262396 + }, + "layer": 46, + "phase": "decode", + "total_routed_tokens": 51952 + }, + { + "actual": { + "expert_utilization": 0.984375, + "load_cv": 1.351786104177472, + "load_gini": 0.536063938828149, + "max_load_ratio": 6.908530951647675 + }, + "actual_vs_frontier_pearson": -0.04368832283746608, + "frontier_simulation": { + "expert_utilization": 1.0, + "load_cv": 0.4952080715601842, + "load_gini": 0.2838945685440406, + "max_load_ratio": 1.9168463196797043 + }, + "layer": 47, + "phase": "decode", + "total_routed_tokens": 51952 + } + ], + "schema": "frontier-routing-mismatch.v1", + "source": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-trace-routing-20260716-v1-20260716T151552356215Z/artifacts/artifacts/trace-routing-v1/raw/routing.json" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-s2-profile-ablation.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-s2-profile-ablation.json new file mode 100644 index 0000000..32d8bdf --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/qwen30-s2-profile-ablation.json @@ -0,0 +1,5222 @@ +{ + "historical_modes": { + "historical-per-tp-calibration": { + "false_feasibility": { + "overall": { + "agreement": 64, + "false_feasible": 21, + "false_infeasible": 7 + }, + "per_cell": { + "tp1_mns16": { + "agreement": 7, + "false_feasible": 1, + "false_infeasible": 0 + }, + "tp1_mns32": { + "agreement": 6, + "false_feasible": 2, + "false_infeasible": 0 + }, + "tp1_mns64": { + "agreement": 6, + "false_feasible": 2, + "false_infeasible": 0 + }, + "tp1_mns8": { + "agreement": 6, + "false_feasible": 0, + "false_infeasible": 2 + }, + "tp2_mns16": { + "agreement": 7, + "false_feasible": 1, + "false_infeasible": 0 + }, + "tp2_mns32": { + "agreement": 3, + "false_feasible": 5, + "false_infeasible": 0 + }, + "tp2_mns64": { + "agreement": 2, + "false_feasible": 6, + "false_infeasible": 0 + }, + "tp2_mns8": { + "agreement": 3, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp4_mns16": { + "agreement": 6, + "false_feasible": 1, + "false_infeasible": 0 + }, + "tp4_mns32": { + "agreement": 6, + "false_feasible": 1, + "false_infeasible": 0 + }, + "tp4_mns64": { + "agreement": 6, + "false_feasible": 1, + "false_infeasible": 0 + }, + "tp4_mns8": { + "agreement": 6, + "false_feasible": 1, + "false_infeasible": 0 + } + } + }, + "metrics": { + "kendall_tau_b": { + "both_tied": 4, + "concordant": 58, + "discordant": 0, + "pair_count": 66, + "real_only_tied": 1, + "sim_only_tied": 3, + "tau_b": 0.9668009539030813 + }, + "named_interactions": { + "relations": { + "tp2_mns32_gt_tp4_mns16": true, + "tp4_mns16_eq_tp4_mns32": true, + "tp4_mns16_eq_tp4_mns64": true, + "tp4_mns16_gt_tp1_mns16": true, + "tp4_mns16_gt_tp2_mns16": true, + "tp4_mns16_gt_tp4_mns8": true + }, + "simulator_top_bucket_cells": [ + "tp2_mns32", + "tp2_mns64" + ], + "tp2_mns32_unique_global_best": false, + "trap_reproduced": true + }, + "pairwise_direction": { + "exact_sign_accuracy": 0.9393939393939394, + "exact_sign_agreement": 62, + "non_tied_concordance": 1.0, + "non_tied_concordant": 58, + "non_tied_pair_count": 58, + "pair_count": 66 + }, + "per_mns_tp_direction": { + "accuracy": 1.0, + "comparisons": 8, + "exact_sign_agreement": 8, + "lines": [ + { + "comparisons": 2, + "exact_sign_agreement": 2, + "line_class_agrees": true, + "mns": 8, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + -1 + ], + "sim_line_class": "non-monotonic" + }, + { + "comparisons": 2, + "exact_sign_agreement": 2, + "line_class_agrees": true, + "mns": 16, + "real_adjacent_signs": [ + -1, + 1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + 1 + ], + "sim_line_class": "non-monotonic" + }, + { + "comparisons": 2, + "exact_sign_agreement": 2, + "line_class_agrees": true, + "mns": 32, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + -1 + ], + "sim_line_class": "non-monotonic" + }, + { + "comparisons": 2, + "exact_sign_agreement": 2, + "line_class_agrees": true, + "mns": 64, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + -1 + ], + "sim_line_class": "non-monotonic" + } + ] + }, + "per_tp_mns_direction": { + "accuracy": 0.6666666666666666, + "comparisons": 9, + "exact_sign_agreement": 6, + "lines": [ + { + "comparisons": 3, + "exact_sign_agreement": 2, + "line_class_agrees": false, + "real_adjacent_signs": [ + 1, + -1, + 0 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + 0, + 0 + ], + "sim_line_class": "increasing", + "tp": 1 + }, + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 0, + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + 1, + 0 + ], + "sim_line_class": "increasing", + "tp": 2 + }, + { + "comparisons": 3, + "exact_sign_agreement": 3, + "line_class_agrees": true, + "real_adjacent_signs": [ + 1, + 0, + 0 + ], + "real_line_class": "increasing", + "sim_adjacent_signs": [ + 1, + 0, + 0 + ], + "sim_line_class": "increasing", + "tp": 4 + } + ] + }, + "tie_buckets": { + "real": { + "buckets": { + "tp1_mns16": 715736, + "tp1_mns32": 695431, + "tp1_mns64": 695431, + "tp1_mns8": 639593, + "tp2_mns16": 692893, + "tp2_mns32": 1000000, + "tp2_mns64": 992385, + "tp2_mns8": 692893, + "tp4_mns16": 743654, + "tp4_mns32": 743654, + "tp4_mns64": 743654, + "tp4_mns8": 390862 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp2_mns32" + ], + "interval": [ + 3.2833333333333328, + 3.2833366166666664 + ] + }, + { + "bucket_id": 992385, + "cells": [ + "tp2_mns64" + ], + "interval": [ + 3.25833075, + 3.258334033333333 + ] + }, + { + "bucket_id": 743654, + "cells": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 2.4416639666666664, + 2.4416672499999996 + ] + }, + { + "bucket_id": 715736, + "cells": [ + "tp1_mns16" + ], + "interval": [ + 2.3499998666666664, + 2.3500031499999996 + ] + }, + { + "bucket_id": 695431, + "cells": [ + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 2.283331783333333, + 2.2833350666666665 + ] + }, + { + "bucket_id": 692893, + "cells": [ + "tp2_mns8", + "tp2_mns16" + ], + "interval": [ + 2.274998683333333, + 2.2750019666666663 + ] + }, + { + "bucket_id": 639593, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 2.0999970166666664, + 2.1000002999999996 + ] + }, + { + "bucket_id": 390862, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 1.2833302333333332, + 1.2833335166666666 + ] + } + ], + "tolerance": 3.283333333333333e-06 + }, + "simulator": { + "buckets": { + "tp1_mns16": 635555, + "tp1_mns32": 635555, + "tp1_mns64": 635555, + "tp1_mns8": 457777, + "tp2_mns16": 613333, + "tp2_mns32": 1000000, + "tp2_mns64": 1000000, + "tp2_mns8": 464444, + "tp4_mns16": 666666, + "tp4_mns32": 666666, + "tp4_mns64": 666666, + "tp4_mns8": 352222 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp2_mns32", + "tp2_mns64" + ], + "interval": [ + 3.7499999999999996, + 3.75000375 + ] + }, + { + "bucket_id": 666666, + "cells": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 2.4999974999999997, + 2.50000125 + ] + }, + { + "bucket_id": 635555, + "cells": [ + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 2.38333125, + 2.3833349999999998 + ] + }, + { + "bucket_id": 613333, + "cells": [ + "tp2_mns16" + ], + "interval": [ + 2.29999875, + 2.3000024999999997 + ] + }, + { + "bucket_id": 464444, + "cells": [ + "tp2_mns8" + ], + "interval": [ + 1.7416649999999998, + 1.7416687499999999 + ] + }, + { + "bucket_id": 457777, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 1.71666375, + 1.7166674999999998 + ] + }, + { + "bucket_id": 352222, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 1.3208324999999999, + 1.32083625 + ] + } + ], + "tolerance": 3.7499999999999997e-06 + } + }, + "top1": { + "candidate_cells": [ + "tp2_mns32", + "tp2_mns64" + ], + "effective_k": 2, + "optimistic_regret": 0.0, + "worst_case_regret": 0.0076142131979695165 + }, + "top5": { + "maximum_exact_five_overlap": 5, + "minimum_exact_five_overlap": 5, + "optimistic_regret": 0.0, + "real_cutoff": { + "boundary": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 3, + "expanded_effective_k": 5, + "higher": [ + "tp2_mns32", + "tp2_mns64" + ] + }, + "simulator_cutoff": { + "boundary": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 3, + "expanded_effective_k": 5, + "higher": [ + "tp2_mns32", + "tp2_mns64" + ] + }, + "worst_case_regret": 0.0 + } + }, + "ranking": [ + { + "cell": "tp2_mns32", + "rank": 1, + "score": 3.75 + }, + { + "cell": "tp2_mns64", + "rank": 2, + "score": 3.75 + }, + { + "cell": "tp4_mns16", + "rank": 3, + "score": 2.5 + }, + { + "cell": "tp4_mns32", + "rank": 4, + "score": 2.5 + }, + { + "cell": "tp4_mns64", + "rank": 5, + "score": 2.5 + }, + { + "cell": "tp1_mns16", + "rank": 6, + "score": 2.3833333333333333 + }, + { + "cell": "tp1_mns32", + "rank": 7, + "score": 2.3833333333333333 + }, + { + "cell": "tp1_mns64", + "rank": 8, + "score": 2.3833333333333333 + }, + { + "cell": "tp2_mns16", + "rank": 9, + "score": 2.3 + }, + { + "cell": "tp2_mns8", + "rank": 10, + "score": 1.7416666666666667 + }, + { + "cell": "tp1_mns8", + "rank": 11, + "score": 1.7166666666666666 + }, + { + "cell": "tp4_mns8", + "rank": 12, + "score": 1.3208333333333333 + } + ], + "simulated_scores": { + "tp1_mns16": 2.3833333333333333, + "tp1_mns32": 2.3833333333333333, + "tp1_mns64": 2.3833333333333333, + "tp1_mns8": 1.7166666666666666, + "tp2_mns16": 2.3, + "tp2_mns32": 3.75, + "tp2_mns64": 3.75, + "tp2_mns8": 1.7416666666666667, + "tp4_mns16": 2.5, + "tp4_mns32": 2.5, + "tp4_mns64": 2.5, + "tp4_mns8": 1.3208333333333333 + } + }, + "historical-profile-only": { + "false_feasibility": { + "overall": { + "agreement": 37, + "false_feasible": 0, + "false_infeasible": 55 + }, + "per_cell": { + "tp1_mns16": { + "agreement": 3, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp1_mns32": { + "agreement": 4, + "false_feasible": 0, + "false_infeasible": 4 + }, + "tp1_mns64": { + "agreement": 4, + "false_feasible": 0, + "false_infeasible": 4 + }, + "tp1_mns8": { + "agreement": 5, + "false_feasible": 0, + "false_infeasible": 3 + }, + "tp2_mns16": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp2_mns32": { + "agreement": 5, + "false_feasible": 0, + "false_infeasible": 3 + }, + "tp2_mns64": { + "agreement": 6, + "false_feasible": 0, + "false_infeasible": 2 + }, + "tp2_mns8": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 7 + }, + "tp4_mns16": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp4_mns32": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp4_mns64": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp4_mns8": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 5 + } + } + }, + "metrics": { + "kendall_tau_b": { + "both_tied": 2, + "concordant": 23, + "discordant": 23, + "pair_count": 66, + "real_only_tied": 3, + "sim_only_tied": 15, + "tau_b": 0.0 + }, + "named_interactions": { + "relations": { + "tp2_mns32_gt_tp4_mns16": false, + "tp4_mns16_eq_tp4_mns32": false, + "tp4_mns16_eq_tp4_mns64": false, + "tp4_mns16_gt_tp1_mns16": false, + "tp4_mns16_gt_tp2_mns16": false, + "tp4_mns16_gt_tp4_mns8": false + }, + "simulator_top_bucket_cells": [ + "tp4_mns32", + "tp4_mns64" + ], + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + "pairwise_direction": { + "exact_sign_accuracy": 0.3787878787878788, + "exact_sign_agreement": 25, + "non_tied_concordance": 0.5, + "non_tied_concordant": 23, + "non_tied_pair_count": 46, + "pair_count": 66 + }, + "per_mns_tp_direction": { + "accuracy": 0.0, + "comparisons": 8, + "exact_sign_agreement": 0, + "lines": [ + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": false, + "mns": 8, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + 0 + ], + "sim_line_class": "decreasing" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": true, + "mns": 16, + "real_adjacent_signs": [ + -1, + 1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + -1 + ], + "sim_line_class": "non-monotonic" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": true, + "mns": 32, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + 1 + ], + "sim_line_class": "non-monotonic" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": true, + "mns": 64, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + 1 + ], + "sim_line_class": "non-monotonic" + } + ] + }, + "per_tp_mns_direction": { + "accuracy": 0.2222222222222222, + "comparisons": 9, + "exact_sign_agreement": 2, + "lines": [ + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 1, + -1, + 0 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0, + 0 + ], + "sim_line_class": "flat", + "tp": 1 + }, + { + "comparisons": 3, + "exact_sign_agreement": 0, + "line_class_agrees": true, + "real_adjacent_signs": [ + 0, + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + -1, + 0 + ], + "sim_line_class": "non-monotonic", + "tp": 2 + }, + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": true, + "real_adjacent_signs": [ + 1, + 0, + 0 + ], + "real_line_class": "increasing", + "sim_adjacent_signs": [ + 0, + 1, + 0 + ], + "sim_line_class": "increasing", + "tp": 4 + } + ] + }, + "tie_buckets": { + "real": { + "buckets": { + "tp1_mns16": 715736, + "tp1_mns32": 695431, + "tp1_mns64": 695431, + "tp1_mns8": 639593, + "tp2_mns16": 692893, + "tp2_mns32": 1000000, + "tp2_mns64": 992385, + "tp2_mns8": 692893, + "tp4_mns16": 743654, + "tp4_mns32": 743654, + "tp4_mns64": 743654, + "tp4_mns8": 390862 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp2_mns32" + ], + "interval": [ + 3.2833333333333328, + 3.2833366166666664 + ] + }, + { + "bucket_id": 992385, + "cells": [ + "tp2_mns64" + ], + "interval": [ + 3.25833075, + 3.258334033333333 + ] + }, + { + "bucket_id": 743654, + "cells": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 2.4416639666666664, + 2.4416672499999996 + ] + }, + { + "bucket_id": 715736, + "cells": [ + "tp1_mns16" + ], + "interval": [ + 2.3499998666666664, + 2.3500031499999996 + ] + }, + { + "bucket_id": 695431, + "cells": [ + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 2.283331783333333, + 2.2833350666666665 + ] + }, + { + "bucket_id": 692893, + "cells": [ + "tp2_mns8", + "tp2_mns16" + ], + "interval": [ + 2.274998683333333, + 2.2750019666666663 + ] + }, + { + "bucket_id": 639593, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 2.0999970166666664, + 2.1000002999999996 + ] + }, + { + "bucket_id": 390862, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 1.2833302333333332, + 1.2833335166666666 + ] + } + ], + "tolerance": 3.283333333333333e-06 + }, + "simulator": { + "buckets": { + "tp1_mns16": 832807, + "tp1_mns32": 832807, + "tp1_mns64": 832807, + "tp1_mns8": 832807, + "tp2_mns16": 902208, + "tp2_mns32": 0, + "tp2_mns64": 0, + "tp2_mns8": 0, + "tp4_mns16": 0, + "tp4_mns32": 1000000, + "tp4_mns64": 1000000, + "tp4_mns8": 0 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 1.3208333333333333, + 1.3208346541666667 + ] + }, + { + "bucket_id": 902208, + "cells": [ + "tp2_mns16" + ], + "interval": [ + 1.1916664, + 1.1916677208333333 + ] + }, + { + "bucket_id": 832807, + "cells": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 1.0999992458333334, + 1.1000005666666666 + ] + }, + { + "bucket_id": 0, + "cells": [ + "tp2_mns8", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16" + ], + "interval": [ + 0.0, + 1.3208333333333333e-06 + ] + } + ], + "tolerance": 1.3208333333333333e-06 + } + }, + "top1": { + "candidate_cells": [ + "tp4_mns32", + "tp4_mns64" + ], + "effective_k": 2, + "optimistic_regret": 0.25634517766497456, + "worst_case_regret": 0.25634517766497456 + }, + "top5": { + "maximum_exact_five_overlap": 2, + "minimum_exact_five_overlap": 2, + "optimistic_regret": 0.25634517766497456, + "real_cutoff": { + "boundary": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 3, + "expanded_effective_k": 5, + "higher": [ + "tp2_mns32", + "tp2_mns64" + ] + }, + "simulator_cutoff": { + "boundary": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64" + ], + "choose_from_boundary": 2, + "expanded_effective_k": 7, + "higher": [ + "tp4_mns32", + "tp4_mns64", + "tp2_mns16" + ] + }, + "worst_case_regret": 0.25634517766497456 + } + }, + "ranking": [ + { + "cell": "tp4_mns32", + "rank": 1, + "score": 1.3208333333333333 + }, + { + "cell": "tp4_mns64", + "rank": 2, + "score": 1.3208333333333333 + }, + { + "cell": "tp2_mns16", + "rank": 3, + "score": 1.1916666666666667 + }, + { + "cell": "tp1_mns16", + "rank": 4, + "score": 1.1 + }, + { + "cell": "tp1_mns32", + "rank": 5, + "score": 1.1 + }, + { + "cell": "tp1_mns64", + "rank": 6, + "score": 1.1 + }, + { + "cell": "tp1_mns8", + "rank": 7, + "score": 1.1 + }, + { + "cell": "tp2_mns32", + "rank": 8, + "score": 0.0 + }, + { + "cell": "tp2_mns64", + "rank": 9, + "score": 0.0 + }, + { + "cell": "tp2_mns8", + "rank": 10, + "score": 0.0 + }, + { + "cell": "tp4_mns16", + "rank": 11, + "score": 0.0 + }, + { + "cell": "tp4_mns8", + "rank": 12, + "score": 0.0 + } + ], + "simulated_scores": { + "tp1_mns16": 1.1, + "tp1_mns32": 1.1, + "tp1_mns64": 1.1, + "tp1_mns8": 1.1, + "tp2_mns16": 1.1916666666666667, + "tp2_mns32": 0.0, + "tp2_mns64": 0.0, + "tp2_mns8": 0.0, + "tp4_mns16": 0.0, + "tp4_mns32": 1.3208333333333333, + "tp4_mns64": 1.3208333333333333, + "tp4_mns8": 0.0 + } + } + }, + "real_scores": { + "tp1_mns16": 2.35, + "tp1_mns32": 2.283333333333333, + "tp1_mns64": 2.283333333333333, + "tp1_mns8": 2.1, + "tp2_mns16": 2.275, + "tp2_mns32": 3.283333333333333, + "tp2_mns64": 3.2583333333333333, + "tp2_mns8": 2.275, + "tp4_mns16": 2.441666666666667, + "tp4_mns32": 2.441666666666667, + "tp4_mns64": 2.441666666666667, + "tp4_mns8": 1.2833333333333334 + }, + "schema": "frontier-qwen30-s2-profile-ablation.v1", + "scope": { + "calibration_a_tp": 1.0, + "cells": 12, + "probes": 92, + "trace_horizon_seconds": 60 + }, + "sources": { + "ground_truth": "/home/gahow/phd/replayserve/docs/assets/simfid_s2r/ground_truth.json", + "historical_metrics": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json", + "replacement_profile_shards": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/comparison/new-profile-only-v2-s2/results-shard-00/metrics.json", + "runs": 46 + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/comparison/new-profile-only-v2-s2/results-shard-01/metrics.json", + "runs": 46 + } + ] + }, + "vllm020_profile_only": { + "SLO-gated": { + "cell_score_details": { + "tp1_mns16": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp1_mns32": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp1_mns64": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp1_mns8": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp2_mns16": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp2_mns32": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp2_mns64": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp2_mns8": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 8, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp4_mns16": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 7, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp4_mns32": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 7, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp4_mns64": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 7, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + }, + "tp4_mns8": { + "feasible": false, + "feasible_anchor_count": 0, + "recorded_anchor_count_after_removal": 7, + "score": 0.0, + "selected_probe_index": null, + "selected_sampling_u": null + } + }, + "false_feasibility": { + "overall": { + "agreement": 30, + "false_feasible": 0, + "false_infeasible": 62 + }, + "per_cell": { + "tp1_mns16": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp1_mns32": { + "agreement": 3, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp1_mns64": { + "agreement": 3, + "false_feasible": 0, + "false_infeasible": 5 + }, + "tp1_mns8": { + "agreement": 4, + "false_feasible": 0, + "false_infeasible": 4 + }, + "tp2_mns16": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 7 + }, + "tp2_mns32": { + "agreement": 5, + "false_feasible": 0, + "false_infeasible": 3 + }, + "tp2_mns64": { + "agreement": 6, + "false_feasible": 0, + "false_infeasible": 2 + }, + "tp2_mns8": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 7 + }, + "tp4_mns16": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp4_mns32": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp4_mns64": { + "agreement": 1, + "false_feasible": 0, + "false_infeasible": 6 + }, + "tp4_mns8": { + "agreement": 2, + "false_feasible": 0, + "false_infeasible": 5 + } + } + }, + "loao": { + "defined_replicates": 92, + "range_semantics": "deterministic LOAO sensitivity ranges, not confidence intervals", + "ranges": { + "pairwise_exact_sign_accuracy": { + "max": 0.07575757575757576, + "min": 0.07575757575757576 + }, + "pairwise_non_tied_concordance": { + "max": null, + "min": null + }, + "tau_b": { + "max": 0.0, + "min": 0.0 + }, + "top1_optimistic_regret": { + "max": 0.0, + "min": 0.0 + }, + "top1_worst_case_regret": { + "max": 0.6091370558375634, + "min": 0.6091370558375634 + }, + "top5_maximum_exact_five_overlap": { + "max": 5, + "min": 5 + }, + "top5_minimum_exact_five_overlap": { + "max": 0, + "min": 0 + }, + "top5_optimistic_regret": { + "max": 0.0, + "min": 0.0 + }, + "top5_worst_case_regret": { + "max": 0.30456852791878175, + "min": 0.30456852791878175 + } + }, + "replicate_count": 92, + "replicates": [ + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 7, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 0, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 1, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 2, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 3, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 4, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 5, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.07575757575757576, + "pairwise_non_tied_concordance": null, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 6, + "tau_b": 0.0, + "top1_optimistic_regret": 0.0, + "top1_worst_case_regret": 0.6091370558375634, + "top5_maximum_exact_five_overlap": 5, + "top5_minimum_exact_five_overlap": 0, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.30456852791878175, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + } + ], + "tp2_mns32_unique_global_best_count": 0, + "trap_reproduced_count": 0, + "undefined": [], + "undefined_replicates": 0 + }, + "metrics": { + "kendall_tau_b": { + "both_tied": 5, + "concordant": 0, + "discordant": 0, + "pair_count": 66, + "real_only_tied": 0, + "sim_only_tied": 61, + "tau_b": 0.0 + }, + "named_interactions": { + "relations": { + "tp2_mns32_gt_tp4_mns16": false, + "tp4_mns16_eq_tp4_mns32": true, + "tp4_mns16_eq_tp4_mns64": true, + "tp4_mns16_gt_tp1_mns16": false, + "tp4_mns16_gt_tp2_mns16": false, + "tp4_mns16_gt_tp4_mns8": false + }, + "simulator_top_bucket_cells": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + "pairwise_direction": { + "exact_sign_accuracy": 0.07575757575757576, + "exact_sign_agreement": 5, + "non_tied_concordance": null, + "non_tied_concordant": 0, + "non_tied_pair_count": 0, + "pair_count": 66 + }, + "per_mns_tp_direction": { + "accuracy": 0.0, + "comparisons": 8, + "exact_sign_agreement": 0, + "lines": [ + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": false, + "mns": 8, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0 + ], + "sim_line_class": "flat" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": false, + "mns": 16, + "real_adjacent_signs": [ + -1, + 1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0 + ], + "sim_line_class": "flat" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": false, + "mns": 32, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0 + ], + "sim_line_class": "flat" + }, + { + "comparisons": 2, + "exact_sign_agreement": 0, + "line_class_agrees": false, + "mns": 64, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0 + ], + "sim_line_class": "flat" + } + ] + }, + "per_tp_mns_direction": { + "accuracy": 0.4444444444444444, + "comparisons": 9, + "exact_sign_agreement": 4, + "lines": [ + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 1, + -1, + 0 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0, + 0 + ], + "sim_line_class": "flat", + "tp": 1 + }, + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 0, + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 0, + 0, + 0 + ], + "sim_line_class": "flat", + "tp": 2 + }, + { + "comparisons": 3, + "exact_sign_agreement": 2, + "line_class_agrees": false, + "real_adjacent_signs": [ + 1, + 0, + 0 + ], + "real_line_class": "increasing", + "sim_adjacent_signs": [ + 0, + 0, + 0 + ], + "sim_line_class": "flat", + "tp": 4 + } + ] + }, + "tie_buckets": { + "real": { + "buckets": { + "tp1_mns16": 715736, + "tp1_mns32": 695431, + "tp1_mns64": 695431, + "tp1_mns8": 639593, + "tp2_mns16": 692893, + "tp2_mns32": 1000000, + "tp2_mns64": 992385, + "tp2_mns8": 692893, + "tp4_mns16": 743654, + "tp4_mns32": 743654, + "tp4_mns64": 743654, + "tp4_mns8": 390862 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp2_mns32" + ], + "interval": [ + 3.2833333333333328, + 3.2833366166666664 + ] + }, + { + "bucket_id": 992385, + "cells": [ + "tp2_mns64" + ], + "interval": [ + 3.25833075, + 3.258334033333333 + ] + }, + { + "bucket_id": 743654, + "cells": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 2.4416639666666664, + 2.4416672499999996 + ] + }, + { + "bucket_id": 715736, + "cells": [ + "tp1_mns16" + ], + "interval": [ + 2.3499998666666664, + 2.3500031499999996 + ] + }, + { + "bucket_id": 695431, + "cells": [ + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 2.283331783333333, + 2.2833350666666665 + ] + }, + { + "bucket_id": 692893, + "cells": [ + "tp2_mns8", + "tp2_mns16" + ], + "interval": [ + 2.274998683333333, + 2.2750019666666663 + ] + }, + { + "bucket_id": 639593, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 2.0999970166666664, + 2.1000002999999996 + ] + }, + { + "bucket_id": 390862, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 1.2833302333333332, + 1.2833335166666666 + ] + } + ], + "tolerance": 3.283333333333333e-06 + }, + "simulator": { + "buckets": { + "tp1_mns16": 0, + "tp1_mns32": 0, + "tp1_mns64": 0, + "tp1_mns8": 0, + "tp2_mns16": 0, + "tp2_mns32": 0, + "tp2_mns64": 0, + "tp2_mns8": 0, + "tp4_mns16": 0, + "tp4_mns32": 0, + "tp4_mns64": 0, + "tp4_mns8": 0 + }, + "occupied_intervals": [ + { + "bucket_id": 0, + "cells": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 0.0, + 1e-09 + ] + } + ], + "tolerance": 1e-09 + } + }, + "top1": { + "candidate_cells": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "effective_k": 12, + "optimistic_regret": 0.0, + "worst_case_regret": 0.6091370558375634 + }, + "top5": { + "maximum_exact_five_overlap": 5, + "minimum_exact_five_overlap": 0, + "optimistic_regret": 0.0, + "real_cutoff": { + "boundary": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 3, + "expanded_effective_k": 5, + "higher": [ + "tp2_mns32", + "tp2_mns64" + ] + }, + "simulator_cutoff": { + "boundary": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 5, + "expanded_effective_k": 12, + "higher": [] + }, + "worst_case_regret": 0.30456852791878175 + } + }, + "ranking": [ + { + "cell": "tp1_mns16", + "rank": 1, + "score": 0.0 + }, + { + "cell": "tp1_mns32", + "rank": 2, + "score": 0.0 + }, + { + "cell": "tp1_mns64", + "rank": 3, + "score": 0.0 + }, + { + "cell": "tp1_mns8", + "rank": 4, + "score": 0.0 + }, + { + "cell": "tp2_mns16", + "rank": 5, + "score": 0.0 + }, + { + "cell": "tp2_mns32", + "rank": 6, + "score": 0.0 + }, + { + "cell": "tp2_mns64", + "rank": 7, + "score": 0.0 + }, + { + "cell": "tp2_mns8", + "rank": 8, + "score": 0.0 + }, + { + "cell": "tp4_mns16", + "rank": 9, + "score": 0.0 + }, + { + "cell": "tp4_mns32", + "rank": 10, + "score": 0.0 + }, + { + "cell": "tp4_mns64", + "rank": 11, + "score": 0.0 + }, + { + "cell": "tp4_mns8", + "rank": 12, + "score": 0.0 + } + ], + "reading": "SLO-gated", + "simulated_scores": { + "tp1_mns16": 0.0, + "tp1_mns32": 0.0, + "tp1_mns64": 0.0, + "tp1_mns8": 0.0, + "tp2_mns16": 0.0, + "tp2_mns32": 0.0, + "tp2_mns64": 0.0, + "tp2_mns8": 0.0, + "tp4_mns16": 0.0, + "tp4_mns32": 0.0, + "tp4_mns64": 0.0, + "tp4_mns8": 0.0 + } + }, + "throughput-proxy": { + "cell_score_details": { + "tp1_mns16": { + "recorded_anchor_count_after_removal": 8, + "score": 0.8453202347375465, + "selected_probe_index": 1, + "selected_sampling_u": 0.25, + "valid_anchor_count": 8 + }, + "tp1_mns32": { + "recorded_anchor_count_after_removal": 8, + "score": 1.4213857973581445, + "selected_probe_index": 0, + "selected_sampling_u": 0.5, + "valid_anchor_count": 8 + }, + "tp1_mns64": { + "recorded_anchor_count_after_removal": 8, + "score": 2.1437448608949605, + "selected_probe_index": 0, + "selected_sampling_u": 0.5, + "valid_anchor_count": 8 + }, + "tp1_mns8": { + "recorded_anchor_count_after_removal": 8, + "score": 0.5319246206192596, + "selected_probe_index": 1, + "selected_sampling_u": 0.25, + "valid_anchor_count": 8 + }, + "tp2_mns16": { + "recorded_anchor_count_after_removal": 8, + "score": 0.6441840113363596, + "selected_probe_index": 4, + "selected_sampling_u": 0.46875, + "valid_anchor_count": 8 + }, + "tp2_mns32": { + "recorded_anchor_count_after_removal": 8, + "score": 1.093439153724593, + "selected_probe_index": 4, + "selected_sampling_u": 0.78125, + "valid_anchor_count": 8 + }, + "tp2_mns64": { + "recorded_anchor_count_after_removal": 8, + "score": 1.7227463064785613, + "selected_probe_index": 2, + "selected_sampling_u": 0.875, + "valid_anchor_count": 8 + }, + "tp2_mns8": { + "recorded_anchor_count_after_removal": 8, + "score": 0.3900292387118045, + "selected_probe_index": 4, + "selected_sampling_u": 0.46875, + "valid_anchor_count": 8 + }, + "tp4_mns16": { + "recorded_anchor_count_after_removal": 7, + "score": 0.4355571852372527, + "selected_probe_index": 5, + "selected_sampling_u": 0.033182214016, + "valid_anchor_count": 7 + }, + "tp4_mns32": { + "recorded_anchor_count_after_removal": 7, + "score": 0.760215065879823, + "selected_probe_index": 5, + "selected_sampling_u": 0.033182214016, + "valid_anchor_count": 7 + }, + "tp4_mns64": { + "recorded_anchor_count_after_removal": 7, + "score": 1.2313111686958826, + "selected_probe_index": 5, + "selected_sampling_u": 0.033182214016, + "valid_anchor_count": 7 + }, + "tp4_mns8": { + "recorded_anchor_count_after_removal": 7, + "score": 0.25160540206585935, + "selected_probe_index": 0, + "selected_sampling_u": 0.034252608017, + "valid_anchor_count": 7 + } + }, + "loao": { + "defined_replicates": 92, + "range_semantics": "deterministic LOAO sensitivity ranges, not confidence intervals", + "ranges": { + "pairwise_exact_sign_accuracy": { + "max": 0.6515151515151515, + "min": 0.6515151515151515 + }, + "pairwise_non_tied_concordance": { + "max": 0.7049180327868853, + "min": 0.7049180327868853 + }, + "tau_b": { + "max": 0.39400626128204935, + "min": 0.39400626128204935 + }, + "top1_optimistic_regret": { + "max": 0.30456852791878175, + "min": 0.30456852791878175 + }, + "top1_worst_case_regret": { + "max": 0.30456852791878175, + "min": 0.30456852791878175 + }, + "top5_maximum_exact_five_overlap": { + "max": 3, + "min": 3 + }, + "top5_minimum_exact_five_overlap": { + "max": 3, + "min": 3 + }, + "top5_optimistic_regret": { + "max": 0.0, + "min": 0.0 + }, + "top5_worst_case_regret": { + "max": 0.0, + "min": 0.0 + } + }, + "replicate_count": 92, + "replicates": [ + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns8", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns16", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns32", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp1_mns64", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns8", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns16", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns32", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp2_mns64", + "removed_probe_index": 7, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns8", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns16", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns32", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 0, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 1, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 2, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 3, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 4, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 5, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + { + "pairwise_exact_sign_accuracy": 0.6515151515151515, + "pairwise_non_tied_concordance": 0.7049180327868853, + "removed_cell_id": "tp4_mns64", + "removed_probe_index": 6, + "tau_b": 0.39400626128204935, + "top1_optimistic_regret": 0.30456852791878175, + "top1_worst_case_regret": 0.30456852791878175, + "top5_maximum_exact_five_overlap": 3, + "top5_minimum_exact_five_overlap": 3, + "top5_optimistic_regret": 0.0, + "top5_worst_case_regret": 0.0, + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + } + ], + "tp2_mns32_unique_global_best_count": 0, + "trap_reproduced_count": 0, + "undefined": [], + "undefined_replicates": 0 + }, + "metrics": { + "kendall_tau_b": { + "both_tied": 0, + "concordant": 43, + "discordant": 18, + "pair_count": 66, + "real_only_tied": 5, + "sim_only_tied": 0, + "tau_b": 0.39400626128204935 + }, + "named_interactions": { + "relations": { + "tp2_mns32_gt_tp4_mns16": true, + "tp4_mns16_eq_tp4_mns32": false, + "tp4_mns16_eq_tp4_mns64": false, + "tp4_mns16_gt_tp1_mns16": false, + "tp4_mns16_gt_tp2_mns16": false, + "tp4_mns16_gt_tp4_mns8": true + }, + "simulator_top_bucket_cells": [ + "tp1_mns64" + ], + "tp2_mns32_unique_global_best": false, + "trap_reproduced": false + }, + "pairwise_direction": { + "exact_sign_accuracy": 0.6515151515151515, + "exact_sign_agreement": 43, + "non_tied_concordance": 0.7049180327868853, + "non_tied_concordant": 43, + "non_tied_pair_count": 61, + "pair_count": 66 + }, + "per_mns_tp_direction": { + "accuracy": 0.5, + "comparisons": 8, + "exact_sign_agreement": 4, + "lines": [ + { + "comparisons": 2, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "mns": 8, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + -1 + ], + "sim_line_class": "decreasing" + }, + { + "comparisons": 2, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "mns": 16, + "real_adjacent_signs": [ + -1, + 1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + -1 + ], + "sim_line_class": "decreasing" + }, + { + "comparisons": 2, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "mns": 32, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + -1 + ], + "sim_line_class": "decreasing" + }, + { + "comparisons": 2, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "mns": 64, + "real_adjacent_signs": [ + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + -1, + -1 + ], + "sim_line_class": "decreasing" + } + ] + }, + "per_tp_mns_direction": { + "accuracy": 0.3333333333333333, + "comparisons": 9, + "exact_sign_agreement": 3, + "lines": [ + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 1, + -1, + 0 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + 1, + 1 + ], + "sim_line_class": "increasing", + "tp": 1 + }, + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": false, + "real_adjacent_signs": [ + 0, + 1, + -1 + ], + "real_line_class": "non-monotonic", + "sim_adjacent_signs": [ + 1, + 1, + 1 + ], + "sim_line_class": "increasing", + "tp": 2 + }, + { + "comparisons": 3, + "exact_sign_agreement": 1, + "line_class_agrees": true, + "real_adjacent_signs": [ + 1, + 0, + 0 + ], + "real_line_class": "increasing", + "sim_adjacent_signs": [ + 1, + 1, + 1 + ], + "sim_line_class": "increasing", + "tp": 4 + } + ] + }, + "tie_buckets": { + "real": { + "buckets": { + "tp1_mns16": 715736, + "tp1_mns32": 695431, + "tp1_mns64": 695431, + "tp1_mns8": 639593, + "tp2_mns16": 692893, + "tp2_mns32": 1000000, + "tp2_mns64": 992385, + "tp2_mns8": 692893, + "tp4_mns16": 743654, + "tp4_mns32": 743654, + "tp4_mns64": 743654, + "tp4_mns8": 390862 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp2_mns32" + ], + "interval": [ + 3.2833333333333328, + 3.2833366166666664 + ] + }, + { + "bucket_id": 992385, + "cells": [ + "tp2_mns64" + ], + "interval": [ + 3.25833075, + 3.258334033333333 + ] + }, + { + "bucket_id": 743654, + "cells": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "interval": [ + 2.4416639666666664, + 2.4416672499999996 + ] + }, + { + "bucket_id": 715736, + "cells": [ + "tp1_mns16" + ], + "interval": [ + 2.3499998666666664, + 2.3500031499999996 + ] + }, + { + "bucket_id": 695431, + "cells": [ + "tp1_mns32", + "tp1_mns64" + ], + "interval": [ + 2.283331783333333, + 2.2833350666666665 + ] + }, + { + "bucket_id": 692893, + "cells": [ + "tp2_mns8", + "tp2_mns16" + ], + "interval": [ + 2.274998683333333, + 2.2750019666666663 + ] + }, + { + "bucket_id": 639593, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 2.0999970166666664, + 2.1000002999999996 + ] + }, + { + "bucket_id": 390862, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 1.2833302333333332, + 1.2833335166666666 + ] + } + ], + "tolerance": 3.283333333333333e-06 + }, + "simulator": { + "buckets": { + "tp1_mns16": 394319, + "tp1_mns32": 663038, + "tp1_mns64": 1000000, + "tp1_mns8": 248128, + "tp2_mns16": 300494, + "tp2_mns32": 510060, + "tp2_mns64": 803615, + "tp2_mns8": 181938, + "tp4_mns16": 203175, + "tp4_mns32": 354620, + "tp4_mns64": 574373, + "tp4_mns8": 117367 + }, + "occupied_intervals": [ + { + "bucket_id": 1000000, + "cells": [ + "tp1_mns64" + ], + "interval": [ + 2.1437448608949605, + 2.1437470046398217 + ] + }, + { + "bucket_id": 803615, + "cells": [ + "tp2_mns64" + ], + "interval": [ + 1.7227455263881037, + 1.7227476701329647 + ] + }, + { + "bucket_id": 663038, + "cells": [ + "tp1_mns32" + ], + "interval": [ + 1.421384305078073, + 1.4213864488229337 + ] + }, + { + "bucket_id": 574373, + "cells": [ + "tp4_mns64" + ], + "interval": [ + 1.2313091669868212, + 1.2313113107316822 + ] + }, + { + "bucket_id": 510060, + "cells": [ + "tp2_mns32" + ], + "interval": [ + 1.0934385037480836, + 1.0934406474929446 + ] + }, + { + "bucket_id": 394319, + "cells": [ + "tp1_mns16" + ], + "interval": [ + 0.84531932980324, + 0.8453214735481008 + ] + }, + { + "bucket_id": 354620, + "cells": [ + "tp4_mns32" + ], + "interval": [ + 0.760214802570571, + 0.7602169463154318 + ] + }, + { + "bucket_id": 300494, + "cells": [ + "tp2_mns16" + ], + "interval": [ + 0.6441824682297703, + 0.6441846119746312 + ] + }, + { + "bucket_id": 248128, + "cells": [ + "tp1_mns8" + ], + "interval": [ + 0.5319231248441448, + 0.5319252685890057 + ] + }, + { + "bucket_id": 203175, + "cells": [ + "tp4_mns16" + ], + "interval": [ + 0.43555536211233364, + 0.4355575058571945 + ] + }, + { + "bucket_id": 181938, + "cells": [ + "tp2_mns8" + ], + "interval": [ + 0.39002865250150737, + 0.3900307962463682 + ] + }, + { + "bucket_id": 117367, + "cells": [ + "tp4_mns8" + ], + "interval": [ + 0.25160490308865885, + 0.25160704683351975 + ] + } + ], + "tolerance": 2.1437448608949606e-06 + } + }, + "top1": { + "candidate_cells": [ + "tp1_mns64" + ], + "effective_k": 1, + "optimistic_regret": 0.30456852791878175, + "worst_case_regret": 0.30456852791878175 + }, + "top5": { + "maximum_exact_five_overlap": 3, + "minimum_exact_five_overlap": 3, + "optimistic_regret": 0.0, + "real_cutoff": { + "boundary": [ + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "choose_from_boundary": 3, + "expanded_effective_k": 5, + "higher": [ + "tp2_mns32", + "tp2_mns64" + ] + }, + "simulator_cutoff": { + "boundary": [ + "tp2_mns32" + ], + "choose_from_boundary": 1, + "expanded_effective_k": 5, + "higher": [ + "tp1_mns64", + "tp2_mns64", + "tp1_mns32", + "tp4_mns64" + ] + }, + "worst_case_regret": 0.0 + } + }, + "ranking": [ + { + "cell": "tp1_mns64", + "rank": 1, + "score": 2.1437448608949605 + }, + { + "cell": "tp2_mns64", + "rank": 2, + "score": 1.7227463064785613 + }, + { + "cell": "tp1_mns32", + "rank": 3, + "score": 1.4213857973581445 + }, + { + "cell": "tp4_mns64", + "rank": 4, + "score": 1.2313111686958826 + }, + { + "cell": "tp2_mns32", + "rank": 5, + "score": 1.093439153724593 + }, + { + "cell": "tp1_mns16", + "rank": 6, + "score": 0.8453202347375465 + }, + { + "cell": "tp4_mns32", + "rank": 7, + "score": 0.760215065879823 + }, + { + "cell": "tp2_mns16", + "rank": 8, + "score": 0.6441840113363596 + }, + { + "cell": "tp1_mns8", + "rank": 9, + "score": 0.5319246206192596 + }, + { + "cell": "tp4_mns16", + "rank": 10, + "score": 0.4355571852372527 + }, + { + "cell": "tp2_mns8", + "rank": 11, + "score": 0.3900292387118045 + }, + { + "cell": "tp4_mns8", + "rank": 12, + "score": 0.25160540206585935 + } + ], + "reading": "throughput-proxy", + "simulated_scores": { + "tp1_mns16": 0.8453202347375465, + "tp1_mns32": 1.4213857973581445, + "tp1_mns64": 2.1437448608949605, + "tp1_mns8": 0.5319246206192596, + "tp2_mns16": 0.6441840113363596, + "tp2_mns32": 1.093439153724593, + "tp2_mns64": 1.7227463064785613, + "tp2_mns8": 0.3900292387118045, + "tp4_mns16": 0.4355571852372527, + "tp4_mns32": 0.760215065879823, + "tp4_mns64": 1.2313111686958826, + "tp4_mns8": 0.25160540206585935 + } + } + } +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-routing-fixture-manifest.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-routing-fixture-manifest.json new file mode 100644 index 0000000..ff78c14 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-routing-fixture-manifest.json @@ -0,0 +1,108 @@ +{ + "contains_prompt_text": false, + "fixture_path": "/tmp/qwen30-routing-fixture-20260716.jsonl", + "fixture_sha256": "e9e7f5b4e0d3a788dcd99d432f939d9e36bff2a64e412649b407b0609f0e39bb", + "prefix_pairs": [ + { + "child_row_id": "17366", + "child_turn": 13, + "parent_row_id": "16914", + "parent_turn": 12, + "trace_hash_common_prefix_blocks": 59 + } + ], + "request_count": 8, + "rows": [ + { + "chat_id": "184516", + "fixture_index": 0, + "input_length": 3791, + "output_length": 73, + "parent_chat_id": "183921", + "prompt_sha256": "6b59c3ecb9a7d8bd3178f10b847a7c16c221fef113df3e80ede55543bc66ea6b", + "row_id": "16914", + "trace_hash_blocks": 60, + "turn": 12 + }, + { + "chat_id": "183512", + "fixture_index": 1, + "input_length": 264, + "output_length": 975, + "parent_chat_id": "-1", + "prompt_sha256": "0af4fe018ad7afc152951995e5cb18f389bacf4076239f5a72165a9c84dd566d", + "row_id": "15910", + "trace_hash_blocks": 5, + "turn": 1 + }, + { + "chat_id": "189951", + "fixture_index": 2, + "input_length": 488, + "output_length": 863, + "parent_chat_id": "-1", + "prompt_sha256": "b84c0580e9cd80e8dd388f176bd51119f06ee4686838d957b4018b8a1feccb15", + "row_id": "22349", + "trace_hash_blocks": 8, + "turn": 1 + }, + { + "chat_id": "177472", + "fixture_index": 3, + "input_length": 1037, + "output_length": 895, + "parent_chat_id": "-1", + "prompt_sha256": "146dc187af0eeb342dd7bd6ebd9453973074209ab4da5b122718b5d9e06d46d1", + "row_id": "9870", + "trace_hash_blocks": 17, + "turn": 1 + }, + { + "chat_id": "177528", + "fixture_index": 4, + "input_length": 1993, + "output_length": 654, + "parent_chat_id": "-1", + "prompt_sha256": "c92ba8acd0d637b796d60b5b01e3ac54bde70a986e83809b434c46f50e5242cf", + "row_id": "9926", + "trace_hash_blocks": 32, + "turn": 1 + }, + { + "chat_id": "193539", + "fixture_index": 5, + "input_length": 4088, + "output_length": 1842, + "parent_chat_id": "-1", + "prompt_sha256": "38eecbf8766bd8432ac41d6a061a74244c4b3f8b20e8dfee7237b5e6c0e9e13e", + "row_id": "25937", + "trace_hash_blocks": 64, + "turn": 1 + }, + { + "chat_id": "177590", + "fixture_index": 6, + "input_length": 7995, + "output_length": 1128, + "parent_chat_id": "-1", + "prompt_sha256": "4bee5b9d1aceec7010a80b8585fb5071b6d468c751e5903abbcd656a8285fcd8", + "row_id": "9988", + "trace_hash_blocks": 125, + "turn": 1 + }, + { + "chat_id": "184968", + "fixture_index": 7, + "input_length": 4017, + "output_length": 72, + "parent_chat_id": "184516", + "prompt_sha256": "6885aff07780bff5906669057d46935d78148fb7f4edd0bf58d44c2ffec76952", + "row_id": "17366", + "trace_hash_blocks": 63, + "turn": 13 + } + ], + "schema_version": "qwen30_trace_routing_fixture.v1", + "source_trace": "/home/gahow/phd/aituner/trace_windows/traces/chat_w20260311_1000.jsonl", + "source_trace_sha256": "f539f38eb0ee0f750e3c23ff47df6eed3faf723a25f1444d55665a85871750b9" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-support/support.json b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-support/support.json new file mode 100644 index 0000000..5bd1e82 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/artifacts/trace-support/support.json @@ -0,0 +1,469 @@ +{ + "block_size": 64, + "cohort_output": "/home/admin/cpfs/wjh/frontier_qwen30_vllm020_profiles/trace-support/cohort.jsonl", + "cohort_sha256": "9ef15a4c8843127a5c52ff8455bdeb9945ac6921f2977dd11070ed6944eed7c8", + "full_downsampled_cohort": { + "hash_blocks_per_request": { + "max": 128.0, + "min": 2.0, + "p10": 4.0, + "p25": 15.0, + "p50": 38.0, + "p75": 82.0, + "p90": 111.0, + "p95": 122.0, + "p99": 127.0 + }, + "input_tokens": { + "max": 8186.0, + "min": 71.0, + "p10": 208.0, + "p25": 933.0, + "p50": 2421.0, + "p75": 5244.0, + "p90": 7103.0, + "p95": 7775.0, + "p99": 8074.0 + }, + "interarrival_s": { + "max": 0.20259999999998257, + "min": 0.06279999999997088, + "p10": 0.0917999999999708, + "p25": 0.10240000000003846, + "p50": 0.1163999999999703, + "p75": 0.13009999999999167, + "p90": 0.14210000000002765, + "p95": 0.15119999999997447, + "p99": 0.17959999999997933 + }, + "multi_turn_fraction": 0.140625, + "output_tokens": { + "max": 128.0, + "min": 128.0, + "p10": 128.0, + "p25": 128.0, + "p50": 128.0, + "p75": 128.0, + "p90": 128.0, + "p95": 128.0, + "p99": 128.0 + }, + "prefix_reuse_upper_bound": { + "any_position_reusable_block_ratio": 0.09274807669777795, + "leading_reusable_blocks_per_request": { + "max": 55.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 1.0, + "p75": 13.0, + "p90": 13.0, + "p95": 13.0, + "p99": 49.0 + }, + "leading_reusable_tokens_per_request": { + "max": 3520.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 64.0, + "p75": 832.0, + "p90": 832.0, + "p95": 832.0, + "p99": 3136.0 + }, + "rows_with_hash_ids": 512, + "semantics": "arrival-ordered, infinite-capacity/no-eviction upper bound; not an observed KV-cache hit rate", + "total_blocks": 25607, + "unique_blocks": 23232 + }, + "request_count": 512, + "sampling_u": { + "max": 0.9979913161468553, + "min": 0.0017297900650431776, + "p10": 0.10365756051317851, + "p25": 0.22877823992809823, + "p50": 0.4667483893139273, + "p75": 0.7338360613813898, + "p90": 0.8969698510725516, + "p95": 0.9558957048665733, + "p99": 0.991130680217395 + } + }, + "limits": [ + "Trace length/hash support does not determine dynamic decode or mixed batch shapes.", + "Those shapes depend jointly on arrival history, SLO pressure, TP execution time, MNS, chunking, and KV eviction.", + "MoE expert routing is not present in the trace and must be measured from model execution." + ], + "loader_contract": { + "completion_tokens_override": 128, + "downsampling": "AITuner _downsample_requests before sampling_u threshold", + "input_length_filter": { + "max": 8192, + "min": 0 + }, + "max_requests_per_probe": 512, + "ordering": "arrival_s", + "replay_time_scale": 0.1 + }, + "schema_version": "qwen30_trace_profile_support.v1", + "study": "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/configs/examples/dash0_qwen30b_a3b_community_vllm020_noharness.json", + "study_sha256": "cd0b9e2385751478f459faeae2cf3d12c58229866176911051421bc043bf99d4", + "threshold_cohorts": { + "0.125": { + "hash_blocks_per_request": { + "max": 127.0, + "min": 2.0, + "p10": 7.0, + "p25": 15.0, + "p50": 30.0, + "p75": 76.0, + "p90": 105.0, + "p95": 111.0, + "p99": 127.0 + }, + "input_tokens": { + "max": 8110.0, + "min": 74.0, + "p10": 392.0, + "p25": 922.0, + "p50": 1864.0, + "p75": 4844.0, + "p90": 6686.0, + "p95": 7054.0, + "p99": 8110.0 + }, + "interarrival_s": { + "max": 6.217200000000002, + "min": 0.10639999999998517, + "p10": 0.16680000000001272, + "p25": 0.30810000000001736, + "p50": 0.6001999999999938, + "p75": 1.1201000000000016, + "p90": 1.8216999999999643, + "p95": 2.419699999999967, + "p99": 6.217200000000002 + }, + "multi_turn_fraction": 0.15151515151515152, + "output_tokens": { + "max": 128.0, + "min": 128.0, + "p10": 128.0, + "p25": 128.0, + "p50": 128.0, + "p75": 128.0, + "p90": 128.0, + "p95": 128.0, + "p99": 128.0 + }, + "prefix_reuse_upper_bound": { + "any_position_reusable_block_ratio": 0.10466666666666667, + "leading_reusable_blocks_per_request": { + "max": 13.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 1.0, + "p75": 13.0, + "p90": 13.0, + "p95": 13.0, + "p99": 13.0 + }, + "leading_reusable_tokens_per_request": { + "max": 832.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 64.0, + "p75": 832.0, + "p90": 832.0, + "p95": 832.0, + "p99": 832.0 + }, + "rows_with_hash_ids": 66, + "semantics": "arrival-ordered, infinite-capacity/no-eviction upper bound; not an observed KV-cache hit rate", + "total_blocks": 3000, + "unique_blocks": 2686 + }, + "request_count": 66, + "sampling_u": { + "max": 0.12321850600323363, + "min": 0.0017297900650431776, + "p10": 0.01599560440427848, + "p25": 0.03636425746001375, + "p50": 0.06451196858246099, + "p75": 0.10162658487005978, + "p90": 0.1138971596689021, + "p95": 0.12001874701158237, + "p99": 0.12321850600323363 + } + }, + "0.25": { + "hash_blocks_per_request": { + "max": 128.0, + "min": 2.0, + "p10": 4.0, + "p25": 14.0, + "p50": 34.0, + "p75": 87.0, + "p90": 107.0, + "p95": 122.0, + "p99": 127.0 + }, + "input_tokens": { + "max": 8149.0, + "min": 72.0, + "p10": 197.0, + "p25": 867.0, + "p50": 2167.0, + "p75": 5568.0, + "p90": 6841.0, + "p95": 7745.0, + "p99": 8110.0 + }, + "interarrival_s": { + "max": 2.3110999999999855, + "min": 0.07340000000003499, + "p10": 0.11439999999997497, + "p25": 0.14210000000002765, + "p50": 0.3411000000000044, + "p75": 0.5650000000000093, + "p90": 0.8215000000000146, + "p95": 1.088799999999992, + "p99": 1.7577999999999996 + }, + "multi_turn_fraction": 0.11888111888111888, + "output_tokens": { + "max": 128.0, + "min": 128.0, + "p10": 128.0, + "p25": 128.0, + "p50": 128.0, + "p75": 128.0, + "p90": 128.0, + "p95": 128.0, + "p99": 128.0 + }, + "prefix_reuse_upper_bound": { + "any_position_reusable_block_ratio": 0.07953890489913544, + "leading_reusable_blocks_per_request": { + "max": 13.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 1.0, + "p75": 13.0, + "p90": 13.0, + "p95": 13.0, + "p99": 13.0 + }, + "leading_reusable_tokens_per_request": { + "max": 832.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 64.0, + "p75": 832.0, + "p90": 832.0, + "p95": 832.0, + "p99": 832.0 + }, + "rows_with_hash_ids": 143, + "semantics": "arrival-ordered, infinite-capacity/no-eviction upper bound; not an observed KV-cache hit rate", + "total_blocks": 6940, + "unique_blocks": 6388 + }, + "request_count": 143, + "sampling_u": { + "max": 0.24941730181025415, + "min": 0.0017297900650431776, + "p10": 0.03095529459918134, + "p25": 0.07431293234041883, + "p50": 0.13301201741611407, + "p75": 0.1994148366998705, + "p90": 0.2292686392189136, + "p95": 0.24107176388894497, + "p99": 0.24671643798343437 + } + }, + "0.5": { + "hash_blocks_per_request": { + "max": 128.0, + "min": 2.0, + "p10": 2.0, + "p25": 15.0, + "p50": 42.0, + "p75": 86.0, + "p90": 109.0, + "p95": 117.0, + "p99": 127.0 + }, + "input_tokens": { + "max": 8149.0, + "min": 71.0, + "p10": 113.0, + "p25": 905.0, + "p50": 2648.0, + "p75": 5490.0, + "p90": 6926.0, + "p95": 7476.0, + "p99": 8074.0 + }, + "interarrival_s": { + "max": 1.2561000000000142, + "min": 0.06749999999997236, + "p10": 0.10019999999999563, + "p25": 0.11840000000002249, + "p50": 0.15819999999994394, + "p75": 0.2599999999999483, + "p90": 0.3867999999999938, + "p95": 0.5083999999999804, + "p99": 0.696700000000007 + }, + "multi_turn_fraction": 0.14130434782608695, + "output_tokens": { + "max": 128.0, + "min": 128.0, + "p10": 128.0, + "p25": 128.0, + "p50": 128.0, + "p75": 128.0, + "p90": 128.0, + "p95": 128.0, + "p99": 128.0 + }, + "prefix_reuse_upper_bound": { + "any_position_reusable_block_ratio": 0.08318264014466546, + "leading_reusable_blocks_per_request": { + "max": 52.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 1.0, + "p75": 13.0, + "p90": 13.0, + "p95": 13.0, + "p99": 24.0 + }, + "leading_reusable_tokens_per_request": { + "max": 3328.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 64.0, + "p75": 832.0, + "p90": 832.0, + "p95": 832.0, + "p99": 1536.0 + }, + "rows_with_hash_ids": 276, + "semantics": "arrival-ordered, infinite-capacity/no-eviction upper bound; not an observed KV-cache hit rate", + "total_blocks": 13825, + "unique_blocks": 12675 + }, + "request_count": 276, + "sampling_u": { + "max": 0.49832600818075473, + "min": 0.0017297900650431776, + "p10": 0.056494245471656455, + "p25": 0.1313132001019981, + "p50": 0.24403265916374703, + "p75": 0.3706937275751198, + "p90": 0.4530058956953501, + "p95": 0.48171254599865826, + "p99": 0.49639910750753047 + } + }, + "1.0": { + "hash_blocks_per_request": { + "max": 128.0, + "min": 2.0, + "p10": 4.0, + "p25": 15.0, + "p50": 38.0, + "p75": 82.0, + "p90": 111.0, + "p95": 122.0, + "p99": 127.0 + }, + "input_tokens": { + "max": 8186.0, + "min": 71.0, + "p10": 208.0, + "p25": 933.0, + "p50": 2421.0, + "p75": 5244.0, + "p90": 7103.0, + "p95": 7775.0, + "p99": 8074.0 + }, + "interarrival_s": { + "max": 0.20259999999998257, + "min": 0.06279999999997088, + "p10": 0.0917999999999708, + "p25": 0.10240000000003846, + "p50": 0.1163999999999703, + "p75": 0.13009999999999167, + "p90": 0.14210000000002765, + "p95": 0.15119999999997447, + "p99": 0.17959999999997933 + }, + "multi_turn_fraction": 0.140625, + "output_tokens": { + "max": 128.0, + "min": 128.0, + "p10": 128.0, + "p25": 128.0, + "p50": 128.0, + "p75": 128.0, + "p90": 128.0, + "p95": 128.0, + "p99": 128.0 + }, + "prefix_reuse_upper_bound": { + "any_position_reusable_block_ratio": 0.09274807669777795, + "leading_reusable_blocks_per_request": { + "max": 55.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 1.0, + "p75": 13.0, + "p90": 13.0, + "p95": 13.0, + "p99": 49.0 + }, + "leading_reusable_tokens_per_request": { + "max": 3520.0, + "min": 0.0, + "p10": 0.0, + "p25": 0.0, + "p50": 64.0, + "p75": 832.0, + "p90": 832.0, + "p95": 832.0, + "p99": 3136.0 + }, + "rows_with_hash_ids": 512, + "semantics": "arrival-ordered, infinite-capacity/no-eviction upper bound; not an observed KV-cache hit rate", + "total_blocks": 25607, + "unique_blocks": 23232 + }, + "request_count": 512, + "sampling_u": { + "max": 0.9979913161468553, + "min": 0.0017297900650431776, + "p10": 0.10365756051317851, + "p25": 0.22877823992809823, + "p50": 0.4667483893139273, + "p75": 0.7338360613813898, + "p90": 0.8969698510725516, + "p95": 0.9558957048665733, + "p99": 0.991130680217395 + } + } + }, + "trace": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/chat_w20260311_1000.jsonl", + "trace_sha256": "f539f38eb0ee0f750e3c23ff47df6eed3faf723a25f1444d55665a85871750b9", + "window_id": "chat_w20260311_1000" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/capture_trace_routing.py b/runs/frontier-qwen30-vllm020-profile-v1/capture_trace_routing.py new file mode 100644 index 0000000..e1f4b99 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/capture_trace_routing.py @@ -0,0 +1,272 @@ +#!/usr/bin/env python3 +"""Capture exact Qwen3 routed-expert IDs from vLLM 0.20 on trace prompts.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import subprocess +from pathlib import Path +from typing import Any + +import numpy as np +import torch +import vllm + + +VLLM_VERSION = "0.20.0" +VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +NUM_EXPERTS = 128 +TOP_K = 8 +NUM_LAYERS = 48 + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--vllm-source", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--fixture", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--routes", type=Path, required=True) + parser.add_argument("--decode-override", type=int) + return parser.parse_args() + + +def git_head(repo: Path) -> str: + return subprocess.check_output( + ["git", "-C", str(repo), "rev-parse", "HEAD"], text=True + ).strip() + + +def sha256(path: Path) -> str: + return hashlib.sha256(path.read_bytes()).hexdigest() + + +def common_prefix(left: list[int], right: list[int]) -> int: + count = 0 + for lhs, rhs in zip(left, right): + if lhs != rhs: + break + count += 1 + return count + + +def distribution(counts: np.ndarray) -> dict[str, Any]: + values = counts.astype(np.float64) + total = float(values.sum()) + mean = float(values.mean()) + probabilities = values[values > 0] / total + entropy = float(-(probabilities * np.log2(probabilities)).sum()) + variance = float(((values - mean) ** 2).mean()) + ordered = np.sort(values) + gini = float( + 2.0 * np.dot(np.arange(1, len(values) + 1), ordered) + / (len(values) * total) + - (len(values) + 1) / len(values) + ) + hottest = np.argsort(values)[-8:][::-1] + return { + "total_routed_tokens": int(total), + "tokens_per_expert_mean": mean, + "load_cv": math.sqrt(variance) / mean, + "load_gini": gini, + "load_entropy_bits": entropy, + "min_load_ratio": float(values.min() / mean), + "max_load_ratio": float(values.max() / mean), + "expert_utilization": float(np.count_nonzero(values) / len(values)), + "hottest_experts": [int(value) for value in hottest], + "hottest_counts": [int(values[value]) for value in hottest], + "counts": counts.astype(int).tolist(), + } + + +def phase_summary(routes: list[np.ndarray]) -> dict[str, Any]: + counts = np.zeros(NUM_EXPERTS, dtype=np.int64) + per_layer = np.zeros((NUM_LAYERS, NUM_EXPERTS), dtype=np.int64) + token_count = 0 + for route in routes: + token_count += route.shape[0] + counts += np.bincount(route.reshape(-1), minlength=NUM_EXPERTS) + for layer in range(NUM_LAYERS): + per_layer[layer] += np.bincount( + route[:, layer, :].reshape(-1), minlength=NUM_EXPERTS + ) + return { + "token_count": token_count, + "all_layers": distribution(counts), + "per_layer": [distribution(row) for row in per_layer], + } + + +def main() -> None: + args = parse_args() + if vllm.__version__ != VLLM_VERSION: + raise SystemExit(f"expected vLLM {VLLM_VERSION}, got {vllm.__version__}") + source_head = git_head(args.vllm_source) + if source_head != VLLM_COMMIT: + raise SystemExit(f"expected vLLM source {VLLM_COMMIT}, got {source_head}") + rows = [json.loads(line) for line in args.fixture.read_text().splitlines() if line] + if not rows: + raise SystemExit("empty routing fixture") + requested_decode = [ + args.decode_override + if args.decode_override is not None + else int(row["output_length"]) + for row in rows + ] + if any(value <= 0 for value in requested_decode): + raise SystemExit("all requested decode lengths must be positive") + + from vllm import LLM, SamplingParams + + llm = LLM( + model=str(args.model), + dtype="bfloat16", + tensor_parallel_size=1, + max_model_len=16384, + max_num_batched_tokens=8192, + max_num_seqs=64, + gpu_memory_utilization=0.90, + enable_chunked_prefill=True, + enable_prefix_caching=True, + enable_return_routed_experts=True, + attention_backend="FLASH_ATTN", + disable_log_stats=False, + ) + sampling = [ + SamplingParams(temperature=0, min_tokens=value, max_tokens=value) + for value in requested_decode + ] + conversations = [ + [{"role": "user", "content": row["prompt"]}] for row in rows + ] + outputs = llm.chat(conversations, sampling_params=sampling, use_tqdm=False) + if len(outputs) != len(rows): + raise SystemExit(f"expected {len(rows)} outputs, got {len(outputs)}") + + prompt_tokens_by_chat: dict[str, list[int]] = {} + prefill_routes: list[np.ndarray] = [] + decode_routes: list[np.ndarray] = [] + raw_routes: dict[str, np.ndarray] = {} + request_summaries = [] + for row, output, decode_tokens in zip(rows, outputs, requested_decode): + completion = output.outputs[0] + routed = completion.routed_experts + if routed is None: + raise SystemExit(f"row {row['row_id']} returned no routed experts") + routed = np.asarray(routed) + prompt_tokens = list(output.prompt_token_ids) + generated_tokens = list(completion.token_ids) + expected = len(prompt_tokens) + len(generated_tokens) - 1 + if routed.shape != (expected, NUM_LAYERS, TOP_K): + raise SystemExit( + f"row {row['row_id']} routes {routed.shape}, expected " + f"{(expected, NUM_LAYERS, TOP_K)}" + ) + if routed.min() < 0 or routed.max() >= NUM_EXPERTS: + raise SystemExit(f"row {row['row_id']} returned invalid expert IDs") + prefill = routed[: len(prompt_tokens)] + decode = routed[len(prompt_tokens) :] + if decode.shape[0] != decode_tokens - 1: + raise SystemExit(f"row {row['row_id']} decode route length mismatch") + prefill_routes.append(prefill) + decode_routes.append(decode) + raw_routes[f"row_{row['row_id']}"] = routed.astype(np.int16) + prompt_tokens_by_chat[str(row["chat_id"])] = prompt_tokens + request_summaries.append( + { + "fixture_index": row["fixture_index"], + "row_id": row["row_id"], + "turn": row["turn"], + "input_length_trace": row["input_length"], + "prompt_tokens_vllm": len(prompt_tokens), + "chat_wrapper_delta": len(prompt_tokens) - int(row["input_length"]), + "generated_tokens": len(generated_tokens), + "requested_decode_tokens": decode_tokens, + "routed_shape": list(routed.shape), + "prompt_sha256": row["prompt_sha256"], + "trace_hash_blocks": len(row["hash_ids"]), + } + ) + + prefix_pairs = [] + by_chat = {str(row["chat_id"]): row for row in rows} + for child in rows: + parent = by_chat.get(str(child["parent_chat_id"])) + if parent is None: + continue + parent_tokens = prompt_tokens_by_chat[str(parent["chat_id"])] + child_tokens = prompt_tokens_by_chat[str(child["chat_id"])] + prefix_pairs.append( + { + "parent_row_id": parent["row_id"], + "child_row_id": child["row_id"], + "trace_hash_common_prefix_blocks": common_prefix( + parent["hash_ids"], child["hash_ids"] + ), + "vllm_token_common_prefix": common_prefix(parent_tokens, child_tokens), + "vllm_full_common_blocks_16": common_prefix( + parent_tokens, child_tokens + ) + // 16, + } + ) + + args.routes.parent.mkdir(parents=True, exist_ok=True) + np.savez_compressed(args.routes, **raw_routes) + payload = { + "schema_version": "qwen30_vllm020_trace_routing.v1", + "environment": { + "vllm_version": vllm.__version__, + "vllm_source_commit": source_head, + "torch_version": torch.__version__, + "torch_cuda": torch.version.cuda, + "gpu": torch.cuda.get_device_name(0), + "model": str(args.model), + "dtype": "bfloat16", + "tensor_parallel_size": 1, + "max_num_batched_tokens": 8192, + "max_num_seqs": 64, + "prefix_caching": True, + "chunked_prefill": True, + "attention_backend": "FLASH_ATTN", + }, + "capture_contract": { + "api": "LLM.chat", + "enable_return_routed_experts": True, + "route_shape": "[prompt_tokens + generated_tokens - 1, layers, topk]", + "decode_policy": ( + f"fixed_override_{args.decode_override}" + if args.decode_override is not None + else "exact_trace_output_length" + ), + "contains_prompt_text": False, + "fixture_sha256": sha256(args.fixture), + "routes_npz": str(args.routes), + }, + "requests": request_summaries, + "prefix_pairs": prefix_pairs, + "phases": { + "prefill": phase_summary(prefill_routes), + "decode": phase_summary(decode_routes), + }, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print( + json.dumps( + { + "requests": len(rows), + "prefill_tokens": payload["phases"]["prefill"]["token_count"], + "decode_tokens": payload["phases"]["decode"]["token_count"], + "prefix_pairs": prefix_pairs, + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/experiment-card.md b/runs/frontier-qwen30-vllm020-profile-v1/experiment-card.md new file mode 100644 index 0000000..4f38358 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/experiment-card.md @@ -0,0 +1,58 @@ +# 实验 EXP-SIMFID-Q30-P020：同栈 per-TP operator profile 能否恢复 Frontier ranking？ + +> **状态：** 完成；H1 rejected，结果支持 H2（2026-07-16） + +## Claim 与决策 + +- **Parent claim：** Frontier 在 Qwen3-30B-A3B 的 TP×MNS surface 上出现 25.63% selection regret，原因可能不是 simulator scheduler 本身，而是 execution profile 与真实 serving stack 不一致。 +- **目的：** 区分 profile provenance mismatch 与 simulator composition/schema mismatch。 +- **Competing hypotheses：** H1：换成与真实 serving 相同的 per-TP operator profiles 后，profile-only ranking 达到 regret ≤5% 且 Kendall τ-b ≥0.8；H2：即使 operator profile 同栈，Frontier 的 operator composition 或 mixed-state abstraction 仍无法保持 ranking。 +- **事前预测：** 若 H1 成立，新 profile-only top set 应从旧结果的 TP4 移到真实 top family TP2，且不需要 end-to-end scalar calibration；若 H2 成立，operator microbench 单项可对齐，但组成后的 config rank 仍错，残差应随 phase、batch state 或 TP action 系统变化。 +- **判定规则：** `P-020` profile-only regret ≤5% 且 τ-b ≥0.8 → 支持 H1；否则进入 stage/action-conditioned residual 分解，禁止用本 surface 的 E2E measurement 拟合 scale。 + +## Setup + +- **自变量：** execution profile root：`P-old`（历史 profile）与 `P-020`（本实验冻结 profile）。`P-cal` 历史 per-TP scalar calibration 只作上界参照。 +- **控制变量：** Frontier simulator code/commit、trace fixtures、KV capacity、SLO、12-cell config surface、random seed 与 analysis code保持完全相同。 +- **System context：** Qwen3-30B-A3B，BF16 weights/activations/KV；community vLLM 0.20.0；dash0 8×NVIDIA H20；TP∈{1,2,4}，DP=PP=EP=1；MBT=8192；MNS∈{8,16,32,64}；prefix caching、chunked prefill 与 async scheduling均启用。 +- **真实 runtime contract：** engine log 已确认 TP1/2/4 均为 FlashAttention 3 attention + FlashInfer CUTLASS unquantized MoE；TP2/4 使用 FlashInfer TRT-LLM all-reduce。旧 profile log 明确为 vLLM 0.11.1，attention backend 为 FlashInfer。 +- **Workload 或 trace：** 历史 ground truth `chat_w20260311_1000`，source 600 s window、replay time scale 0.1、input 0–8192、fixed output 128、prefix cache on。当前本地 recovered artifact 无法重建 exact source requests；本次只替换 frozen simulator profile，不重跑或改写 ground truth。 +- **Baselines：** old profile-only；new `P-020` profile-only；historical frozen per-TP calibrated。 +- **Metrics：** real/sim capacity per GPU、selected top set、worst selected-config regret、Kendall τ-b、informative-pair direction、逐 TP operator-time ratio、profile coverage/extrapolation ratio。 + +## 预期产物与 review + +- **实际数据：** TP1/2/4 attention、KV-cache update、linear、MoE、collective raw profiles；冻结 manifest/sha256；92/92 new profile-only runs；old/new/calibrated comparison；scheduler graph-mode 与 native MoE routing diagnostics。 +- **Figure：** `../../docs/assets/simulator-fidelity/qwen30-vllm020-profile-ablation.png`；四个 panels 分别展示完整 ranking、routing skew、phase-dependent graph mode 和 held-out P1 boundary accuracy。图中只使用实测或冻结分析结果。 +- **人工 review：** 已批准。用户明确要求推进 5 个步骤，并要求 smoke 通过后直接完整运行。 +- **Review 意见：** 不能把 TP–MNS 耦合拆成互不相关问题；profile 只冻结 execution counterfactual 的证据，最终仍以完整 config ranking 判断。 + +## 复现信息 + +- **Code：** AITuner 当前 branch `codex/fidelity-prefix-pilot-20260714`；Frontier canonical commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e`；vLLM commit/tag `88d34c6409e9fb3c7b8ca0c04756f061d2099eb1` / `0.20.0`。 +- **Environment：** `/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1`；hardware/driver/package freeze 写入每个 profile artifact。 +- **产物路径：** remote `/home/admin/cpfs/wjh/frontier_qwen30_vllm020_profiles/`; local harvest `runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/`。 +- **已知 deviation：** 历史 real ground truth 与本次 profile 都来自 dash0；早期文档中的主机 provenance 标注错误。driver 从历史 570.133.20 漂移到当前 580.95.05，因此结果仍不能解释为严格的逐 kernel latency 复现，只能作为同一 H20 主机上 profile-stack alignment 的对照。 + +## 结果 + +- **Step 1—同栈固定：** dash0 8×H20；Qwen3-30B-A3B BF16；community vLLM `0.20.0+cu129` / source commit `88d34c6409e9fb3c7b8ca0c04756f061d2099eb1`。真实 serving 与 profiles 使用同一主机和 model family；历史 driver 570.133.20 到当前 580.95.05 的漂移保留为限制。 +- **Step 2—per-TP profiles：** profile-v2 冻结 attention 132 rows、直接测量 fused-mixed diagnostic 30 rows、linear 36 rows、MoE 72 rows、all-reduce 24 rows。attention/linear/MoE 通过单 H20 local-shard shape 表达 TP；all-reduce 使用真实 TP2/4 ranks。base comparison 保持 historical analytical CC backend，measured all-reduce 仅作独立诊断。 +- **Step 3—trace execution coverage：** 6 个 P1 cells 共 161,161 scheduler/model steps：pure decode 151,471（100% FULL graph），pure prefill 224（192 NONE、30 PIECEWISE、2 FULL），true mixed 9,466（8,623 NONE、842 PIECEWISE、1 FULL）。exact-trace 8-request native routing probe 表明实际 median max/mean expert load 为 prefill 6.12、decode 5.64，而 Frontier fixed prior 为 1.84；actual-vs-prior Pearson median 接近 0。该小 cohort 用于机制否证，不作为 routing population estimate。 +- **Step 4—冻结 profile-only：** profile-v1 因 true-mixed schema 缺少 `attn_decode_in_mixed` 在第一个 mixed batch fail。profile-v2 使用 same-TP pure prefill/decode ratio 做 total-conserving compatibility split，同时保留直接实测 fused total。两个 46-run CPU shards 均 PASS，92/92 runs 无 crash，总 simulator runtime 2,394.9 s（0.665 CPU-hour）。 +- **Step 5—ranking comparison：** real optimum 为 TP2/MNS32。old profile-only top set 为 TP4/MNS32/64，τ-b=0、worst regret=25.63%；new vLLM 0.20 profile-only 在所有 92 anchors 上都 SLO-infeasible，12 configs 全并列，τ-b=0、exact pair sign=7.58%、worst tie-break regret=60.91%；historical frozen per-TP calibration top set 为 TP2/MNS32/64，τ-b=0.9668、worst regret=0.76%。在 held-out P1 12 boundary labels 上，new profile-only accuracy 41.67%、pass-rate MAE 69.84%，也没有优于 old profile-only；calibration 分别为 83.33% 和 11.75%。 +- **Same-state delta：** 在 `tp1_mns64` 的 batch 0/layer 0、profile-dependent trajectories 尚未分叉时，new/old layer component sum 已为 2.70×；MoE grouped GEMM 3.76×、RoPE 7.39×、attention prefill 2.67×。误差不是可由一个 version correction 解释的一致 scale。 +- **异常与 schema gap：** Frontier 原生 attention profiler 只实现 `FLASHINFER`/`NO_OP`，不能调用真实 serving 的 `FLASH_ATTN`；其 MoE wrapper声明并硬编码 vLLM 0.10.x API。vLLM 0.20.0 的 FA3 mixed prefill+decode 是一个 fused varlen call，而 Frontier schema要求 `attn_prefill` 和 `attn_decode_in_mixed` 两项。兼容 split 是归因，不是观测。 +- **Interpretation：** H1 被否证。同栈 isolated operator profile 不等价于真实 execution counterfactual。TP 与 MNS 共同改变 local shape、collective、queue/batch composition、phase mixture、routing 和 CUDA graph regime；static operator sum 在 scheduler 推进后产生 feedback amplification。历史 per-TP calibration 是 composition 之后的外部 E2E scale，能吸收这些残差，但不是 Frontier 原生 per-TP profiling 或机制解释。 +- **Claim update：** 支持 H2。新的 research target 是验证 execution-context-conditioned composition，而不是继续把更多 isolated kernel rows 当作 fidelity 的充分条件。 +- **下一步：** 分别做 graph-conditioned step profile、fused mixed total、trace-conditioned routing 和 measured collective 四个单变量消融，再回到完整 TP×MNS joint surface 评测；frozen calibration 仅作上界，不参与 mechanism fitting。 + +## 产物索引 + +- Frozen profile：`frozen/profile-v2/manifest.json` +- Full-surface comparison：`artifacts/qwen30-s2-profile-ablation.json` +- P1 boundary comparison：`artifacts/qwen30-p1-profile-ablation.json` +- Scheduler/graph-mode summary：`artifacts/qwen30-p1b-opprof-summary.json` +- Native routing mismatch：`artifacts/qwen30-routing-mismatch.json` +- Same-state operator delta：`artifacts/qwen30-initial-op-trace-delta.json` +- Figure script：`plot_profile_ablation.py` diff --git a/runs/frontier-qwen30-vllm020-profile-v1/extract_routing_fixture.py b/runs/frontier-qwen30-vllm020-profile-v1/extract_routing_fixture.py new file mode 100644 index 0000000..2021ddc --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/extract_routing_fixture.py @@ -0,0 +1,162 @@ +#!/usr/bin/env python3 +"""Extract a prompt-bearing routing fixture while emitting a prompt-free manifest.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +from pathlib import Path +from typing import Any + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--trace", type=Path, required=True) + parser.add_argument("--row-ids", type=int, nargs="+", required=True) + parser.add_argument("--parent-of", type=int, nargs="*", default=[]) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + return parser.parse_args() + + +def common_prefix(left: list[Any], right: list[Any]) -> int: + count = 0 + for lhs, rhs in zip(left, right): + if lhs != rhs: + break + count += 1 + return count + + +def main() -> None: + args = parse_args() + target_ids = set(args.row_ids) + parent_targets = set(args.parent_of) + if not parent_targets.issubset(target_ids): + raise SystemExit("--parent-of must be a subset of --row-ids") + + trace_digest = hashlib.sha256() + by_row: dict[int, dict[str, Any]] = {} + by_chat_id: dict[str, dict[str, Any]] = {} + with args.trace.open("rb") as handle: + row_id = 0 + while True: + offset = handle.tell() + line = handle.readline() + if not line: + break + trace_digest.update(line) + row = json.loads(line) + meta = { + "row_id": row_id, + "offset": offset, + "chat_id": str(row.get("chat_id")), + "parent_chat_id": str(row.get("parent_chat_id")), + "turn": int(row.get("turn", 1)), + "input_length": int(row["input_length"]), + "output_length": int(row["output_length"]), + "hash_ids": row.get("hash_ids") or [], + } + by_chat_id[meta["chat_id"]] = meta + if row_id in target_ids: + by_row[row_id] = meta + row_id += 1 + + missing = target_ids - set(by_row) + if missing: + raise SystemExit(f"missing target row IDs: {sorted(missing)}") + parent_rows: list[dict[str, Any]] = [] + for row_id in args.parent_of: + parent_id = by_row[row_id]["parent_chat_id"] + parent = by_chat_id.get(parent_id) + if parent is None: + raise SystemExit(f"row {row_id} parent chat {parent_id} is absent") + parent_rows.append(parent) + + # Put parents first so online prefix caching can materialize their shared + # blocks before descendants are admitted. + ordered_meta = parent_rows + [by_row[row_id] for row_id in args.row_ids] + if len({row["row_id"] for row in ordered_meta}) != len(ordered_meta): + raise SystemExit("fixture rows must be unique") + + output_rows: list[dict[str, Any]] = [] + with args.trace.open("rb") as handle: + for fixture_index, meta in enumerate(ordered_meta): + handle.seek(meta["offset"]) + source = json.loads(handle.readline()) + prompt = source.get("prompt") + if not isinstance(prompt, str) or not prompt: + raise SystemExit(f"row {meta['row_id']} has no prompt") + output_rows.append( + { + "fixture_index": fixture_index, + "row_id": str(meta["row_id"]), + "prompt": prompt, + "prompt_sha256": hashlib.sha256(prompt.encode()).hexdigest(), + "input_length": meta["input_length"], + "output_length": meta["output_length"], + "turn": meta["turn"], + "chat_id": meta["chat_id"], + "parent_chat_id": meta["parent_chat_id"], + "hash_ids": meta["hash_ids"], + } + ) + + args.output.parent.mkdir(parents=True, exist_ok=True) + with args.output.open("w") as handle: + for row in output_rows: + handle.write(json.dumps(row, sort_keys=True) + "\n") + + pair_coverage = [] + by_chat = {row["chat_id"]: row for row in output_rows} + for child in output_rows: + parent = by_chat.get(child["parent_chat_id"]) + if parent is None: + continue + pair_coverage.append( + { + "parent_row_id": parent["row_id"], + "child_row_id": child["row_id"], + "parent_turn": parent["turn"], + "child_turn": child["turn"], + "trace_hash_common_prefix_blocks": common_prefix( + parent["hash_ids"], child["hash_ids"] + ), + } + ) + fixture_digest = hashlib.sha256(args.output.read_bytes()).hexdigest() + manifest = { + "schema_version": "qwen30_trace_routing_fixture.v1", + "source_trace": str(args.trace.resolve()), + "source_trace_sha256": trace_digest.hexdigest(), + "fixture_path": str(args.output.resolve()), + "fixture_sha256": fixture_digest, + "contains_prompt_text": False, + "request_count": len(output_rows), + "rows": [ + { + key: row[key] + for key in ( + "fixture_index", + "row_id", + "prompt_sha256", + "input_length", + "output_length", + "turn", + "chat_id", + "parent_chat_id", + ) + } + | {"trace_hash_blocks": len(row["hash_ids"])} + for row in output_rows + ], + "prefix_pairs": pair_coverage, + } + args.manifest.parent.mkdir(parents=True, exist_ok=True) + args.manifest.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps({"requests": len(output_rows), "prefix_pairs": pair_coverage})) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/extract_trace_profile_support.py b/runs/frontier-qwen30-vllm020-profile-v1/extract_trace_profile_support.py new file mode 100644 index 0000000..e3e37c6 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/extract_trace_profile_support.py @@ -0,0 +1,215 @@ +#!/usr/bin/env python3 +"""Freeze the exact request cohort and its operator-profile support. + +This calls AITuner's production trace loader, including its input-length +filter, uniform max-request downsampling, output override, and sampling-u +threshold semantics. Prefix reuse below is a no-eviction upper bound; the +actual cache state remains scheduler/config dependent. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from pathlib import Path +from typing import Any, Iterable + +from aituner.spec import load_study_spec +from aituner.trace import load_trace_requests, select_requests_for_threshold + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--study", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--cohort-output", type=Path, required=True) + parser.add_argument( + "--thresholds", + type=float, + nargs="+", + default=[0.125, 0.25, 0.5, 1.0], + ) + return parser.parse_args() + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def nearest_rank(values: Iterable[float], percentile: float) -> float: + ordered = sorted(values) + if not ordered: + return 0.0 + index = min( + len(ordered) - 1, + max(0, math.ceil(percentile / 100.0 * len(ordered)) - 1), + ) + return float(ordered[index]) + + +def distribution(values: list[float]) -> dict[str, float]: + return { + "min": min(values, default=0.0), + "p10": nearest_rank(values, 10), + "p25": nearest_rank(values, 25), + "p50": nearest_rank(values, 50), + "p75": nearest_rank(values, 75), + "p90": nearest_rank(values, 90), + "p95": nearest_rank(values, 95), + "p99": nearest_rank(values, 99), + "max": max(values, default=0.0), + } + + +def prefix_upper_bound(requests: list[Any], block_size: int) -> dict[str, Any]: + seen: set[Any] = set() + total_blocks = 0 + reusable_blocks = 0 + leading_reusable_blocks: list[float] = [] + reusable_tokens: list[float] = [] + rows_with_hashes = 0 + for request in requests: + hashes = request.metadata.get("hash_ids") + if not isinstance(hashes, list): + continue + rows_with_hashes += 1 + total_blocks += len(hashes) + reusable = sum(hash_id in seen for hash_id in hashes) + reusable_blocks += reusable + leading = 0 + for hash_id in hashes: + if hash_id not in seen: + break + leading += 1 + leading_reusable_blocks.append(float(leading)) + reusable_tokens.append( + float(min(request.prompt_tokens_hint or 0, leading * block_size)) + ) + seen.update(hashes) + return { + "semantics": ( + "arrival-ordered, infinite-capacity/no-eviction upper bound; " + "not an observed KV-cache hit rate" + ), + "rows_with_hash_ids": rows_with_hashes, + "total_blocks": total_blocks, + "unique_blocks": len(seen), + "any_position_reusable_block_ratio": ( + reusable_blocks / total_blocks if total_blocks else 0.0 + ), + "leading_reusable_blocks_per_request": distribution( + leading_reusable_blocks + ), + "leading_reusable_tokens_per_request": distribution(reusable_tokens), + } + + +def summarize(requests: list[Any], block_size: int) -> dict[str, Any]: + input_lengths = [float(request.prompt_tokens_hint or 0) for request in requests] + output_lengths = [ + float(request.completion_tokens_hint or 0) for request in requests + ] + hash_counts = [ + float(len(request.metadata.get("hash_ids") or [])) for request in requests + ] + arrivals = [request.arrival_s for request in requests] + interarrivals = [ + max(0.0, arrivals[index] - arrivals[index - 1]) + for index in range(1, len(arrivals)) + ] + return { + "request_count": len(requests), + "input_tokens": distribution(input_lengths), + "output_tokens": distribution(output_lengths), + "hash_blocks_per_request": distribution(hash_counts), + "interarrival_s": distribution(interarrivals), + "sampling_u": distribution([request.sampling_u for request in requests]), + "multi_turn_fraction": ( + sum( + isinstance(request.metadata.get("turn"), (int, float)) + and request.metadata["turn"] > 1 + for request in requests + ) + / len(requests) + if requests + else 0.0 + ), + "prefix_reuse_upper_bound": prefix_upper_bound(requests, block_size), + } + + +def cohort_row(request: Any) -> dict[str, Any]: + return { + "row_id": request.row_id, + "arrival_s": request.arrival_s, + "sampling_u": request.sampling_u, + "input_tokens": request.prompt_tokens_hint, + "output_tokens": request.completion_tokens_hint, + "hash_ids": request.metadata.get("hash_ids"), + "turn": request.metadata.get("turn"), + "parent_chat_id": request.metadata.get("parent_chat_id"), + "type": request.metadata.get("type"), + } + + +def main() -> None: + args = parse_args() + study = load_study_spec(args.study) + window, cohort = load_trace_requests(study, study_spec_path=args.study) + block_size = int(window.source_payload.get("block_size") or 1) + + args.cohort_output.parent.mkdir(parents=True, exist_ok=True) + with args.cohort_output.open("w", encoding="utf-8") as handle: + for request in cohort: + handle.write(json.dumps(cohort_row(request), sort_keys=True) + "\n") + + payload = { + "schema_version": "qwen30_trace_profile_support.v1", + "study": str(args.study.resolve()), + "study_sha256": sha256_file(args.study), + "trace": str(window.trace_path), + "trace_sha256": sha256_file(window.trace_path), + "window_id": window.window_id, + "block_size": block_size, + "loader_contract": { + "input_length_filter": { + "min": study.trace.input_length_filter.min_input_tokens, + "max": study.trace.input_length_filter.max_input_tokens, + } + if study.trace.input_length_filter is not None + else None, + "completion_tokens_override": study.trace.completion_tokens_override, + "max_requests_per_probe": study.trace.max_requests_per_probe, + "replay_time_scale": study.trace.replay_time_scale, + "ordering": "arrival_s", + "downsampling": "AITuner _downsample_requests before sampling_u threshold", + }, + "full_downsampled_cohort": summarize(cohort, block_size), + "threshold_cohorts": { + str(threshold): summarize( + select_requests_for_threshold(cohort, threshold=threshold), + block_size, + ) + for threshold in args.thresholds + }, + "limits": [ + "Trace length/hash support does not determine dynamic decode or mixed batch shapes.", + "Those shapes depend jointly on arrival history, SLO pressure, TP execution time, MNS, chunking, and KV eviction.", + "MoE expert routing is not present in the trace and must be measured from model execution.", + ], + } + payload["cohort_output"] = str(args.cohort_output.resolve()) + payload["cohort_sha256"] = sha256_file(args.cohort_output) + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps(payload, indent=2, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/fleet.toml b/runs/frontier-qwen30-vllm020-profile-v1/fleet.toml new file mode 100644 index 0000000..e560549 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/fleet.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-qwen30-vllm020-profile-v1/fleet-state" +artifacts_dir = "runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 1024 +gpu_free_utilization_pct = 10 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-qwen30-vllm020-profile-v1" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-qwen30-vllm020-profile-v1" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/freeze_frontier_profiles.py b/runs/frontier-qwen30-vllm020-profile-v1/freeze_frontier_profiles.py new file mode 100644 index 0000000..c6f8a2a --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/freeze_frontier_profiles.py @@ -0,0 +1,822 @@ +#!/usr/bin/env python3 +"""Freeze vLLM 0.20 microprofiles into Frontier-compatible CSV inputs.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +import re +import shutil +from pathlib import Path +from typing import Any + + +STAT_NAMES = ("min", "max", "mean", "median", "std") +ATTENTION_OPS = ( + "attn_input_reshape", + "attn_kv_cache_save", + "attn_prefill", + "attn_decode", + "attn_output_reshape", +) +ATTENTION_METADATA = ( + "n_embd", + "n_q_head", + "n_kv_head", + "block_size", + "num_tensor_parallel_workers", + "max_model_len", + "batch_size", + "prefill_chunk_size", + "kv_cache_size", + "is_prefill", + "attention_backend", + "is_mixed_batch", + "mode", + "seq_lens", + "total_tokens", + "max_seq_len", + "min_seq_len", + "avg_seq_len", + "equal_seq_len", + "seq_len_variance", + "seq_len_std", + "seq_len_cv", + "is_chunked_prefill_sample", + "chunk_start_token", + "chunk_end_token", + "total_prefill_tokens", + "profiling_precision", + "model_arch", + "quant_signature", + "measurement_type", + "is_true_mixed_batch", + "prefill_seq_lens", + "prefill_kv_cache_sizes", + "decode_kv_cache_sizes", + "num_prefill_seqs", + "num_decode_seqs", + "decode_batch_size", + "total_batch_size", + "total_decode_tokens", + "decode_avg_kv_cache_size", + "batch_composition_ratio", + "batch_spec", + "projection_policy", +) +MOE_OPS = ( + "moe_gating_linear", + "moe_gating_routing_topk", + "moe_shuffling", + "moe_grouped_gemm", +) +MOE_METADATA = ( + "num_tokens", + "num_experts", + "num_experts_per_device", + "expert_parallel_size", + "routing_runtime_path", + "routing_assignment_policy", + "routing_weight_policy", + "routing_uses_router_logits", + "gating_runtime_context", + "gating_runtime_context_impl", + "router_topk", + "hidden_dim", + "expert_hidden_dim", + "use_gated", + "num_tensor_parallel_workers", + "total_routed_tokens", + "model_expansion_ratio", + "tokens_per_expert_avg", + "tokens_to_experts_ratio", + "expert_utilization", + "min_load_ratio", + "load_imbalance_cv", + "max_load_ratio", + "load_entropy", + "load_gini_coefficient", + "load_distribution", + "seed", + "moe_grouped_gemm_backend", + "measurement_type", + "profiling_precision", + "model_arch", + "quant_signature", + "router_median_nonadditivity_ratio", + "projection_policy", +) + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--linear", type=Path, required=True) + parser.add_argument("--attention", type=Path, nargs="+", required=True) + parser.add_argument("--moe", type=Path, required=True) + parser.add_argument("--router", type=Path, required=True) + parser.add_argument("--allreduce", type=Path, nargs=2) + parser.add_argument("--allreduce-frozen", type=Path) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument( + "--measurement-type", + choices=("CUDA_EVENT", "KERNEL_ONLY"), + default="CUDA_EVENT", + ) + parser.add_argument( + "--frontier-commit", + default="d9cfeb6d8791fbf2f295dd9744c56a666171776e", + ) + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def load_json(path: Path) -> dict[str, Any]: + return json.loads(path.read_text()) + + +def stat_columns(prefix: str, stats: dict[str, float]) -> dict[str, float]: + return {f"time_stats.{prefix}.{name}": float(stats[name]) for name in STAT_NAMES} + + +def zero_stat_columns(prefix: str) -> dict[str, float]: + return {f"time_stats.{prefix}.{name}": 0.0 for name in STAT_NAMES} + + +def attention_core_stats(raw: dict[str, Any]) -> dict[str, float]: + # vLLM's benchmark result exports aggregate mean but not the raw samples. + # Preserve that mean as Frontier's training target and record the proxy in + # the manifest rather than inventing an unobserved median. + return { + "min": 1000.0 * float(raw["min_time"]), + "max": 1000.0 * float(raw["max_time"]), + "mean": 1000.0 * float(raw["mean_time"]), + "median": 1000.0 * float(raw["mean_time"]), + "std": 1000.0 * float(raw["std_time"]), + } + + +def kv_update_stats(raw: dict[str, Any]) -> dict[str, float]: + stats = raw["kv_cache_update_time"] + return {name: float(stats[f"{name}_ms"]) for name in STAT_NAMES} + + +def parse_size(value: str, suffix: str) -> int: + return int(value) * (1024 if suffix == "k" else 1) + + +def parse_batch_spec(spec: str) -> list[tuple[int, int]]: + requests: list[tuple[int, int]] = [] + pattern = re.compile(r"^(?:(\d+))?q(\d+)(k?)(?:s(\d+)(k?))?$") + for segment in spec.split("_"): + match = pattern.match(segment) + if match is None: + raise ValueError(f"invalid vLLM batch spec: {spec}") + count = int(match.group(1) or 1) + query = parse_size(match.group(2), match.group(3)) + kv = ( + parse_size(match.group(4), match.group(5)) + if match.group(4) + else query + ) + requests.extend([(query, kv)] * count) + return requests + + +def write_csv(path: Path, fieldnames: list[str], rows: list[dict[str, Any]]) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + with path.open("w", newline="") as handle: + writer = csv.DictWriter( + handle, + fieldnames=fieldnames, + extrasaction="raise", + lineterminator="\n", + ) + writer.writeheader() + writer.writerows(rows) + + +def freeze_attention( + inputs: list[Path], output: Path, *, measurement_type: str +) -> tuple[int, int, list[str]]: + rows: list[dict[str, Any]] = [] + mixed_rows: list[dict[str, Any]] = [] + seen_tps: set[int] = set() + raw_by_tp: dict[int, list[dict[str, Any]]] = {} + for path in inputs: + payload = load_json(path) + if payload.get("schema_version") != "qwen30_vllm020_flashattn_raw.v1": + raise ValueError(f"unexpected attention schema in {path}") + if payload["environment"].get("vllm_version") != "0.20.0": + raise ValueError(f"unexpected vLLM version in {path}") + expected_method = ( + "record_function" if measurement_type == "KERNEL_ONLY" else "cuda_event" + ) + if payload["environment"].get("profile_method", "cuda_event") != expected_method: + raise ValueError( + f"attention profile method mismatch in {path}: expected {expected_method}" + ) + for raw in payload["rows"]: + if raw.get("error") is not None: + raise ValueError(f"failed attention row in {path}: {raw['error']}") + tp = int(raw["tensor_parallel_size"]) + seen_tps.add(tp) + raw_by_tp.setdefault(tp, []).append(raw) + + def pure_reference_ms( + tp: int, requests: list[tuple[int, int]], *, decode_phase: bool + ) -> float: + candidates: list[tuple[list[tuple[int, int]], float]] = [] + for candidate in raw_by_tp[tp]: + parsed = parse_batch_spec(candidate["config"]["batch_spec"]) + is_decode = all(query == 1 for query, _ in parsed) + if is_decode != decode_phase: + continue + if decode_phase and not is_decode: + continue + if not decode_phase and any(query == 1 for query, _ in parsed): + continue + candidates.append((parsed, 1000.0 * float(candidate["mean_time"]))) + for parsed, mean_ms in candidates: + if parsed == requests: + return mean_ms + if not decode_phase: + raise ValueError(f"no exact pure prefill reference for TP{tp}: {requests}") + if len({kv for _, kv in requests}) != 1: + raise ValueError(f"decode interpolation requires one KV length: {requests}") + target_batch = len(requests) + target_kv = requests[0][1] + points = sorted( + (parsed[0][1], mean_ms) + for parsed, mean_ms in candidates + if len(parsed) == target_batch + and len({kv for _, kv in parsed}) == 1 + ) + if not points: + raise ValueError(f"no pure decode reference for TP{tp}: {requests}") + if target_kv <= points[0][0]: + return points[0][1] + if target_kv >= points[-1][0]: + return points[-1][1] + for (left_kv, left_ms), (right_kv, right_ms) in zip(points, points[1:]): + if left_kv <= target_kv <= right_kv: + fraction = (target_kv - left_kv) / (right_kv - left_kv) + return left_ms + fraction * (right_ms - left_ms) + raise AssertionError("unreachable decode interpolation") + + for tp in sorted(raw_by_tp): + for raw in raw_by_tp[tp]: + spec = raw["config"]["batch_spec"] + requests = parse_batch_spec(spec) + prefill = [(q, kv) for q, kv in requests if q > 1] + decode = [(q, kv) for q, kv in requests if q == 1] + core = attention_core_stats(raw) + kv_stats = kv_update_stats(raw) + if prefill and decode: + prefill_reference_ms = pure_reference_ms( + tp, prefill, decode_phase=False + ) + decode_reference_ms = pure_reference_ms( + tp, decode, decode_phase=True + ) + reference_total_ms = prefill_reference_ms + decode_reference_ms + prefill_share = prefill_reference_ms / reference_total_ms + decode_share = decode_reference_ms / reference_total_ms + projected_prefill = { + name: value * prefill_share for name, value in core.items() + } + projected_decode = { + name: value * decode_share for name, value in core.items() + } + row = {} + for op in ATTENTION_OPS: + row.update(zero_stat_columns(op)) + row.update(stat_columns("attn_kv_cache_save", kv_stats)) + row.update(stat_columns("attn_prefill", projected_prefill)) + row.update(stat_columns("attn_decode", projected_decode)) + prefill_queries = [q for q, _ in prefill] + prefill_contexts = [kv - q for q, kv in prefill] + decode_kv_lengths = [kv for _, kv in decode] + total_batch = len(requests) + row.update( + { + "n_embd": 2048, + "n_q_head": 32, + "n_kv_head": 4, + "block_size": 16, + "num_tensor_parallel_workers": tp, + "max_model_len": 40960, + "batch_size": total_batch, + "prefill_chunk_size": 0, + "kv_cache_size": 0, + "is_prefill": True, + "attention_backend": "FLASH_ATTN", + "is_mixed_batch": False, + "mode": "true_mixed_fused_projected", + "seq_lens": "", + "total_tokens": sum(prefill_queries) + len(decode), + "max_seq_len": "", + "min_seq_len": "", + "avg_seq_len": "", + "equal_seq_len": "", + "seq_len_variance": "", + "seq_len_std": "", + "seq_len_cv": "", + "is_chunked_prefill_sample": False, + "chunk_start_token": "", + "chunk_end_token": "", + "total_prefill_tokens": sum(prefill_queries), + "profiling_precision": "BF16", + "model_arch": "generic", + "quant_signature": "none", + "measurement_type": measurement_type, + "is_true_mixed_batch": True, + "prefill_seq_lens": json.dumps(prefill_queries), + "prefill_kv_cache_sizes": json.dumps(prefill_contexts), + "decode_kv_cache_sizes": json.dumps(decode_kv_lengths), + "num_prefill_seqs": len(prefill), + "num_decode_seqs": len(decode), + "decode_batch_size": len(decode), + "total_batch_size": total_batch, + "total_decode_tokens": len(decode), + "decode_avg_kv_cache_size": ( + sum(decode_kv_lengths) / len(decode_kv_lengths) + ), + "batch_composition_ratio": len(prefill) / total_batch, + "batch_spec": spec, + "projection_policy": ( + "fused_total_conserving_projection_by_same_tp_pure_" + "prefill_decode_reference_ratio" + ), + } + ) + rows.append(row) + mixed_rows.append( + { + "num_tensor_parallel_workers": tp, + "batch_spec": spec, + "num_prefill_seqs": len(prefill), + "num_decode_seqs": len(decode), + "total_prefill_tokens": sum(q for q, _ in prefill), + "total_decode_tokens": len(decode), + "decode_avg_kv_cache_size": sum(kv for _, kv in decode) + / len(decode), + "attention_core_mean_ms": core["mean"], + "attention_core_mean_as_median_ms": core["median"], + "kv_cache_update_median_ms": kv_stats["median"], + "pure_prefill_reference_mean_ms": prefill_reference_ms, + "pure_decode_reference_mean_ms": decode_reference_ms, + "projected_prefill_mean_ms": projected_prefill["mean"], + "projected_decode_mean_ms": projected_decode["mean"], + "projection_sum_error_ms": ( + projected_prefill["mean"] + + projected_decode["mean"] + - core["mean"] + ), + "representation": ( + "one_fused_FA3_call_projected_for_Frontier_with_" + "total_conservation" + ), + } + ) + continue + + is_decode = bool(decode) + queries = [q for q, _ in requests] + contexts = [kv if is_decode else kv - q for q, kv in requests] + avg_query = sum(queries) / len(queries) + variance = sum((query - avg_query) ** 2 for query in queries) / len(queries) + std = math.sqrt(variance) + avg_context = sum(contexts) / len(contexts) + row: dict[str, Any] = {} + for op in ATTENTION_OPS: + row.update(zero_stat_columns(op)) + row.update(stat_columns("attn_kv_cache_save", kv_stats)) + row.update( + stat_columns("attn_decode" if is_decode else "attn_prefill", core) + ) + row.update( + { + "n_embd": 2048, + "n_q_head": 32, + "n_kv_head": 4, + "block_size": 16, + "num_tensor_parallel_workers": tp, + "max_model_len": 40960, + "batch_size": len(requests), + "prefill_chunk_size": 0 if is_decode else sum(queries), + "kv_cache_size": avg_context, + "is_prefill": not is_decode, + "attention_backend": "FLASH_ATTN", + "is_mixed_batch": False, + "mode": "vllm020_batch_spec", + "seq_lens": json.dumps(queries), + "total_tokens": sum(queries), + "max_seq_len": max(queries), + "min_seq_len": min(queries), + "avg_seq_len": avg_query, + "equal_seq_len": len(set(queries)) == 1, + "seq_len_variance": variance, + "seq_len_std": std, + "seq_len_cv": std / avg_query if avg_query else 0.0, + "is_chunked_prefill_sample": (not is_decode and avg_context > 0), + "chunk_start_token": avg_context if not is_decode else 0, + "chunk_end_token": avg_context + sum(queries) if not is_decode else 0, + "total_prefill_tokens": 0 if is_decode else sum(queries), + "profiling_precision": "BF16", + "model_arch": "generic", + "quant_signature": "none", + "measurement_type": measurement_type, + "is_true_mixed_batch": False, + "prefill_seq_lens": "", + "prefill_kv_cache_sizes": "", + "decode_kv_cache_sizes": "", + "num_prefill_seqs": "", + "num_decode_seqs": "", + "decode_batch_size": "", + "total_batch_size": "", + "total_decode_tokens": "", + "decode_avg_kv_cache_size": "", + "batch_composition_ratio": "", + "batch_spec": spec, + "projection_policy": ( + "measured_FA3_core_plus_measured_KV;reshape_assumed_zero;" + "mean_as_median" + ), + } + ) + rows.append(row) + + if seen_tps != {1, 2, 4}: + raise ValueError(f"attention TP coverage mismatch: {seen_tps}") + attention_fields = [ + f"time_stats.{op}.{stat}" for op in ATTENTION_OPS for stat in STAT_NAMES + ] + list(ATTENTION_METADATA) + write_csv(output / "attention.csv", attention_fields, rows) + mixed_fields = [ + "num_tensor_parallel_workers", + "batch_spec", + "num_prefill_seqs", + "num_decode_seqs", + "total_prefill_tokens", + "total_decode_tokens", + "decode_avg_kv_cache_size", + "attention_core_mean_ms", + "attention_core_mean_as_median_ms", + "kv_cache_update_median_ms", + "pure_prefill_reference_mean_ms", + "pure_decode_reference_mean_ms", + "projected_prefill_mean_ms", + "projected_decode_mean_ms", + "projection_sum_error_ms", + "representation", + ] + write_csv(output / "attention_true_mixed_fused.csv", mixed_fields, mixed_rows) + return len(rows), len(mixed_rows), sorted(seen_tps) + + +def load_features(counts: list[int]) -> dict[str, float]: + total = sum(counts) + count = len(counts) + mean = total / count + variance = sum((value - mean) ** 2 for value in counts) / count + probabilities = [value / total for value in counts if value > 0] + entropy = -sum(probability * math.log2(probability) for probability in probabilities) + sorted_counts = sorted(counts) + gini = ( + 2 * sum((index + 1) * value for index, value in enumerate(sorted_counts)) + / (count * total) + - (count + 1) / count + ) + return { + "total_routed_tokens": total, + "num_experts_per_device": count, + "hidden_dim": 2048, + "expert_hidden_dim": 768, + "router_topk": 8, + "model_expansion_ratio": 768 / 2048, + "tokens_per_expert_avg": mean, + "tokens_to_experts_ratio": mean, + "expert_utilization": sum(value > 0 for value in counts) / count, + "min_load_ratio": min(counts) / mean, + "load_imbalance_cv": math.sqrt(variance) / mean, + "max_load_ratio": max(counts) / mean, + "load_entropy": entropy, + "load_gini_coefficient": gini, + } + + +def freeze_moe( + moe_path: Path, router_path: Path, output: Path, *, measurement_type: str +) -> int: + moe = load_json(moe_path) + router = load_json(router_path) + if moe.get("schema_version") != "qwen30_vllm020_moe_raw.v1": + raise ValueError(f"unexpected MoE schema in {moe_path}") + if router.get("schema_version") != "qwen30_vllm020_router_raw.v1": + raise ValueError(f"unexpected router schema in {router_path}") + expected_method = ( + "record_function" if measurement_type == "KERNEL_ONLY" else "cuda_event" + ) + for payload, label in ((moe, "moe"), (router, "router")): + if payload["environment"].get("profile_method", "cuda_event") != expected_method: + raise ValueError( + f"{label} profile method mismatch in {payload}: expected {expected_method}" + ) + router_by_tokens = {int(row["num_tokens"]): row for row in router["rows"]} + rows: list[dict[str, Any]] = [] + seen_pairs: set[tuple[int, int, str]] = set() + for raw in moe["rows"]: + tp = int(raw["tensor_parallel_size"]) + num_tokens = int(raw["num_tokens"]) + routing_mode = str(raw["routing_mode"]) + key = (tp, num_tokens, routing_mode) + if key in seen_pairs: + raise ValueError(f"duplicate MoE row: {key}") + seen_pairs.add(key) + router_row = router_by_tokens[num_tokens] + counts = [int(value) for value in raw["routing_load"]["counts"]] + if sum(counts) != num_tokens * 8 or len(counts) != 128: + raise ValueError(f"invalid routing counts for {key}") + row: dict[str, Any] = {} + row.update(stat_columns("moe_gating_linear", router_row["gate_linear_time_ms"])) + row.update( + stat_columns( + "moe_gating_routing_topk", router_row["routing_topk_time_ms"] + ) + ) + row.update(zero_stat_columns("moe_shuffling")) + row.update(stat_columns("moe_grouped_gemm", raw["time_ms"])) + row.update(load_features(counts)) + row.update( + { + "num_tokens": num_tokens, + "num_experts": 128, + "expert_parallel_size": 1, + "routing_runtime_path": "standard_fused_topk", + "routing_assignment_policy": ( + "logit_topk" + if routing_mode == "uniform_random_logits" + else "fixed_hotset8" + ), + "routing_weight_policy": "softmax_renorm", + "routing_uses_router_logits": routing_mode == "uniform_random_logits", + "gating_runtime_context": "standalone_legacy", + "gating_runtime_context_impl": "vllm020_replicated_linear", + "use_gated": True, + "num_tensor_parallel_workers": tp, + "load_distribution": routing_mode, + "seed": 20260716, + "moe_grouped_gemm_backend": raw["backend"], + "measurement_type": measurement_type, + "profiling_precision": "BF16", + "model_arch": "generic", + "quant_signature": "none", + "router_median_nonadditivity_ratio": router_row[ + "median_nonadditivity_ratio" + ], + "projection_policy": ( + "measured_gate+topk+modular_expert;shuffling_zero_because_" + "expert_measurement_includes_prepare_finalize" + ), + } + ) + rows.append(row) + + tokens = {int(row["num_tokens"]) for row in router["rows"]} + modes = {str(row["routing_mode"]) for row in moe["rows"]} + expected = {(tp, tokens_value, mode) for tp in (1, 2, 4) for tokens_value in tokens for mode in modes} + actual = { + (int(row["num_tensor_parallel_workers"]), int(row["num_tokens"]), str(row["load_distribution"])) + for row in rows + } + if actual != expected: + raise ValueError( + f"MoE TP/token/routing coverage mismatch: missing={expected - actual}, extra={actual - expected}" + ) + moe_fields = [ + f"time_stats.{op}.{stat}" for op in MOE_OPS for stat in STAT_NAMES + ] + list(MOE_METADATA) + write_csv(output / "moe.csv", moe_fields, rows) + return len(rows) + + +def freeze_allreduce(inputs: list[Path], output: Path) -> int: + rows: list[dict[str, Any]] = [] + environments: list[dict[str, Any]] = [] + for path in inputs: + payload = load_json(path) + if payload.get("schema_version") != "qwen30_vllm020_allreduce_raw.v1": + raise ValueError(f"unexpected all-reduce schema in {path}") + rows.extend(payload["rows"]) + environments.append(payload["environment"]) + if {(row["tensor_parallel_size"], row["num_tokens"]) for row in rows} != { + (tp, tokens) + for tp in (2, 4) + for tokens in (1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192) + }: + raise ValueError("all-reduce TP/token coverage mismatch") + (output / "allreduce.json").write_text( + json.dumps( + { + "schema_version": "qwen30_vllm020_allreduce_frozen.v1", + "environment": environments, + "rows": sorted( + rows, + key=lambda row: ( + row["tensor_parallel_size"], + row["num_tokens"], + ), + ), + "frontier_consumption": ( + "diagnostic_only_in_base_profile_only_run; measured lookup " + "requires a separate CC-backend injection ablation" + ), + }, + indent=2, + sort_keys=True, + ) + + "\n" + ) + return len(rows) + + +def main() -> None: + args = parse_args() + if args.allreduce is not None and args.allreduce_frozen is not None: + raise SystemExit("provide either --allreduce or --allreduce-frozen, not both") + all_inputs = [args.linear, *args.attention, args.moe, args.router] + if args.allreduce is not None: + all_inputs.extend(args.allreduce) + if args.allreduce_frozen is not None: + all_inputs.append(args.allreduce_frozen) + for path in all_inputs: + if not path.is_file(): + raise SystemExit(f"missing input: {path}") + args.output.mkdir(parents=True, exist_ok=False) + + linear_output = args.output / "linear_op.csv" + shutil.copyfile(args.linear, linear_output) + with linear_output.open(newline="") as handle: + linear_rows = list(csv.DictReader(handle)) + if not linear_rows: + raise ValueError("linear profile has no rows") + if {row.get("measurement_type") for row in linear_rows} != {args.measurement_type}: + raise ValueError( + f"linear measurement family mismatch: expected {args.measurement_type}" + ) + + attention_rows, mixed_rows, attention_tps = freeze_attention( + list(args.attention), args.output, measurement_type=args.measurement_type + ) + moe_rows = freeze_moe( + args.moe, args.router, args.output, measurement_type=args.measurement_type + ) + allreduce_rows = 0 + allreduce_source = "not_included" + if args.allreduce is not None: + allreduce_rows = freeze_allreduce(list(args.allreduce), args.output) + allreduce_source = "raw_vllm020_measurements" + elif args.allreduce_frozen is not None: + shutil.copyfile(args.allreduce_frozen, args.output / "allreduce.json") + allreduce_rows = len(load_json(args.allreduce_frozen).get("rows", [])) + allreduce_source = "carried_forward_frozen_measurements" + + output_files = [ + linear_output, + args.output / "attention.csv", + args.output / "attention_true_mixed_fused.csv", + args.output / "moe.csv", + ] + if (args.output / "allreduce.json").is_file(): + output_files.append(args.output / "allreduce.json") + batch_composition_augmented = len(args.attention) > 3 + long_context_augmented = args.measurement_type == "KERNEL_ONLY" or any( + "long-context" in path.name for path in args.attention + ) + long_context_coverage: dict[str, Any] = {"included": long_context_augmented} + if long_context_augmented: + with (args.output / "attention.csv").open(newline="") as handle: + frozen_attention = list(csv.DictReader(handle)) + by_tp = {} + for tp in (1, 2, 4): + decode_kv = { + int(float(row["kv_cache_size"])) + for row in frozen_attention + if int(row["num_tensor_parallel_workers"]) == tp + and row["is_prefill"].lower() == "false" + } + required_decode = ( + {128, 1024, 2048, 4096, 8192, 16384, 32768, 40960} + if args.measurement_type == "KERNEL_ONLY" + else {16384, 32768, 40960} + ) + if not required_decode.issubset(decode_kv): + raise ValueError(f"decode KV coverage mismatch for TP{tp}") + by_tp[str(tp)] = { + "decode_kv_lengths": sorted(decode_kv), + } + if args.measurement_type != "KERNEL_ONLY": + mixed_kv = { + int(float(row["decode_avg_kv_cache_size"])) + for row in frozen_attention + if int(row["num_tensor_parallel_workers"]) == tp + and row.get("is_true_mixed_batch", "").lower() == "true" + } + if not {16384, 32768}.issubset(mixed_kv): + raise ValueError(f"long-context mixed coverage mismatch for TP{tp}") + by_tp[str(tp)]["true_mixed_decode_avg_kv_lengths"] = sorted(mixed_kv) + long_context_coverage["by_tp"] = by_tp + manifest = { + "schema_version": ( + "frontier_qwen30_vllm020_kernel_only_profile.v1" + if args.measurement_type == "KERNEL_ONLY" + else ( + "frontier_qwen30_vllm020_frozen_profile.v4" + if long_context_augmented + else ( + "frontier_qwen30_vllm020_frozen_profile.v3" + if batch_composition_augmented + else "frontier_qwen30_vllm020_frozen_profile.v2" + ) + ) + ), + "profile_id": ( + "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-" + "fused-mixed-total-conserving" + + ("-kernel-only-record-function" if args.measurement_type == "KERNEL_ONLY" else "") + + ("-pure-prefill-batch-composition" if batch_composition_augmented else "") + + ("-long-context-decode-mixed" if long_context_augmented else "") + ), + "environment_contract": { + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "dtype": "bfloat16", + "vllm_version": "0.20.0", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "frontier_commit": args.frontier_commit, + "tensor_parallel_sizes": [1, 2, 4], + }, + "row_counts": { + "linear": len(linear_rows), + "attention_frontier_compatible": attention_rows, + "attention_true_mixed_fused_diagnostic": mixed_rows, + "moe": moe_rows, + "allreduce": allreduce_rows, + }, + "attention_tp_coverage": attention_tps, + "attention_long_context_coverage": long_context_coverage, + "projection_contract": { + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "attention": ( + "Pure prefill/extend/decode FA3 core plus separately measured KV update; " + "input/output reshape assumed zero; exported mean is used as median target; " + "true mixed rows use a total-conserving compatibility projection" + ), + "attention_true_mixed": ( + "The directly measured fused total is preserved in diagnostics. Frontier's " + "two targets are projected by the same-TP pure prefill/decode reference " + "ratio, with projected prefill + decode exactly equal to the fused total; " + "the split is a schema compatibility attribution, not an observation" + ), + "attention_pure_prefill_batch_composition": ( + "Direct FA3 measurements for 2/4 requests at query length 2048 and " + "2/4/8/16 requests at query length 512 for each TP; included only " + "when batch-composition attention inputs are supplied" + if batch_composition_augmented + else "not included" + ), + "attention_long_context": ( + "Direct FA3 measurements at decode KV lengths 16K/32K/40K, " + "plus true-mixed prefill/decode measurements at 16K/32K, " + "for TP1/2/4" + if long_context_augmented + else "not included" + ), + "moe": ( + "Replicated gate and fused top-k plus TP-local modular expert kernel; " + "expert measurement already includes prepare/finalize so shuffling is zero" + ), + "allreduce": ( + "Frozen exact runtime measurements; source=" + allreduce_source + ), + }, + "inputs": {str(path.resolve()): sha256(path) for path in all_inputs}, + "outputs": {path.name: sha256(path) for path in output_files}, + } + manifest_path = args.output / "manifest.json" + manifest_path.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps(manifest["row_counts"], sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frontier_vllm020_compat.py b/runs/frontier-qwen30-vllm020-profile-v1/frontier_vllm020_compat.py new file mode 100644 index 0000000..4366c9c --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frontier_vllm020_compat.py @@ -0,0 +1,92 @@ +#!/usr/bin/env python3 +"""Run Frontier's linear profiler against the vLLM 0.20 RoPE API. + +Frontier passes the pre-v0.20 ``get_rope`` arguments separately, while vLLM +0.20 carries the same values in ``rope_parameters``. Keep both repositories +unchanged and adapt only that call boundary for the profiling experiment. +""" + +from __future__ import annotations + +import inspect +import os +from typing import Any + +import torch +from vllm import _custom_ops as vllm_ops +from vllm.model_executor.layers.layernorm import ( + GemmaRMSNorm as VllmGemmaRMSNorm, + fused_add_rms_norm as vllm_fused_add_rms_norm, +) +from vllm.model_executor.layers.rotary_embedding import get_rope as vllm_get_rope + +from frontier.profiling.common.layers import layernorm as frontier_layernorm +from frontier.profiling.common.layers import rotary_embedding as frontier_rope + + +def _vllm020_get_rope_adapter( + *, + head_size: int, + rotary_dim: int, + max_position: int, + base: int | float, + is_neox_style: bool, + rope_scaling: dict[str, Any] | None, + dtype: torch.dtype | None = None, +) -> Any: + rope_parameters = dict(rope_scaling or {}) + rope_parameters["rope_theta"] = base + rope_parameters["rope_dim"] = rotary_dim + return vllm_get_rope( + head_size=head_size, + max_position=max_position, + is_neox_style=is_neox_style, + rope_parameters=rope_parameters, + dtype=dtype, + ) + + +def _vllm020_rms_norm_adapter( + x: torch.Tensor, + weight: torch.Tensor, + variance_epsilon: float, +) -> torch.Tensor: + output = torch.empty_like(x) + vllm_ops.rms_norm(output, x, weight, variance_epsilon) + return output + + +def main() -> None: + parameters = inspect.signature(vllm_get_rope).parameters + if "rope_parameters" not in parameters or "rotary_dim" in parameters: + raise RuntimeError( + "Expected the vLLM 0.20 get_rope API with rope_parameters; " + f"found {inspect.signature(vllm_get_rope)}" + ) + + frontier_rope._VLLM_GET_ROPE = _vllm020_get_rope_adapter + frontier_rope._VLLM_GET_ROPE_IMPORT_ERROR = None + frontier_layernorm.HAS_VLLM_RMSNORM = True + frontier_layernorm.VllmGemmaRMSNorm = VllmGemmaRMSNorm + frontier_layernorm.vllm_rms_norm = _vllm020_rms_norm_adapter + frontier_layernorm.vllm_fused_add_rms_norm = vllm_fused_add_rms_norm + + from frontier.profiling.linear_op.main import main as frontier_main + from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config + + model_root = os.environ.get("MODEL_ROOT") + if not model_root: + raise RuntimeError("MODEL_ROOT must point to the profiled Qwen checkpoint") + model_config = ModelConfig( + model=model_root, + dtype="bfloat16", + max_model_len=8192, + skip_tokenizer_init=True, + generation_config="vllm", + ) + with set_current_vllm_config(VllmConfig(model_config=model_config)): + frontier_main() + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/allreduce.json b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/allreduce.json new file mode 100644 index 0000000..aae1f0a --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/allreduce.json @@ -0,0 +1,824 @@ +{ + "environment": [ + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + } + ], + "frontier_consumption": "diagnostic_only_in_base_profile_only_run; measured lookup requires a separate CC-backend injection ablation", + "rows": [ + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08288000151515007, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.4872959852218628, + "mean": 0.1310015980154276, + "median": 0.07679999992251396, + "min": 0.06217600032687187, + "std": 0.12790721677293843 + }, + { + "max": 0.4402880072593689, + "mean": 0.12842560112476348, + "median": 0.08288000151515007, + "min": 0.0655680000782013, + "std": 0.11220176524616535 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0793600007891655, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.12716799974441528, + "mean": 0.07871360033750534, + "median": 0.0759200006723404, + "min": 0.06032000109553337, + "std": 0.019331314939874535 + }, + { + "max": 0.12380799651145935, + "mean": 0.08059840016067028, + "median": 0.0793600007891655, + "min": 0.06217600032687187, + "std": 0.01687088356254092 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0713919997215271, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12697599828243256, + "mean": 0.0767391998320818, + "median": 0.07078400254249573, + "min": 0.05910399928689003, + "std": 0.018775178979463278 + }, + { + "max": 0.11430399864912033, + "mean": 0.07594559974968433, + "median": 0.0713919997215271, + "min": 0.06124800071120262, + "std": 0.015719922150631036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08056000247597694, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.1037760004401207, + "mean": 0.07954559996724128, + "median": 0.08056000247597694, + "min": 0.05955199897289276, + "std": 0.0135697420393132 + }, + { + "max": 0.10608000308275223, + "mean": 0.07971520014107228, + "median": 0.07593599706888199, + "min": 0.06028800085186958, + "std": 0.015319849772775456 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0865279994904995, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.14470399916172028, + "mean": 0.09121599942445754, + "median": 0.0865279994904995, + "min": 0.06441599875688553, + "std": 0.024893837894277456 + }, + { + "max": 0.12438400089740753, + "mean": 0.08531199917197227, + "median": 0.08031999692320824, + "min": 0.06364800035953522, + "std": 0.01878029830059533 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07135999947786331, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.11753600090742111, + "mean": 0.07606079950928687, + "median": 0.07135999947786331, + "min": 0.05843200162053108, + "std": 0.01755519771639284 + }, + { + "max": 0.1103999987244606, + "mean": 0.07607359997928143, + "median": 0.07073600217700005, + "min": 0.05721599981188774, + "std": 0.016904445220949783 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07321599870920181, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.11740799993276596, + "mean": 0.07749119997024537, + "median": 0.07203200086951256, + "min": 0.05862399935722351, + "std": 0.017701381594822835 + }, + { + "max": 0.11382400244474411, + "mean": 0.07733759954571724, + "median": 0.07321599870920181, + "min": 0.059039998799562454, + "std": 0.017188890882557036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09025600180029869, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.13980799913406372, + "mean": 0.0950367994606495, + "median": 0.09025600180029869, + "min": 0.06815999746322632, + "std": 0.022829835127539378 + }, + { + "max": 0.14764800667762756, + "mean": 0.09710080176591873, + "median": 0.08720000088214874, + "min": 0.07152000069618225, + "std": 0.02500574954063789 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08083200082182884, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.15574400126934052, + "mean": 0.08760640025138855, + "median": 0.07846399769186974, + "min": 0.07097599655389786, + "std": 0.024487311423551025 + }, + { + "max": 0.16284799575805664, + "mean": 0.08963519930839539, + "median": 0.08083200082182884, + "min": 0.07100799679756165, + "std": 0.026047320562445356 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.10891199856996536, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.1582079976797104, + "mean": 0.11541439890861512, + "median": 0.10836799815297127, + "min": 0.09062399715185165, + "std": 0.018285136316576037 + }, + { + "max": 0.1578879952430725, + "mean": 0.11537599861621857, + "median": 0.10891199856996536, + "min": 0.0960640013217926, + "std": 0.018246538626977286 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1703840047121048, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.19327999651432037, + "mean": 0.1707327976822853, + "median": 0.1703840047121048, + "min": 0.14815999567508698, + "std": 0.014211056022719618 + }, + { + "max": 0.19276799261569977, + "mean": 0.1658592015504837, + "median": 0.16379200667142868, + "min": 0.14560000598430634, + "std": 0.013840249648693638 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.25539200007915497, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.2807359993457794, + "mean": 0.25750079900026324, + "median": 0.25539200007915497, + "min": 0.24624000489711761, + "std": 0.008925204570802302 + }, + { + "max": 0.2863999903202057, + "mean": 0.2585055992007256, + "median": 0.255280002951622, + "min": 0.24371199309825897, + "std": 0.012059738582656496 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1021759994328022, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.9770879745483398, + "mean": 0.1974783968180418, + "median": 0.10099200159311295, + "min": 0.05913599953055382, + "std": 0.2660581738745569 + }, + { + "max": 0.892799973487854, + "mean": 0.18164799660444259, + "median": 0.1021759994328022, + "min": 0.06435199826955795, + "std": 0.23947520188197013 + }, + { + "max": 0.6467199921607971, + "mean": 0.15839359983801843, + "median": 0.10100800171494484, + "min": 0.06800000369548798, + "std": 0.16617013141866102 + }, + { + "max": 0.6725760102272034, + "mean": 0.15686400160193442, + "median": 0.10044800117611885, + "min": 0.06063999980688095, + "std": 0.17523222161300497 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12694399803876877, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.6659520268440247, + "mean": 0.1927648030221462, + "median": 0.12694399803876877, + "min": 0.07609599828720093, + "std": 0.16697784531907736 + }, + { + "max": 0.695360004901886, + "mean": 0.19356480240821838, + "median": 0.11726400256156921, + "min": 0.0796160027384758, + "std": 0.17588096678867862 + }, + { + "max": 0.5939840078353882, + "mean": 0.1868800014257431, + "median": 0.12379200011491776, + "min": 0.07427199929952621, + "std": 0.1463231714943902 + }, + { + "max": 0.6635839939117432, + "mean": 0.1874335989356041, + "median": 0.12014400213956833, + "min": 0.07526399940252304, + "std": 0.16680959677760304 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09161599725484848, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12665599584579468, + "mean": 0.09443839862942696, + "median": 0.09124799817800522, + "min": 0.06431999802589417, + "std": 0.0205690155775906 + }, + { + "max": 0.1303039938211441, + "mean": 0.09712959825992584, + "median": 0.09161599725484848, + "min": 0.07648000121116638, + "std": 0.019072048129173236 + }, + { + "max": 0.13836799561977386, + "mean": 0.09821119979023933, + "median": 0.09148800000548363, + "min": 0.0727040022611618, + "std": 0.021314066545189116 + }, + { + "max": 0.12992000579833984, + "mean": 0.09272959977388381, + "median": 0.08931199833750725, + "min": 0.06406400352716446, + "std": 0.021360803830354765 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08580800145864487, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.13449600338935852, + "mean": 0.08960640132427215, + "median": 0.08299200236797333, + "min": 0.06796800345182419, + "std": 0.020633597898445998 + }, + { + "max": 0.14735999703407288, + "mean": 0.09248319901525974, + "median": 0.08580800145864487, + "min": 0.05913599953055382, + "std": 0.025960234928829564 + }, + { + "max": 0.13705599308013916, + "mean": 0.08947199806571007, + "median": 0.08460799977183342, + "min": 0.0634239986538887, + "std": 0.02122838946992339 + }, + { + "max": 0.13846400380134583, + "mean": 0.0859104000031948, + "median": 0.08308799937367439, + "min": 0.05974400043487549, + "std": 0.022373631424433445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09867199882864952, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.1279039978981018, + "mean": 0.09748799875378608, + "median": 0.09678399935364723, + "min": 0.0655359998345375, + "std": 0.02315719144614622 + }, + { + "max": 0.1356479972600937, + "mean": 0.10018239840865135, + "median": 0.09532799944281578, + "min": 0.06185600161552429, + "std": 0.02330700253650042 + }, + { + "max": 0.13142399489879608, + "mean": 0.09778879955410957, + "median": 0.09492799639701843, + "min": 0.06560000032186508, + "std": 0.02307579212430045 + }, + { + "max": 0.1276479959487915, + "mean": 0.09611519873142242, + "median": 0.09867199882864952, + "min": 0.0642239972949028, + "std": 0.02273612181973431 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09646400064229965, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.14060799777507782, + "mean": 0.09770880043506622, + "median": 0.09115200117230415, + "min": 0.06950400024652481, + "std": 0.024752645089849798 + }, + { + "max": 0.14377599954605103, + "mean": 0.09824960008263588, + "median": 0.08999999985098839, + "min": 0.07103999704122543, + "std": 0.025403407389046027 + }, + { + "max": 0.13680000603199005, + "mean": 0.09993600100278854, + "median": 0.09646400064229965, + "min": 0.06790400296449661, + "std": 0.022297424273985882 + }, + { + "max": 0.1391039937734604, + "mean": 0.09769919961690902, + "median": 0.09601600095629692, + "min": 0.06835199892520905, + "std": 0.023899922100804445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08377600088715553, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.1464959979057312, + "mean": 0.09276160076260567, + "median": 0.08377600088715553, + "min": 0.06777600198984146, + "std": 0.024751086465295658 + }, + { + "max": 0.14319999516010284, + "mean": 0.09080640003085136, + "median": 0.080400001257658, + "min": 0.06796800345182419, + "std": 0.023365718881708488 + }, + { + "max": 0.1382399946451187, + "mean": 0.09063360020518303, + "median": 0.08193599805235863, + "min": 0.06627199798822403, + "std": 0.02372618650854302 + }, + { + "max": 0.14313599467277527, + "mean": 0.09044799953699112, + "median": 0.08128000050783157, + "min": 0.06652799993753433, + "std": 0.02462486862033686 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1128000020980835, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.1600320041179657, + "mean": 0.11426239982247352, + "median": 0.1128000020980835, + "min": 0.07657600194215775, + "std": 0.030240087702350687 + }, + { + "max": 0.15881599485874176, + "mean": 0.11206399947404862, + "median": 0.10628800094127655, + "min": 0.0772159993648529, + "std": 0.029933135345483627 + }, + { + "max": 0.15612800419330597, + "mean": 0.10761600062251091, + "median": 0.09860799834132195, + "min": 0.07689599692821503, + "std": 0.02634237020678647 + }, + { + "max": 0.15865600109100342, + "mean": 0.11094079986214637, + "median": 0.10979199782013893, + "min": 0.07583999633789062, + "std": 0.028619217028542445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08755199983716011, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.11999999731779099, + "mean": 0.0843871995806694, + "median": 0.08755199983716011, + "min": 0.06332799792289734, + "std": 0.016638543890716024 + }, + { + "max": 0.1218239963054657, + "mean": 0.08518079966306687, + "median": 0.08032000064849854, + "min": 0.06393600255250931, + "std": 0.019900899429956945 + }, + { + "max": 0.11849600076675415, + "mean": 0.0843968003988266, + "median": 0.08702399954199791, + "min": 0.06297600269317627, + "std": 0.017031908773433545 + }, + { + "max": 0.12300799787044525, + "mean": 0.0846304003149271, + "median": 0.08139199763536453, + "min": 0.06195199862122536, + "std": 0.020106523698622265 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12361599877476692, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.17155200242996216, + "mean": 0.12168959975242614, + "median": 0.12055999785661697, + "min": 0.09609600156545639, + "std": 0.022565485532483928 + }, + { + "max": 0.9246399998664856, + "mean": 0.19978560134768486, + "median": 0.12361599877476692, + "min": 0.09715200215578079, + "std": 0.24230694305662265 + }, + { + "max": 0.9317439794540405, + "mean": 0.20037759989500045, + "median": 0.12327999994158745, + "min": 0.09603200107812881, + "std": 0.24450903278000383 + }, + { + "max": 0.9321280121803284, + "mean": 0.19875840097665787, + "median": 0.12230399996042252, + "min": 0.0950080007314682, + "std": 0.24519252220785093 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.20030399411916733, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.3261440098285675, + "mean": 0.20826880186796187, + "median": 0.19974400103092194, + "min": 0.1409280002117157, + "std": 0.051397264855718945 + }, + { + "max": 0.3248000144958496, + "mean": 0.20548800230026246, + "median": 0.1979840025305748, + "min": 0.141184002161026, + "std": 0.04980002399656717 + }, + { + "max": 0.32547199726104736, + "mean": 0.21280319690704347, + "median": 0.20030399411916733, + "min": 0.14127999544143677, + "std": 0.05304243085685509 + }, + { + "max": 0.26047998666763306, + "mean": 0.1969312012195587, + "median": 0.18079999834299088, + "min": 0.14057600498199463, + "std": 0.042174123638424224 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.2924960106611252, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.45052799582481384, + "mean": 0.3013375997543335, + "median": 0.28273600339889526, + "min": 0.21334399282932281, + "std": 0.07214223764527236 + }, + { + "max": 0.4466240108013153, + "mean": 0.2968191936612129, + "median": 0.27796798944473267, + "min": 0.21241599321365356, + "std": 0.07400678240849741 + }, + { + "max": 0.3830080032348633, + "mean": 0.2946112036705017, + "median": 0.2924960106611252, + "min": 0.21084800362586975, + "std": 0.05323627615746332 + }, + { + "max": 0.4609600007534027, + "mean": 0.3054272010922432, + "median": 0.289792001247406, + "min": 0.21062399446964264, + "std": 0.07608482904865672 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + } + ], + "schema_version": "qwen30_vllm020_allreduce_frozen.v1" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention.csv new file mode 100644 index 0000000..e53b40b --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention.csv @@ -0,0 +1,103 @@ +time_stats.attn_input_reshape.min,time_stats.attn_input_reshape.max,time_stats.attn_input_reshape.mean,time_stats.attn_input_reshape.median,time_stats.attn_input_reshape.std,time_stats.attn_kv_cache_save.min,time_stats.attn_kv_cache_save.max,time_stats.attn_kv_cache_save.mean,time_stats.attn_kv_cache_save.median,time_stats.attn_kv_cache_save.std,time_stats.attn_prefill.min,time_stats.attn_prefill.max,time_stats.attn_prefill.mean,time_stats.attn_prefill.median,time_stats.attn_prefill.std,time_stats.attn_decode.min,time_stats.attn_decode.max,time_stats.attn_decode.mean,time_stats.attn_decode.median,time_stats.attn_decode.std,time_stats.attn_output_reshape.min,time_stats.attn_output_reshape.max,time_stats.attn_output_reshape.mean,time_stats.attn_output_reshape.median,time_stats.attn_output_reshape.std,n_embd,n_q_head,n_kv_head,block_size,num_tensor_parallel_workers,max_model_len,batch_size,prefill_chunk_size,kv_cache_size,is_prefill,attention_backend,is_mixed_batch,mode,seq_lens,total_tokens,max_seq_len,min_seq_len,avg_seq_len,equal_seq_len,seq_len_variance,seq_len_std,seq_len_cv,is_chunked_prefill_sample,chunk_start_token,chunk_end_token,total_prefill_tokens,profiling_precision,model_arch,quant_signature,measurement_type,is_true_mixed_batch,prefill_seq_lens,prefill_kv_cache_sizes,decode_kv_cache_sizes,num_prefill_seqs,num_decode_seqs,decode_batch_size,total_batch_size,total_decode_tokens,decode_avg_kv_cache_size,batch_composition_ratio,batch_spec,projection_policy +0.0,0.0,0.0,0.0,0.0,0.01414399966597557,0.028863999992609024,0.019705599918961526,0.01771199982613325,0.005157200849836681,0.047968000173568726,0.07046400010585785,0.05810240097343922,0.05810240097343922,0.007477463486041561,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04947200044989586,0.020412799902260303,0.01635199971497059,0.010107497379722417,0.046560000628232956,0.08323200047016144,0.05587520003318787,0.05587520003318787,0.011126758739503428,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.022207999601960182,0.017033600155264138,0.015312000177800655,0.002819235991970241,0.05104000121355057,0.07692799717187881,0.056396800279617305,0.056396800279617305,0.007481982178637539,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.022272000089287758,0.01706880023702979,0.01616000011563301,0.002460889579319197,0.06931199878454208,0.0838719978928566,0.07432000041007995,0.07432000041007995,0.004777766433175866,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.018592000007629395,0.028543999418616295,0.02095999978482723,0.019183999858796597,0.003198175496053494,0.12179200351238251,0.15408000349998474,0.1307712011039257,0.1307712011039257,0.00858807797538298,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027775999158620834,0.03385600075125694,0.030131200328469276,0.029680000618100166,0.0021152558103575215,0.32678401470184326,0.3450239896774292,0.33396480381488797,0.33396480381488797,0.0045872424917606375,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04438399896025658,0.05084799975156784,0.046540799736976626,0.04531199857592583,0.002277905811237223,1.0959680080413818,1.1151360273361206,1.0999775886535645,1.0999775886535645,0.005694403246120485,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.078015998005867,0.08691199868917465,0.08114239946007729,0.08019199967384338,0.00292795706334475,4.070400238037109,4.113152027130127,4.087088012695312,4.087088012695312,0.013660567012509554,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.05167999863624573,0.022115200012922286,0.017583999782800674,0.010340094822340818,0.05196800082921982,0.09011200070381165,0.06328320093452933,0.06328320093452933,0.012557341255467452,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01583999954164028,0.026623999699950218,0.018927999772131443,0.017376000061631203,0.003514650316260619,0.06681600213050842,0.07993599772453308,0.0725280001759529,0.0725280001759529,0.004343502558613716,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01648000068962574,0.030880000442266464,0.01945280022919178,0.017967999912798405,0.004096211183007485,0.1311360001564026,0.1546880006790161,0.13908160030841826,0.13908160030841826,0.007511906874366178,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.03558399900794029,0.020851199887692927,0.018559999763965607,0.005235911594130716,0.32950401306152344,0.350271999835968,0.33912960588932034,0.33912960588932034,0.006027400986663648,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.019328000023961067,0.040608000010252,0.022790400311350822,0.020704000256955624,0.006113051965778337,1.1415679454803467,1.1518720388412476,1.144483208656311,1.144483208656311,0.0032332311374389127,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015807999297976494,0.030688000842928886,0.019596799835562707,0.01774400006979704,0.004343384771033462,0.0,0.0,0.0,0.0,0.0,0.049056001007556915,0.07580800354480743,0.05948160067200661,0.05948160067200661,0.009031541471446955,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.02486399933695793,0.01923839971423149,0.018079999834299088,0.0032282528537266424,0.0,0.0,0.0,0.0,0.0,0.05142400041222572,0.07353600114583969,0.059328000620007516,0.059328000620007516,0.0073307807735143084,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016543999314308167,0.03977600112557411,0.021379199624061585,0.018511999398469925,0.006593576176246171,0.0,0.0,0.0,0.0,0.0,0.0488319993019104,0.06435199826955795,0.05479039996862411,0.05479039996862411,0.005672522998491864,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01635199971497059,0.02844800055027008,0.019267200119793416,0.017952000722289085,0.0035068687666949577,0.0,0.0,0.0,0.0,0.0,0.049855999648571014,0.07798399776220322,0.05986879989504815,0.05986879989504815,0.01043914754878828,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016383999958634377,0.026079999282956123,0.01923519968986511,0.017791999503970146,0.0032161974331284568,0.0,0.0,0.0,0.0,0.0,0.058111999183893204,0.1045759990811348,0.06708480007946492,0.06708480007946492,0.013479022462646494,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.04057599976658821,0.019100800156593323,0.015455999877303839,0.007512281243011577,0.0,0.0,0.0,0.0,0.0,0.05363199859857559,0.07782399654388428,0.06090559959411622,0.06090559959411622,0.007544620176348091,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.02191999927163124,0.016684799920767546,0.01563199982047081,0.0024293118621811216,0.0,0.0,0.0,0.0,0.0,0.0629120022058487,0.07891199737787247,0.06891520097851753,0.06891520097851753,0.005472695665695425,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.038943998515605927,0.018313600029796363,0.01561600062996149,0.007127270260115769,0.0,0.0,0.0,0.0,0.0,0.08675199747085571,0.10662399977445602,0.09391999915242194,0.09391999915242194,0.006988099589086635,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.054687999188899994,0.021439999900758268,0.01539199985563755,0.012052764849597775,0.0,0.0,0.0,0.0,0.0,0.13488000631332397,0.1528639942407608,0.1431359991431236,0.1431359991431236,0.005436271464033599,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.041919998824596405,0.01899839974939823,0.015343999955803156,0.007989843526623287,0.0,0.0,0.0,0.0,0.0,0.06176000088453293,0.08374399691820145,0.06747519969940186,0.06747519969940186,0.0066067747128778,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016256000846624374,0.11353600025177002,0.042761600017547606,0.028960000723600388,0.029104301538020762,0.0,0.0,0.0,0.0,0.0,0.09734400361776352,0.14422400295734406,0.11392960175871848,0.11392960175871848,0.013198594600417867,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.034143999218940735,0.018918400071561335,0.01643200032413006,0.005500943993080684,0.0,0.0,0.0,0.0,0.0,0.12918399274349213,0.15087999403476715,0.13807999789714814,0.13807999789714814,0.007658330538677587,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.03641600161790848,0.0198208000510931,0.01780799962580204,0.0057264128169845765,0.0,0.0,0.0,0.0,0.0,0.22099199891090393,0.23904000222682953,0.2293503984808922,0.2293503984808922,0.004861342907006028,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.035840000957250595,0.018908800091594458,0.015792000107467175,0.0064374817924757475,0.0,0.0,0.0,0.0,0.0,0.10134399682283401,0.12201599776744843,0.10896319895982742,0.10896319895982742,0.006336330809165179,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.03417599946260452,0.017846399918198586,0.014800000004470348,0.006495007539635255,0.0,0.0,0.0,0.0,0.0,0.13126400113105774,0.15561600029468536,0.1389280006289482,0.1389280006289482,0.008381472811075022,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03519999980926514,0.019168000388890504,0.01600000075995922,0.005995522477654695,0.0,0.0,0.0,0.0,0.0,0.21728000044822693,0.2343679964542389,0.2231455981731415,0.2231455981731415,0.004720730646739123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.03670400008559227,0.018441599886864425,0.015023999847471714,0.006596535162793127,0.0,0.0,0.0,0.0,0.0,0.39321601390838623,0.4524799883365631,0.4058080047369003,0.4058080047369003,0.01578349755088941,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.022143999114632607,0.016912000067532063,0.015168000012636185,0.0028156089295136347,0.0,0.0,0.0,0.0,0.0,0.15587200224399567,0.3079040050506592,0.17838079929351805,0.17838079929351805,0.04355865575265927,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.02112000063061714,0.015516800060868263,0.01488000014796853,0.001940870731593904,0.0,0.0,0.0,0.0,0.0,0.21587200462818146,0.23561599850654602,0.22250880002975468,0.22250880002975468,0.006181951170646666,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.039264000952243805,0.023609600123018028,0.02131200022995472,0.007236979625711548,0.0,0.0,0.0,0.0,0.0,0.408735990524292,0.470335990190506,0.4336863994598388,0.4336863994598388,0.01844662383160074,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.025407999753952026,0.016227199975401164,0.014960000291466713,0.0031617375441736185,0.0,0.0,0.0,0.0,0.0,0.7412800192832947,0.7627840042114258,0.7464000046253203,0.7464000046253203,0.006112167448837547,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.02364799939095974,0.01809599995613098,0.016352000646293163,0.0035481127058959038,0.04822399839758873,0.08566399663686752,0.05961279980838299,0.05961279980838299,0.011445665413968877,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.020864000543951988,0.01668160008266568,0.015664000064134598,0.0025769063833097584,0.049695998430252075,0.08057600259780884,0.05882879942655563,0.05882879942655563,0.009515126108519331,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.028704000636935234,0.017430400010198355,0.015056000091135502,0.004349294613335555,0.049855999648571014,0.07366400212049484,0.05459520071744919,0.05459520071744919,0.0069610920757925574,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.03161599859595299,0.017788799852132796,0.015711999963968992,0.005005385723318659,0.06102399900555611,0.08179199695587158,0.06650560013949873,0.06650560013949873,0.006947995595801105,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.04416000097990036,0.019200000166893005,0.015488000120967627,0.008561241323364038,0.08054400235414505,0.09196799993515015,0.08607039973139763,0.08607039973139763,0.004145329035483754,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.023391999304294586,0.019667199812829494,0.018463999964296818,0.0022577669687832585,0.18729600310325623,0.20585599541664124,0.19546559900045393,0.19546559900045393,0.006339068824303663,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.026240000501275063,0.03446400165557861,0.028297600522637367,0.027088000439107418,0.0025148441522922374,0.5754240155220032,0.5889919996261597,0.5800191938877105,0.5800191938877105,0.003858869273829596,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04182400181889534,0.047200001776218414,0.043036799877882004,0.0423360001295805,0.0017073405772076728,2.063199996948242,2.0787200927734375,2.067151999473572,2.067151999473572,0.004959651271127835,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.05648000165820122,0.02270399993285537,0.017935999669134617,0.012377915150727689,0.049536000937223434,0.07196799665689468,0.056015999615192415,0.056015999615192415,0.0070476637552742884,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01408000010997057,0.021344000473618507,0.01611520005390048,0.014928000047802925,0.0025499884993961702,0.07072000205516815,0.2642880082130432,0.1588256008923054,0.1588256008923054,0.053220347086102376,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01360000018030405,0.03014400042593479,0.016975999902933837,0.015056000091135502,0.004715159697364111,0.08393599838018417,0.11036799848079681,0.09160000011324881,0.09160000011324881,0.007911669434472792,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.020767999812960625,0.016233599931001663,0.015392000321298838,0.002202615907995427,0.1998399943113327,0.22070400416851044,0.20855360180139543,0.20855360180139543,0.00689307200230667,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.033440001308918,0.018684800155460833,0.016048000194132328,0.00553991005639174,0.6043199896812439,0.635807991027832,0.6126143991947173,0.6126143991947173,0.008745933953408096,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.03359999880194664,0.017811199743300678,0.014944000169634819,0.005926140483565472,0.0,0.0,0.0,0.0,0.0,0.045471999794244766,0.07539200037717819,0.054758400097489356,0.054758400097489356,0.010253548506101549,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02409599907696247,0.01641279999166727,0.01473599998280406,0.003347158638713987,0.0,0.0,0.0,0.0,0.0,0.0461760014295578,0.07529599964618683,0.05460800044238568,0.05460800044238568,0.009748937798340135,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.028416000306606293,0.018611199874430894,0.01598400017246604,0.005209875093863647,0.0,0.0,0.0,0.0,0.0,0.048128001391887665,0.07897599786520004,0.061353600397706036,0.061353600397706036,0.010488153655157845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.025728000327944756,0.016092800162732603,0.01512000011280179,0.003300888123052605,0.0,0.0,0.0,0.0,0.0,0.04864000156521797,0.07648000121116638,0.05810560062527656,0.05810560062527656,0.009473544218404408,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.03551999852061272,0.018588799890130757,0.016064000315964222,0.006071057437992447,0.0,0.0,0.0,0.0,0.0,0.04822399839758873,0.07862400263547897,0.05660480037331582,0.05660480037331582,0.009565394213730401,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.028831999748945236,0.01699519995599985,0.015024000313133001,0.004285000744868188,0.0,0.0,0.0,0.0,0.0,0.04854400083422661,0.06719999760389328,0.05778240002691746,0.05778240002691746,0.006852125554679805,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.024927999824285507,0.0161183999851346,0.01521599991247058,0.0029774808524673907,0.0,0.0,0.0,0.0,0.0,0.05379199981689453,0.08966399729251862,0.06270079985260964,0.06270079985260964,0.009983591277092696,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03001599945127964,0.017648000083863736,0.01547200046479702,0.004585126309484848,0.0,0.0,0.0,0.0,0.0,0.061664000153541565,0.07692799717187881,0.06704320013523103,0.06704320013523103,0.005500191798490629,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.026367999613285065,0.016947199776768684,0.015039999969303608,0.0037951566103550205,0.0,0.0,0.0,0.0,0.0,0.08799999952316284,0.111455999314785,0.0964031994342804,0.0964031994342804,0.007397541615558088,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01369599997997284,0.021247999742627144,0.015359999984502793,0.014640000183135271,0.0020942770950814317,0.0,0.0,0.0,0.0,0.0,0.051711998879909515,0.07065600156784058,0.058054400235414506,0.058054400235414506,0.006633034815910025,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.0307839997112751,0.017148799914866685,0.015039999969303608,0.004882374686312284,0.0,0.0,0.0,0.0,0.0,0.061919998377561576,0.07843200117349625,0.06628479920327664,0.06628479920327664,0.004962852689801192,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013887999579310417,0.02969600073993206,0.017500799987465142,0.015232000034302473,0.00467273319705314,0.0,0.0,0.0,0.0,0.0,0.08819200098514557,0.11097600311040878,0.09493440166115762,0.09493440166115762,0.007509235042985577,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.022112000733613968,0.01751680001616478,0.016047999262809753,0.0030306621792915785,0.0,0.0,0.0,0.0,0.0,0.13065600395202637,0.15110400319099426,0.13857279866933822,0.13857279866933822,0.00750137841249771,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.04012800008058548,0.01892479993402958,0.01550400024279952,0.007459545267816961,0.0,0.0,0.0,0.0,0.0,0.06278400123119354,0.08259200304746628,0.0704512007534504,0.0704512007534504,0.005979055984382744,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.02236800082027912,0.016332800220698118,0.014928000047802925,0.002784044363186731,0.0,0.0,0.0,0.0,0.0,0.1003199964761734,0.1279360055923462,0.10921279862523078,0.10921279862523078,0.00862769617046716,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0225600004196167,0.016128000058233737,0.014800000004470348,0.002958953332547708,0.0,0.0,0.0,0.0,0.0,0.13116799294948578,0.14812800288200378,0.13783999979496003,0.13783999979496003,0.005696384361148053,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.029983999207615852,0.017468800116330386,0.01508800033479929,0.0047379550962483065,0.0,0.0,0.0,0.0,0.0,0.217631995677948,0.2447360008955002,0.22715839892625808,0.22715839892625808,0.008831828221847138,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02304000034928322,0.016512000095099212,0.014512000139802694,0.0035026774828624254,0.0,0.0,0.0,0.0,0.0,0.11020799726247787,0.12307199835777283,0.11600959971547126,0.11600959971547126,0.004667637266950902,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.042080000042915344,0.017510399967432023,0.014607999939471483,0.00823117883530167,0.0,0.0,0.0,0.0,0.0,0.15702399611473083,0.17587199807167053,0.16399359852075576,0.16399359852075576,0.006588299074676393,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0208320003002882,0.015299199987202883,0.01462399959564209,0.001916768086505386,0.0,0.0,0.0,0.0,0.0,0.21779200434684753,0.2415360063314438,0.2260768011212349,0.2260768011212349,0.007251352080685236,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014015999622642994,0.021088000386953354,0.015619200188666582,0.01473599998280406,0.002013227452302141,0.0,0.0,0.0,0.0,0.0,0.3959999978542328,0.4152640104293823,0.40332479774951924,0.40332479774951924,0.006942401431914052,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.022784000262618065,0.016435200069099664,0.015519999898970127,0.0023688301421469523,0.04879999905824661,0.09139200299978256,0.06054079942405224,0.06054079942405224,0.012152608702448775,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014816000126302242,0.02844800055027008,0.017008000146597625,0.015696000307798386,0.003941466294662679,0.047807998955249786,0.07356800138950348,0.05626560002565384,0.05626560002565384,0.00842179125412236,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.053408000618219376,0.019353600218892097,0.01532800029963255,0.01138302400841626,0.048448000103235245,0.0785600021481514,0.0556256003677845,0.0556256003677845,0.009348274502616908,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.03139200061559677,0.01823679991066456,0.016159999649971724,0.004860071238302512,0.055424001067876816,0.08505599945783615,0.0640383992344141,0.0640383992344141,0.00921448636178623,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.026528000831604004,0.017324799951165915,0.01536000007763505,0.0036004822686428305,0.07660800218582153,0.0942080020904541,0.08209280073642732,0.08209280073642732,0.00515895587669752,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.021247999742627144,0.016403199825435876,0.01532800029963255,0.001995897022647934,0.11395200341939926,0.15113599598407745,0.12431039959192276,0.12431039959192276,0.011164431123683732,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017184000462293625,0.028672000393271446,0.019865600019693376,0.017823999747633934,0.0035798491315929977,0.3171840012073517,0.3341119885444641,0.3261695951223373,0.3261695951223373,0.005111046452878918,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024639999493956566,0.030400000512599945,0.02656640000641346,0.02556800004094839,0.0020189846603237303,1.0648640394210815,1.0828479528427124,1.0715327858924866,1.0715327858924866,0.005558639263049851,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014976000413298607,0.021695999428629875,0.017305599898099898,0.01593599934130907,0.0025718046517268054,0.04956800118088722,0.07932800054550171,0.06228480041027069,0.06228480041027069,0.01027160349757827,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01539199985563755,0.03580800071358681,0.019686400331556796,0.017136000096797943,0.005918387811304003,0.05158400163054466,0.10713600367307663,0.06364160068333148,0.06364160068333148,0.015832957809696766,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02223999984562397,0.016672000009566545,0.015887999907135963,0.0020934945946034563,0.06521599739789963,0.08902399986982346,0.07520959973335266,0.07520959973335266,0.007913840684102929,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015231999568641186,0.04028800129890442,0.019971200078725816,0.01646399963647127,0.007351305886577286,0.14467200636863708,0.16844800114631653,0.15363519936800005,0.15363519936800005,0.008174435899956223,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015519999898970127,0.02304000034928322,0.01775679988786578,0.016784000210464,0.0025077243712082584,0.33740800619125366,0.35343998670578003,0.3445120006799698,0.3445120006799698,0.00445648463590358,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.02393599972128868,0.01809599995613098,0.01654400024563074,0.002998393102466254,0.0,0.0,0.0,0.0,0.0,0.0504320003092289,0.0843840017914772,0.060083200410008426,0.060083200410008426,0.00986959318572296,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.030047999694943428,0.019510399922728537,0.017680000513792038,0.004065185265963332,0.0,0.0,0.0,0.0,0.0,0.05004800111055374,0.07036799937486649,0.059315200522542,0.059315200522542,0.006537768821329647,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.02393599972128868,0.018035199772566558,0.016512000001966953,0.0030667689116777724,0.0,0.0,0.0,0.0,0.0,0.05100800096988678,0.06735999882221222,0.058387200161814694,0.058387200161814694,0.005832787739241381,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.026208000257611275,0.017500799987465142,0.01646399963647127,0.0030666103306165714,0.0,0.0,0.0,0.0,0.0,0.05023999884724617,0.07254400104284286,0.057254400476813315,0.057254400476813315,0.006890853653068151,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.027327999472618103,0.0179776000790298,0.01648000068962574,0.003783417447531392,0.0,0.0,0.0,0.0,0.0,0.04819199815392494,0.07100799679756165,0.05621119923889638,0.05621119923889638,0.007824391484124725,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,127.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.036448001861572266,0.019136000238358975,0.016704000532627106,0.006076530095624803,0.0,0.0,0.0,0.0,0.0,0.047488000243902206,0.08441600203514099,0.0588383998721838,0.0588383998721838,0.011275940726332522,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015135999768972397,0.033952001482248306,0.02119360016658902,0.018000000156462193,0.007136646614305304,0.0,0.0,0.0,0.0,0.0,0.04931199923157692,0.06992000341415405,0.05755840018391609,0.05755840018391609,0.007297587694315226,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.034272000193595886,0.0204415999352932,0.017311999574303627,0.00695404701803013,0.0,0.0,0.0,0.0,0.0,0.05215999856591225,0.0735040009021759,0.06117440015077591,0.06117440015077591,0.007381118436894922,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03743999823927879,0.02012479966506362,0.01775999926030636,0.006181045509079057,0.0,0.0,0.0,0.0,0.0,0.06355199962854385,0.08508799970149994,0.07019200026988984,0.07019200026988984,0.0062246580442117845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,1023.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.032607998698949814,0.01959999995306134,0.017487999983131886,0.004882475068460749,0.0,0.0,0.0,0.0,0.0,0.04918399825692177,0.0865280032157898,0.05973760038614274,0.05973760038614274,0.011546847942113974,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02675200067460537,0.017430400010198355,0.016560000367462635,0.003243135094239819,0.0,0.0,0.0,0.0,0.0,0.052671998739242554,0.08367999643087387,0.06238719932734965,0.06238719932734965,0.011207824780630104,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.03612799942493439,0.019327999837696553,0.01688000001013279,0.006058251425153085,0.0,0.0,0.0,0.0,0.0,0.06364800035953522,0.07878399640321732,0.06935679838061332,0.06935679838061332,0.004515588328677643,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03711999952793121,0.019670399930328132,0.017152000218629837,0.006165994646522264,0.0,0.0,0.0,0.0,0.0,0.09071999788284302,0.11507199704647064,0.10252480059862136,0.10252480059862136,0.008782544051535657,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,2047.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.026528000831604004,0.018396800104528665,0.01601599995046854,0.004279788048986283,0.0,0.0,0.0,0.0,0.0,0.05331199988722801,0.07977599650621414,0.06076480001211167,0.06076480001211167,0.008761579122069606,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.02643200010061264,0.01726400004699826,0.015855999663472176,0.003210008417242029,0.0,0.0,0.0,0.0,0.0,0.062431998550891876,0.08246400207281113,0.06970879957079888,0.06970879957079888,0.007016341676068233,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.02223999984562397,0.0166015999391675,0.015887999907135963,0.00204913969129354,0.0,0.0,0.0,0.0,0.0,0.10127999633550644,0.1141119971871376,0.10621120035648347,0.10621120035648347,0.004029318311754605,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03363199904561043,0.019305599946528675,0.016671999357640743,0.0055445582307981234,0.0,0.0,0.0,0.0,0.0,0.1319040060043335,0.15839999914169312,0.14040640145540234,0.14040640145540234,0.007998041073596942,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,4095.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.04447999969124794,0.018908800091594458,0.0157279996201396,0.0086814937461721,0.0,0.0,0.0,0.0,0.0,0.06428799778223038,0.08982399851083755,0.07349760085344315,0.07349760085344315,0.008605284512197258,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.028672000393271446,0.01890560006722808,0.016080000437796116,0.004797584627815021,0.0,0.0,0.0,0.0,0.0,0.11123199760913849,0.14115199446678162,0.11942399889230729,0.11942399889230729,0.008513394706791027,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.024927999824285507,0.016915200091898442,0.015216000378131866,0.003374147499442635,0.0,0.0,0.0,0.0,0.0,0.15887999534606934,0.18111999332904816,0.16934399753808976,0.16934399753808976,0.007415181260761123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.026944000273942947,0.017737600207328796,0.015199999790638685,0.00415598714375255,0.0,0.0,0.0,0.0,0.0,0.2192319929599762,0.23472000658512115,0.22809920012950893,0.22809920012950893,0.004730841376327335,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,8191.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention_true_mixed_fused.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention_true_mixed_fused.csv new file mode 100644 index 0000000..1b7a751 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/attention_true_mixed_fused.csv @@ -0,0 +1,31 @@ +num_tensor_parallel_workers,batch_spec,num_prefill_seqs,num_decode_seqs,total_prefill_tokens,total_decode_tokens,decode_avg_kv_cache_size,attention_core_mean_ms,attention_core_mean_as_median_ms,kv_cache_update_median_ms,representation +1,q64_8q1s512,1,8,64,8,511.0,0.06604800000786781,0.06604800000786781,0.021359999664127827,one_fused_FA3_call_not_split_into_prefill_decode +1,q128_8q1s1k,1,8,128,8,1023.0,0.07483199909329416,0.07483199909329416,0.024255999363958836,one_fused_FA3_call_not_split_into_prefill_decode +1,q128_16q1s1k,1,16,128,16,1023.0,0.10597119852900506,0.10597119852900506,0.018112000077962875,one_fused_FA3_call_not_split_into_prefill_decode +1,q256_16q1s2k,1,16,256,16,2047.0,0.13697599917650224,0.13697599917650224,0.01673599984496832,one_fused_FA3_call_not_split_into_prefill_decode +1,q256_32q1s2k,1,32,256,32,2047.0,0.2163648009300232,0.2163648009300232,0.01515199989080429,one_fused_FA3_call_not_split_into_prefill_decode +1,q512_32q1s4k,1,32,512,32,4095.0,0.3746495962142944,0.3746495962142944,0.016191999427974224,one_fused_FA3_call_not_split_into_prefill_decode +1,q512_64q1s4k,1,64,512,64,4095.0,0.6718560099601746,0.6718560099601746,0.024848000146448612,one_fused_FA3_call_not_split_into_prefill_decode +1,q1k_64q1s8k,1,64,1024,64,8191.0,1.3472000002861022,1.3472000002861022,0.019600000232458115,one_fused_FA3_call_not_split_into_prefill_decode +1,q2k_32q1s4k,1,32,2048,32,4095.0,0.6351647913455962,0.6351647913455962,0.028256000019609928,one_fused_FA3_call_not_split_into_prefill_decode +1,q4k_16q1s4k,1,16,4096,16,4095.0,1.2515872120857237,1.2515872120857237,0.04395199939608574,one_fused_FA3_call_not_split_into_prefill_decode +2,q64_8q1s512,1,8,64,8,511.0,0.05652800053358078,0.05652800053358078,0.019600000232458115,one_fused_FA3_call_not_split_into_prefill_decode +2,q128_8q1s1k,1,8,128,8,1023.0,0.07362559959292413,0.07362559959292413,0.014431999996304512,one_fused_FA3_call_not_split_into_prefill_decode +2,q128_16q1s1k,1,16,128,16,1023.0,0.0736224003136158,0.0736224003136158,0.014479999896138906,one_fused_FA3_call_not_split_into_prefill_decode +2,q256_16q1s2k,1,16,256,16,2047.0,0.10047360062599181,0.10047360062599181,0.01425600005313754,one_fused_FA3_call_not_split_into_prefill_decode +2,q256_32q1s2k,1,32,256,32,2047.0,0.14032640159130094,0.14032640159130094,0.01566399959847331,one_fused_FA3_call_not_split_into_prefill_decode +2,q512_32q1s4k,1,32,512,32,4095.0,0.22105600088834762,0.22105600088834762,0.016032000072300434,one_fused_FA3_call_not_split_into_prefill_decode +2,q512_64q1s4k,1,64,512,64,4095.0,0.36867519915103913,0.36867519915103913,0.015263999812304974,one_fused_FA3_call_not_split_into_prefill_decode +2,q1k_64q1s8k,1,64,1024,64,8191.0,0.6853824079036711,0.6853824079036711,0.015856000129133463,one_fused_FA3_call_not_split_into_prefill_decode +2,q2k_32q1s4k,1,32,2048,32,4095.0,0.3503839999437332,0.3503839999437332,0.018432000651955605,one_fused_FA3_call_not_split_into_prefill_decode +2,q4k_16q1s4k,1,16,4096,16,4095.0,0.7185311973094941,0.7185311973094941,0.026575999334454536,one_fused_FA3_call_not_split_into_prefill_decode +4,q64_8q1s512,1,8,64,8,511.0,0.0562208004295826,0.0562208004295826,0.015520000364631414,one_fused_FA3_call_not_split_into_prefill_decode +4,q128_8q1s1k,1,8,128,8,1023.0,0.061199999228119854,0.061199999228119854,0.018240000121295452,one_fused_FA3_call_not_split_into_prefill_decode +4,q128_16q1s1k,1,16,128,16,1023.0,0.06104319989681244,0.06104319989681244,0.016207999549806118,one_fused_FA3_call_not_split_into_prefill_decode +4,q256_16q1s2k,1,16,256,16,2047.0,0.07356479987502099,0.07356479987502099,0.015584000386297703,one_fused_FA3_call_not_split_into_prefill_decode +4,q256_32q1s2k,1,32,256,32,2047.0,0.09806400015950203,0.09806400015950203,0.015008000191301107,one_fused_FA3_call_not_split_into_prefill_decode +4,q512_32q1s4k,1,32,512,32,4095.0,0.1509471982717514,0.1509471982717514,0.014864000026136637,one_fused_FA3_call_not_split_into_prefill_decode +4,q512_64q1s4k,1,64,512,64,4095.0,0.21687040030956264,0.21687040030956264,0.015343999955803156,one_fused_FA3_call_not_split_into_prefill_decode +4,q1k_64q1s8k,1,64,1024,64,8191.0,0.3848575979471207,0.3848575979471207,0.015647999942302704,one_fused_FA3_call_not_split_into_prefill_decode +4,q2k_32q1s4k,1,32,2048,32,4095.0,0.2785151988267898,0.2785151988267898,0.01726400014013052,one_fused_FA3_call_not_split_into_prefill_decode +4,q4k_16q1s4k,1,16,4096,16,4095.0,0.456512001156807,0.456512001156807,0.017680000513792038,one_fused_FA3_call_not_split_into_prefill_decode diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/linear_op.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/linear_op.csv new file mode 100644 index 0000000..2803f02 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/linear_op.csv @@ -0,0 +1,37 @@ +time_stats.emb.min,time_stats.emb.max,time_stats.emb.mean,time_stats.emb.median,time_stats.emb.std,time_stats.input_layernorm.min,time_stats.input_layernorm.max,time_stats.input_layernorm.mean,time_stats.input_layernorm.median,time_stats.input_layernorm.std,time_stats.attn_pre_proj.min,time_stats.attn_pre_proj.max,time_stats.attn_pre_proj.mean,time_stats.attn_pre_proj.median,time_stats.attn_pre_proj.std,time_stats.attn_rope.min,time_stats.attn_rope.max,time_stats.attn_rope.mean,time_stats.attn_rope.median,time_stats.attn_rope.std,time_stats.attn_post_proj.min,time_stats.attn_post_proj.max,time_stats.attn_post_proj.mean,time_stats.attn_post_proj.median,time_stats.attn_post_proj.std,time_stats.post_attention_layernorm.min,time_stats.post_attention_layernorm.max,time_stats.post_attention_layernorm.mean,time_stats.post_attention_layernorm.median,time_stats.post_attention_layernorm.std,n_head,n_kv_head,n_embd,n_expanded_embd,vocab_size,use_gated_mlp,use_qk_norm,attn_output_gate,num_tokens,num_tensor_parallel_workers,padded_n_embd,padded_n_expanded_embd,model_arch,is_step2_mini,share_expert_dim,share_q_dim,measurement_type,profiling_precision,quant_signature +0.029184000566601753,0.06780800223350525,0.03157280012965202,0.030736000277101994,0.005874173435341216,0.033215999603271484,0.04825599864125252,0.03443359974771738,0.0337119996547699,0.0031825678429048183,1.438431978225708,1.505568027496338,1.446228802204132,1.4429279565811157,0.014128607642643025,0.538752019405365,0.5440319776535034,0.5416463971138,0.5420799851417542,0.0016099306164432847,1.0073280334472656,1.0163840055465698,1.0098415970802308,1.0081279873847961,0.0032980457197329728,0.04022400081157684,0.041728001087903976,0.04091359991580248,0.04081599973142147,0.0004728505416767937,32,4,2048,768,151936,True,True,False,8192,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.01360000018030405,0.06784000247716904,0.017755200061947106,0.0179840000346303,0.00837681421401443,0.01836800016462803,0.03651199862360954,0.019606399815529585,0.018719999119639397,0.0038821958848767424,0.7512000203132629,0.8208960294723511,0.7566704005002975,0.75382399559021,0.014793092586577971,0.28995200991630554,0.29337599873542786,0.29135999977588656,0.29150401055812836,0.000998381071258815,0.5149760246276855,0.5169600248336792,0.5160208016633987,0.5158880054950714,0.0005038652783291977,0.02051199972629547,0.021503999829292297,0.021067200042307378,0.021104000508785248,0.00023542855306902367,32,4,2048,768,151936,True,True,False,4096,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.0080960001796484,0.04281599819660187,0.011092799971811474,0.011039999779313803,0.005489135752949302,0.012223999947309494,0.026335999369621277,0.013187199970707298,0.01247999956831336,0.0030236510562153375,0.3928639888763428,0.45372799038887024,0.3976895987987518,0.39528000354766846,0.012905176716136006,0.1547199934720993,0.15884800255298615,0.15712319910526276,0.1573439985513687,0.001165121945135037,0.26633599400520325,0.268095999956131,0.26719200164079665,0.2671840041875839,0.0004242740199415328,0.013024000450968742,0.013887999579310417,0.013489600038155913,0.013520000036805868,0.0002382817554057738,32,4,2048,768,151936,True,True,False,2048,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.00825599953532219,0.04755200073122978,0.02398160002194345,0.01961600035429001,0.00874683840888523,0.018880000337958336,0.03868800029158592,0.021590400114655496,0.0208320003002882,0.004057015751746236,0.24316799640655518,0.2710399925708771,0.2521967992186546,0.25065599381923676,0.007998418528894075,0.09644799679517746,0.19120000302791595,0.10407840013504029,0.09963199868798256,0.020043722414992166,0.13600000739097595,0.18892799317836761,0.15760480016469955,0.15760000050067902,0.0112223677907762,0.009664000011980534,0.010015999898314476,0.009836799977347255,0.009824000298976898,9.016971952948177e-05,32,4,2048,768,151936,True,True,False,1024,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017376000061631203,0.044704001396894455,0.026318399980664254,0.027312000282108784,0.007565344034680866,0.01836800016462803,0.03014400042593479,0.021276800055056812,0.020655999891459942,0.002632977855097951,0.1438719928264618,0.17132799327373505,0.152497598528862,0.15012799948453903,0.007947150319625347,0.1430719941854477,0.19305600225925446,0.16630879789590836,0.1685439944267273,0.014628024163894684,0.08899199962615967,0.10467199981212616,0.0943599995225668,0.09374399855732918,0.003472669494074113,0.007615999784320593,0.007935999892652035,0.007769599952735007,0.0077760000713169575,7.680004540222077e-05,32,4,2048,768,151936,True,True,False,512,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016992000862956047,0.0544000007212162,0.02573199989274144,0.026016000658273697,0.00803323401720434,0.018432000651955605,0.02348800003528595,0.020648000109940768,0.02062400057911873,0.0013939985047930988,0.10220800340175629,0.1361600011587143,0.11850560046732425,0.11684799939393997,0.010637754898360304,0.16710400581359863,0.21110400557518005,0.19078560024499894,0.19409599900245667,0.013714352646558832,0.06265600025653839,0.0740479975938797,0.06842879951000214,0.0690080001950264,0.0031292240446560557,0.00979200005531311,0.033440001308918,0.01864320016466081,0.017280000261962414,0.006957998188876383,32,4,2048,768,151936,True,True,False,256,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017152000218629837,0.04396799951791763,0.025907999789342284,0.02598400041460991,0.007714554737915267,0.018400000408291817,0.03667199984192848,0.024132800102233887,0.02112000063061714,0.006120348733354326,0.10678400099277496,0.1363839954137802,0.1193264003843069,0.11583999916911125,0.009838647443214228,0.17017599940299988,0.22748799622058868,0.18853759989142418,0.18433599919080734,0.015728078443174653,0.04569600149989128,0.06652799993753433,0.05192639995366335,0.05151999928057194,0.004516605694976449,0.021856000646948814,0.026335999369621277,0.02384479995816946,0.023599999956786633,0.0011301153013314744,32,4,2048,768,151936,True,True,False,128,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017343999817967415,1.0683200359344482,0.05748240072280168,0.029504000209271908,0.16366824814254827,0.018688000738620758,0.3317759931087494,0.03685439983382821,0.021151999942958355,0.06767422466731164,0.10255999863147736,0.9434880018234253,0.16412640027701855,0.11956800147891045,0.17964606281728834,0.1714559942483902,2.1306240558624268,0.3011296011507511,0.1926399990916252,0.42310127734378766,0.03574400022625923,0.6859520077705383,0.08389280084520578,0.04279999993741512,0.14321647071615612,0.020479999482631683,0.1831360012292862,0.033024000097066165,0.023856000043451786,0.03470987082429836,32,4,2048,768,151936,True,True,False,64,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016095999628305435,0.05142400041222572,0.026363200135529043,0.02676799986511469,0.008637725852473854,0.01849599927663803,0.03577600046992302,0.022193600237369538,0.020848000422120094,0.004615266181181815,0.10540799796581268,0.15014399588108063,0.12211520001292228,0.11896000057458878,0.012772013396624768,0.17315199971199036,0.21478399634361267,0.1881632000207901,0.1873439997434616,0.011761164657572015,0.03481600061058998,0.058079998940229416,0.042200000025331974,0.03969600051641464,0.006461281521769989,0.02143999934196472,0.038816001266241074,0.02466559996828437,0.023856000043451786,0.0035418033748569927,32,4,2048,768,151936,True,True,False,32,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015904000028967857,0.03753599897027016,0.023127999808639287,0.02527999971061945,0.0054230027890305385,0.018400000408291817,0.03001599945127964,0.020488000102341176,0.020096000283956528,0.002585688394137697,0.10355199873447418,0.1438400000333786,0.11536479964852334,0.11124800145626068,0.011136028294919255,0.16502399742603302,0.2072959989309311,0.18646399974822997,0.19075199961662292,0.013296437761247597,0.03142400085926056,0.05215999856591225,0.03888959977775812,0.03750399872660637,0.0057399302067536314,0.020128000527620316,0.04064000025391579,0.023937600292265417,0.023648000322282314,0.004144197400898248,32,4,2048,768,151936,True,True,False,16,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015200000256299973,0.037151999771595,0.02144480012357235,0.02195199951529503,0.005576364091933697,0.017952000722289085,0.0226879995316267,0.019934400077909233,0.019952000118792057,0.0011812499470458758,0.10063999891281128,0.13468800485134125,0.11595199964940547,0.1207519993185997,0.011974301621092394,0.16332800686359406,0.20748800039291382,0.18162400051951408,0.1767839938402176,0.01474120743720334,0.03222399950027466,0.043455999344587326,0.03829439990222454,0.03859200142323971,0.0031378131240041122,0.020959999412298203,0.03587200120091438,0.023795200139284135,0.023423999547958374,0.0031339489695198443,32,4,2048,768,151936,True,True,False,8,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.014944000169634819,0.04022400081157684,0.02162720002233982,0.022463999688625336,0.006017219317026815,0.018464000895619392,0.026528000831604004,0.021264000236988066,0.020896000787615776,0.001934095017586082,0.10156799852848053,0.1703999936580658,0.12565439902245998,0.1244799979031086,0.01641776722785409,0.1653759926557541,0.23865599930286407,0.1969360001385212,0.19223999977111816,0.02058903050274278,0.03254399821162224,0.06752000004053116,0.0443536002188921,0.041519999504089355,0.00984829071098989,0.020096000283956528,0.040031999349594116,0.026934400014579297,0.02478400059044361,0.00562071702648593,32,4,2048,768,151936,True,True,False,1,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.7531200051307678,0.8461440205574036,0.7618160009384155,0.7576479911804199,0.019464233617982506,0.2922559976577759,0.2985599935054779,0.2953856036067009,0.29576000571250916,0.001575901077689139,0.510047972202301,0.5140479803085327,0.5121696025133133,0.5123839974403381,0.0012263137313476844,,,,,,32,4,2048,768,151936,True,True,False,8192,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.39529600739479065,0.47279998660087585,0.40216960161924364,0.39825600385665894,0.0162749796240819,0.15625600516796112,0.16211199760437012,0.15959519892930984,0.15988799929618835,0.0011493418942396922,0.2635200023651123,0.2642880082130432,0.2639120012521744,0.26392000913619995,0.0001903593401384135,,,,,,32,4,2048,768,151936,True,True,False,4096,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2531839907169342,0.32419198751449585,0.26924319565296173,0.26049599051475525,0.01657194262368116,0.10051199793815613,0.2375359982252121,0.12411200068891048,0.10311999917030334,0.039466165428540506,0.15881599485874176,0.19289599359035492,0.16896959990262986,0.1640480011701584,0.009162416086418127,,,,,,32,4,2048,768,151936,True,True,False,2048,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15302400290966034,0.1780800074338913,0.16284480094909667,0.16113600134849548,0.00740355661356144,0.14521600306034088,0.20233599841594696,0.17807039842009545,0.1796799972653389,0.013908448560094403,0.0907519981265068,0.09750399738550186,0.09460479989647866,0.09478399902582169,0.0017965487667361475,,,,,,32,4,2048,768,151936,True,True,False,1024,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11740799993276596,0.17958399653434753,0.13262080028653145,0.12878400087356567,0.014146060532423056,0.17468799650669098,0.21161599457263947,0.1910431995987892,0.18966399878263474,0.01270903647700971,0.05926400050520897,0.07577600330114365,0.06530559975653887,0.06404799968004227,0.00479458462514625,,,,,,32,4,2048,768,151936,True,True,False,512,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11260800063610077,0.17209599912166595,0.13610880002379416,0.13809599727392197,0.017675922458993677,0.18729600310325623,0.26822400093078613,0.20815680101513861,0.2078079953789711,0.01700718593658771,0.04499199986457825,0.06537599861621857,0.0507551996037364,0.04787199944257736,0.005727123232692158,,,,,,32,4,2048,768,151936,True,True,False,256,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11257600039243698,0.23452800512313843,0.13866880126297473,0.13964799791574478,0.026590047697062115,0.1828799992799759,0.24751999974250793,0.20735519900918006,0.20670399814844131,0.01500438131586336,0.03654399886727333,0.05593600124120712,0.0408239996060729,0.03892800025641918,0.004864409450710354,,,,,,32,4,2048,768,151936,True,True,False,128,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11734399944543839,0.18726399540901184,0.13890240006148816,0.13308800011873245,0.019086167340006257,0.16991999745368958,0.2443840056657791,0.19185120090842248,0.19257599860429764,0.018017536159219673,0.03033600002527237,0.04956800118088722,0.038387199863791466,0.03750400058925152,0.00500897018702887,,,,,,32,4,2048,768,151936,True,True,False,64,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11420799791812897,0.18115200102329254,0.13938880078494548,0.1393439993262291,0.018141313962922536,0.1693439930677414,0.221343994140625,0.19187839925289155,0.19438399374485016,0.01430752121272052,0.03017600066959858,0.06019200012087822,0.03866560012102127,0.0364960003644228,0.0074199790031205266,,,,,,32,4,2048,768,151936,True,True,False,32,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11276800185441971,0.16223999857902527,0.1331360016018152,0.13305599987506866,0.01432493740801146,0.17187200486660004,0.23625600337982178,0.19287680014967917,0.1913280040025711,0.01754628831589704,0.029823999851942062,0.07574400305747986,0.03900959976017475,0.036927999928593636,0.009277465083962879,,,,,,32,4,2048,768,151936,True,True,False,16,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1130559965968132,0.19814400374889374,0.13379519879817964,0.12531199678778648,0.020074427302248836,0.16841599345207214,0.21139200031757355,0.1893615983426571,0.19092799723148346,0.012418720676762711,0.029503999277949333,0.07558400183916092,0.040144000016152856,0.03728000074625015,0.010223239196498205,,,,,,32,4,2048,768,151936,True,True,False,8,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.115167997777462,0.2640640139579773,0.13805920109152794,0.13118399679660797,0.03132849683515479,0.17103999853134155,0.2977280020713806,0.19899839907884598,0.1976960003376007,0.028928046763067948,0.0297279991209507,0.05990400165319443,0.03834720011800528,0.0363520011305809,0.0070885330713495905,,,,,,32,4,2048,768,151936,True,True,False,1,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.3940800130367279,0.46483200788497925,0.39923040121793746,0.3957759886980057,0.015108903906233569,0.16022400557994843,0.1634880006313324,0.16203359961509706,0.16228799521923065,0.0010220720912414007,0.26073598861694336,0.2627840042114258,0.26138080209493636,0.2613760083913803,0.00041640363494172775,,,,,,32,4,2048,768,151936,True,True,False,8192,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2536959946155548,0.27161601185798645,0.25960480123758317,0.2577280104160309,0.005142017404245015,0.09849599748849869,0.10281600058078766,0.10057279989123344,0.10063999891281128,0.0008836232237268523,0.13913600146770477,0.17606399953365326,0.15875840038061143,0.15988799929618835,0.010058952112389172,,,,,,32,4,2048,768,151936,True,True,False,4096,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15142400562763214,0.1780479997396469,0.16139679849147798,0.1602879986166954,0.0070656055731385115,0.14601600170135498,0.23343999683856964,0.17913119941949845,0.179967999458313,0.023065274309176566,0.09388799965381622,0.12108799815177917,0.10317599996924401,0.10073599964380264,0.007364345618470178,,,,,,32,4,2048,768,151936,True,True,False,2048,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12198399752378464,0.18406400084495544,0.13770400024950505,0.13232000172138214,0.01623164885687898,0.1773120015859604,0.20688000321388245,0.1870912007987499,0.18535999953746796,0.00825628058448234,0.05766399949789047,0.06739199906587601,0.06187200043350458,0.061216000467538834,0.003004312467037863,,,,,,32,4,2048,768,151936,True,True,False,1024,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11737599968910217,0.18614399433135986,0.13109439946711063,0.12771200388669968,0.01372168725934724,0.17587199807167053,0.2699519991874695,0.19926720038056372,0.19075199961662292,0.02327478982490538,0.041728001087903976,0.06224000081419945,0.05000480003654957,0.049375999718904495,0.006185850944273961,,,,,,32,4,2048,768,151936,True,True,False,512,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1159679964184761,0.1610880047082901,0.12988320142030715,0.12494400516152382,0.011865375783184065,0.1767680048942566,0.27379199862480164,0.21035519987344742,0.21275199949741364,0.023206964017550125,0.037087999284267426,0.062144000083208084,0.04439679980278015,0.04283200018107891,0.006155226392511492,,,,,,32,4,2048,768,151936,True,True,False,256,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12015999853610992,0.158720001578331,0.13573280088603495,0.1343199983239174,0.010666830836014414,0.1737920045852661,0.22127999365329742,0.20261440128087999,0.20321600139141083,0.011153965849067301,0.03587200120091438,0.04944000020623207,0.038265600241720675,0.037328001111745834,0.0030431580113575636,,,,,,32,4,2048,768,151936,True,True,False,128,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1212799996137619,0.15139199793338776,0.13183839991688728,0.13014400005340576,0.00851207547022104,0.17257599532604218,0.2250880002975464,0.1872655987739563,0.18193599581718445,0.013638284975248818,0.02969600073993206,0.0525440014898777,0.03840640028938651,0.03444799967110157,0.007598511754254174,,,,,,32,4,2048,768,151936,True,True,False,64,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11270400136709213,0.1624639928340912,0.13508000001311302,0.13180799782276154,0.014243564451606775,0.17052799463272095,0.23715199530124664,0.1953311987221241,0.1966560035943985,0.01500670718978398,0.030239999294281006,0.05270399898290634,0.03807039987295866,0.03742399998009205,0.004906165602021684,,,,,,32,4,2048,768,151936,True,True,False,32,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12319999933242798,0.17919999361038208,0.14059039913117885,0.14156799763441086,0.016217662982107223,0.1701119989156723,0.21987199783325195,0.18903039917349815,0.1870879977941513,0.014117854507841239,0.030368000268936157,0.06406400352716446,0.04045119984075427,0.03710399940609932,0.00878942205983628,,,,,,32,4,2048,768,151936,True,True,False,16,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11296000331640244,0.16761599481105804,0.13843199908733367,0.13814399391412735,0.014925286935582404,0.1711679995059967,0.21561600267887115,0.1906527981162071,0.1876479983329773,0.011803992622137974,0.0306560005992651,0.09216000139713287,0.041129599791020155,0.036847999319434166,0.013739721468154687,,,,,,32,4,2048,768,151936,True,True,False,8,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11929599940776825,0.2008959949016571,0.14535359852015972,0.13971199840307236,0.024769473061576282,0.16803200542926788,0.25123199820518494,0.19573760256171227,0.19366399943828583,0.023196011081705884,0.03049599938094616,0.08089599758386612,0.04739360017701984,0.0453919991850853,0.01286389846889463,,,,,,32,4,2048,768,151936,True,True,False,1,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/manifest.json b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/manifest.json new file mode 100644 index 0000000..6873602 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/manifest.json @@ -0,0 +1,53 @@ +{ + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "674e010598eea4c06bdee04ccba413f4c34194fbe2d70b143c6ec2169b638cc5", + "attention_true_mixed_fused.csv": "92cd22f8b669abc7af488cf3d067ed880a6cced2dba1a3213a93a4f02a0001a8", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target", + "attention_true_mixed": "Preserved as fused-total diagnostics and excluded from attention.csv because Frontier requires an unobservable prefill/decode split", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 102, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v1" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/moe.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/moe.csv new file mode 100644 index 0000000..e5bf507 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v1/moe.csv @@ -0,0 +1,73 @@ +time_stats.moe_gating_linear.min,time_stats.moe_gating_linear.max,time_stats.moe_gating_linear.mean,time_stats.moe_gating_linear.median,time_stats.moe_gating_linear.std,time_stats.moe_gating_routing_topk.min,time_stats.moe_gating_routing_topk.max,time_stats.moe_gating_routing_topk.mean,time_stats.moe_gating_routing_topk.median,time_stats.moe_gating_routing_topk.std,time_stats.moe_shuffling.min,time_stats.moe_shuffling.max,time_stats.moe_shuffling.mean,time_stats.moe_shuffling.median,time_stats.moe_shuffling.std,time_stats.moe_grouped_gemm.min,time_stats.moe_grouped_gemm.max,time_stats.moe_grouped_gemm.mean,time_stats.moe_grouped_gemm.median,time_stats.moe_grouped_gemm.std,num_tokens,num_experts,num_experts_per_device,expert_parallel_size,routing_runtime_path,routing_assignment_policy,routing_weight_policy,routing_uses_router_logits,gating_runtime_context,gating_runtime_context_impl,router_topk,hidden_dim,expert_hidden_dim,use_gated,num_tensor_parallel_workers,total_routed_tokens,model_expansion_ratio,tokens_per_expert_avg,tokens_to_experts_ratio,expert_utilization,min_load_ratio,load_imbalance_cv,max_load_ratio,load_entropy,load_gini_coefficient,load_distribution,seed,moe_grouped_gemm_backend,measurement_type,profiling_precision,model_arch,quant_signature,router_median_nonadditivity_ratio,projection_policy +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.33926400542259216,0.405023992061615,0.36780479848384856,0.36507199704647064,0.01690507644474779,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,1.1943039894104004,1.286784052848816,1.228384006023407,1.2273280024528503,0.02832547242381263,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.3984375,0.0,1.346291201783626,4.0,5.59375,0.661865234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,1.679744005203247,1.766144037246704,1.7095808148384095,1.7015680074691772,0.02438921262998535,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.625,0.0,1.015504800579495,5.0,6.15516433212955,0.529052734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,2.1576640605926514,2.2921600341796875,2.2097824096679686,2.188944101333618,0.045572321842012986,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.875,0.0,0.6343057228182637,2.5,6.64370748444639,0.35369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,2.422368049621582,2.5130879878997803,2.4516672134399413,2.434159994125366,0.03278900287381846,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.984375,0.0,0.4921254921257382,2.25,6.817190042344769,0.272369384765625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,2.2217600345611572,2.289599895477295,2.2571327924728393,2.263375997543335,0.021660416089449488,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,1.0,0.125,0.3486861500690843,1.875,6.908192310183997,0.197662353515625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,2.18668794631958,2.318079948425293,2.2218016147613526,2.211087942123413,0.035380897213135316,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,1.0,0.4375,0.2525504668006971,1.875,6.953347743053017,0.1410369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,2.2291839122772217,2.3929600715637207,2.2908096313476562,2.2804640531539917,0.04479348924786221,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,1.0,0.65625,0.15765965680164504,1.5625,6.98229848728205,0.08779525756835938,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,2.0678720474243164,2.1297600269317627,2.0837119817733765,2.0779199600219727,0.017880044357986735,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,1.0,0.625,0.12169081635504074,1.3125,6.9892029662356325,0.06879425048828125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,2.916032075881958,3.0819520950317383,2.9805248022079467,2.9656319618225098,0.05482195799019572,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,1.0,0.796875,0.07935434147688751,1.1796875,6.9954297964750305,0.044734954833984375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,4.421599864959717,4.535359859466553,4.486294317245483,4.497056007385254,0.036990243787549344,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,1.0,0.8203125,0.060849326483103046,1.17578125,6.9973188375859685,0.033740997314453125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,7.302591800689697,7.402751922607422,7.354758310317993,7.3464319705963135,0.032142662400335566,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,1.0,0.890625,0.0412323087266341,1.08984375,6.998772433185578,0.02334284782409668,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.35280001163482666,0.39692801237106323,0.37662720382213594,0.37196800112724304,0.013401318050665304,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.4692479968070984,0.5523840188980103,0.5134752035140991,0.5100640058517456,0.02291433464135784,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.34652799367904663,0.4119040071964264,0.3789471983909607,0.38550400733947754,0.02073945105803335,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.31462401151657104,0.7456960082054138,0.38617280423641204,0.34545600414276123,0.12230201266253077,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.32521599531173706,0.419871985912323,0.36325119733810424,0.34968000650405884,0.03161798848223672,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.289792001247406,0.4663360118865967,0.4091839998960495,0.41655999422073364,0.0446001986506615,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.3761279881000519,0.4416320025920868,0.40686399936676027,0.40540799498558044,0.02257778769899645,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.7172480225563049,0.8663039803504944,0.7723807990550995,0.7591840028762817,0.04164772379451242,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.0195519924163818,1.2216639518737793,1.1253888130187988,1.1453600525856018,0.06548005322243594,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7490559816360474,1.9644800424575806,1.8529024004936219,1.814303994178772,0.08042565617327288,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,3.2479360103607178,3.385279893875122,3.296070408821106,3.2800960540771484,0.04525529026442046,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,6.344799995422363,6.517856121063232,6.464438438415527,6.478623867034912,0.05116674443145098,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.2648000121116638,0.325439989566803,0.28852800130844114,0.28390398621559143,0.01933778377635077,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.7347840070724487,0.862496018409729,0.7769344031810761,0.769216001033783,0.03290485328796285,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.421875,0.0,1.346291201783626,6.0,5.652114648336087,0.636962890625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.9198399782180786,0.9646080136299133,0.9412063956260681,0.9411839842796326,0.014939365085478117,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.5703125,0.0,1.118033988749895,5.0,6.008641773518898,0.580810546875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,1.2796800136566162,1.3484159708023071,1.3006976008415223,1.2929120063781738,0.020807998177176254,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.8828125,0.0,0.6959705453537527,3.0,6.60872850615583,0.38055419921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,1.3630399703979492,1.4430400133132935,1.3909215927124023,1.3892319798469543,0.022335744492366926,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.984375,0.0,0.5201036555341637,3.0,6.798826509158851,0.28302001953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,1.27948796749115,1.3904000520706177,1.3176063895225525,1.309440016746521,0.038060887827312775,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,1.0,0.25,0.3511282039725661,1.875,6.91002266305238,0.1970977783203125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,1.264415979385376,1.3145920038223267,1.2791999936103822,1.2753440141677856,0.014130605249568332,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,1.0,0.375,0.24692938483248605,1.6875,6.955481130775285,0.13909912109375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,1.3081920146942139,1.347648024559021,1.3292255997657776,1.329967975616455,0.014558863679016933,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,1.0,0.625,0.17143053326165383,1.5625,6.9786675275754035,0.09520339965820312,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.242751955986023,1.3112000226974487,1.2747935891151427,1.266207993030548,0.021093073517695057,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,1.0,0.78125,0.11000099875256815,1.296875,6.991308871213679,0.062183380126953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7388160228729248,1.8077759742736816,1.772764801979065,1.772704005241394,0.021056644077284283,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,1.0,0.78125,0.0864630150197678,1.1796875,6.994552526394139,0.048796653747558594,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,2.6563520431518555,2.7063679695129395,2.6785055875778196,2.6791679859161377,0.01639963463052944,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,1.0,0.8671875,0.06127686514721937,1.16015625,6.997291583027146,0.03497934341430664,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,4.386879920959473,4.452256202697754,4.4108480453491214,4.406303882598877,0.019768161791937636,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,1.0,0.884765625,0.041723768525324195,1.1171875,6.998746434318934,0.02298593521118164,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.24208000302314758,0.4028480052947998,0.3041536003351212,0.277103990316391,0.05660721484881584,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.2447039932012558,0.30502399802207947,0.26446720361709597,0.26265600323677063,0.016744548364435372,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.2337920069694519,0.2881599962711334,0.26074880361557007,0.264384001493454,0.016469850143940968,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.23369599878787994,0.28591999411582947,0.25465920120477675,0.25385600328445435,0.01593204652619194,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.2295999974012375,0.26556798815727234,0.24674240052700042,0.2497600018978119,0.010345732066199003,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.21721599996089935,0.29020801186561584,0.2394208014011383,0.2346400022506714,0.018747330357768585,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2717759907245636,0.305184006690979,0.28813759982585907,0.28809599578380585,0.01183422600545854,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.3917759954929352,0.43772798776626587,0.41130879521369934,0.4131519943475723,0.012992473640805227,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.6176639795303345,0.7009919881820679,0.642767995595932,0.6330719888210297,0.024074084919938756,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.0820800065994263,1.1674879789352417,1.1034304022789,1.0977439880371094,0.0233067292981566,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.9809919595718384,2.0415360927581787,2.003715181350708,1.992751955986023,0.022066076434645737,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,3.790112018585205,3.8651199340820312,3.829139161109924,3.8230879306793213,0.025177464160110564,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.212351992726326,0.24383999407291412,0.22760000079870224,0.22723200172185898,0.01050568575837594,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.47494399547576904,0.5184000134468079,0.4920704007148743,0.49169600009918213,0.011991064701471855,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.3984375,0.0,1.3919410907075054,6.0,5.570159765557392,0.667236328125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.6360960006713867,0.7004479765892029,0.6608384013175964,0.6572319865226746,0.020416877242438597,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.625,0.0,1.0307764064044151,4.0,6.138251855282827,0.5382080078125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.780896008014679,0.8301439881324768,0.80346559882164,0.8030399978160858,0.016801230312128875,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.859375,0.0,0.6903350635742038,3.0,6.5943747091218174,0.38067626953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.8607040047645569,0.9195200204849243,0.8783008038997651,0.8751039803028107,0.01719059253115595,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.9765625,0.0,0.49410588440130926,2.75,6.814452474347134,0.271270751953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.833952009677887,0.894752025604248,0.8619967997074127,0.863215982913971,0.018716378851797198,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,1.0,0.25,0.33693529145074724,2.125,6.9186075263155535,0.1867218017578125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.834879994392395,0.8871039748191833,0.8651552021503448,0.8638879954814911,0.015262430894400969,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,1.0,0.375,0.25567294018677456,1.8125,6.952441049154937,0.14349365234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.8518080115318298,0.9097599983215332,0.8810272097587586,0.8751040101051331,0.017005819633271906,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,1.0,0.625,0.1747801353218523,1.53125,6.978069554482723,0.09820938110351562,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.8470079898834229,0.9010239839553833,0.8694015920162201,0.8716959953308105,0.016138881171190216,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,1.0,0.65625,0.1158122428154187,1.3125,6.9901908183358845,0.06445503234863281,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.1698240041732788,1.2311359643936157,1.1888479948043824,1.1890720129013062,0.017978797794492758,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,1.0,0.78125,0.08347181893108634,1.1796875,6.994921772573154,0.046871185302734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.7702720165252686,1.8097599744796753,1.7919103980064393,1.7956640124320984,0.012641295676021557,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,1.0,0.78125,0.06866734477822484,1.20703125,6.996602562938728,0.03801727294921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.968672037124634,3.0278079509735107,2.9899007797241213,2.9824799299240112,0.01816414122631667,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,1.0,0.8984375,0.04399546833977376,1.126953125,6.998607314922362,0.024699926376342773,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.19574399292469025,0.2512960135936737,0.21939200013875962,0.2199999988079071,0.017532156418212565,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.20483200252056122,0.24006399512290955,0.22215040028095245,0.22433599829673767,0.00969639786132892,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.20559999346733093,0.24726399779319763,0.22126719802618028,0.22207999974489212,0.01328093478485499,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.20003199577331543,0.2301120012998581,0.21453119963407516,0.21598400175571442,0.010402239855151332,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.19551999866962433,0.22972799837589264,0.21238719969987868,0.21488000452518463,0.010706095835489097,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.19420799612998962,0.2903999984264374,0.2211231991648674,0.21476799994707108,0.025955008886196004,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2290560007095337,0.289247989654541,0.2543327987194061,0.24939200282096863,0.01663236753503115,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.30831998586654663,0.36953601241111755,0.3324000000953674,0.3288639932870865,0.018617999572156707,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.462911993265152,0.5497919917106628,0.4893856018781662,0.4816960096359253,0.023348152887178286,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,0.7662079930305481,0.8717759847640991,0.788454395532608,0.7744799852371216,0.03174746482604812,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.363935947418213,1.4143040180206299,1.3812703967094422,1.3798720240592957,0.014770450075530007,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.5507519245147705,2.680704116821289,2.579859209060669,2.566223978996277,0.03673283558365955,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/allreduce.json b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/allreduce.json new file mode 100644 index 0000000..aae1f0a --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/allreduce.json @@ -0,0 +1,824 @@ +{ + "environment": [ + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + { + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm" + }, + "gpu": "NVIDIA H20", + "model": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B", + "torch_cuda": "12.9", + "torch_version": "2.11.0+cu129", + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + } + ], + "frontier_consumption": "diagnostic_only_in_base_profile_only_run; measured lookup requires a separate CC-backend injection ablation", + "rows": [ + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08288000151515007, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.4872959852218628, + "mean": 0.1310015980154276, + "median": 0.07679999992251396, + "min": 0.06217600032687187, + "std": 0.12790721677293843 + }, + { + "max": 0.4402880072593689, + "mean": 0.12842560112476348, + "median": 0.08288000151515007, + "min": 0.0655680000782013, + "std": 0.11220176524616535 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0793600007891655, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.12716799974441528, + "mean": 0.07871360033750534, + "median": 0.0759200006723404, + "min": 0.06032000109553337, + "std": 0.019331314939874535 + }, + { + "max": 0.12380799651145935, + "mean": 0.08059840016067028, + "median": 0.0793600007891655, + "min": 0.06217600032687187, + "std": 0.01687088356254092 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0713919997215271, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12697599828243256, + "mean": 0.0767391998320818, + "median": 0.07078400254249573, + "min": 0.05910399928689003, + "std": 0.018775178979463278 + }, + { + "max": 0.11430399864912033, + "mean": 0.07594559974968433, + "median": 0.0713919997215271, + "min": 0.06124800071120262, + "std": 0.015719922150631036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08056000247597694, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.1037760004401207, + "mean": 0.07954559996724128, + "median": 0.08056000247597694, + "min": 0.05955199897289276, + "std": 0.0135697420393132 + }, + { + "max": 0.10608000308275223, + "mean": 0.07971520014107228, + "median": 0.07593599706888199, + "min": 0.06028800085186958, + "std": 0.015319849772775456 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.0865279994904995, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.14470399916172028, + "mean": 0.09121599942445754, + "median": 0.0865279994904995, + "min": 0.06441599875688553, + "std": 0.024893837894277456 + }, + { + "max": 0.12438400089740753, + "mean": 0.08531199917197227, + "median": 0.08031999692320824, + "min": 0.06364800035953522, + "std": 0.01878029830059533 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07135999947786331, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.11753600090742111, + "mean": 0.07606079950928687, + "median": 0.07135999947786331, + "min": 0.05843200162053108, + "std": 0.01755519771639284 + }, + { + "max": 0.1103999987244606, + "mean": 0.07607359997928143, + "median": 0.07073600217700005, + "min": 0.05721599981188774, + "std": 0.016904445220949783 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.07321599870920181, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.11740799993276596, + "mean": 0.07749119997024537, + "median": 0.07203200086951256, + "min": 0.05862399935722351, + "std": 0.017701381594822835 + }, + { + "max": 0.11382400244474411, + "mean": 0.07733759954571724, + "median": 0.07321599870920181, + "min": 0.059039998799562454, + "std": 0.017188890882557036 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09025600180029869, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.13980799913406372, + "mean": 0.0950367994606495, + "median": 0.09025600180029869, + "min": 0.06815999746322632, + "std": 0.022829835127539378 + }, + { + "max": 0.14764800667762756, + "mean": 0.09710080176591873, + "median": 0.08720000088214874, + "min": 0.07152000069618225, + "std": 0.02500574954063789 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08083200082182884, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.15574400126934052, + "mean": 0.08760640025138855, + "median": 0.07846399769186974, + "min": 0.07097599655389786, + "std": 0.024487311423551025 + }, + { + "max": 0.16284799575805664, + "mean": 0.08963519930839539, + "median": 0.08083200082182884, + "min": 0.07100799679756165, + "std": 0.026047320562445356 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.10891199856996536, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.1582079976797104, + "mean": 0.11541439890861512, + "median": 0.10836799815297127, + "min": 0.09062399715185165, + "std": 0.018285136316576037 + }, + { + "max": 0.1578879952430725, + "mean": 0.11537599861621857, + "median": 0.10891199856996536, + "min": 0.0960640013217926, + "std": 0.018246538626977286 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1703840047121048, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.19327999651432037, + "mean": 0.1707327976822853, + "median": 0.1703840047121048, + "min": 0.14815999567508698, + "std": 0.014211056022719618 + }, + { + "max": 0.19276799261569977, + "mean": 0.1658592015504837, + "median": 0.16379200667142868, + "min": 0.14560000598430634, + "std": 0.013840249648693638 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.25539200007915497, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.2807359993457794, + "mean": 0.25750079900026324, + "median": 0.25539200007915497, + "min": 0.24624000489711761, + "std": 0.008925204570802302 + }, + { + "max": 0.2863999903202057, + "mean": 0.2585055992007256, + "median": 0.255280002951622, + "min": 0.24371199309825897, + "std": 0.012059738582656496 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 2 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1021759994328022, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1, + "payload_bytes": 4096, + "per_rank_time_ms": [ + { + "max": 0.9770879745483398, + "mean": 0.1974783968180418, + "median": 0.10099200159311295, + "min": 0.05913599953055382, + "std": 0.2660581738745569 + }, + { + "max": 0.892799973487854, + "mean": 0.18164799660444259, + "median": 0.1021759994328022, + "min": 0.06435199826955795, + "std": 0.23947520188197013 + }, + { + "max": 0.6467199921607971, + "mean": 0.15839359983801843, + "median": 0.10100800171494484, + "min": 0.06800000369548798, + "std": 0.16617013141866102 + }, + { + "max": 0.6725760102272034, + "mean": 0.15686400160193442, + "median": 0.10044800117611885, + "min": 0.06063999980688095, + "std": 0.17523222161300497 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12694399803876877, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8, + "payload_bytes": 32768, + "per_rank_time_ms": [ + { + "max": 0.6659520268440247, + "mean": 0.1927648030221462, + "median": 0.12694399803876877, + "min": 0.07609599828720093, + "std": 0.16697784531907736 + }, + { + "max": 0.695360004901886, + "mean": 0.19356480240821838, + "median": 0.11726400256156921, + "min": 0.0796160027384758, + "std": 0.17588096678867862 + }, + { + "max": 0.5939840078353882, + "mean": 0.1868800014257431, + "median": 0.12379200011491776, + "min": 0.07427199929952621, + "std": 0.1463231714943902 + }, + { + "max": 0.6635839939117432, + "mean": 0.1874335989356041, + "median": 0.12014400213956833, + "min": 0.07526399940252304, + "std": 0.16680959677760304 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09161599725484848, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 16, + "payload_bytes": 65536, + "per_rank_time_ms": [ + { + "max": 0.12665599584579468, + "mean": 0.09443839862942696, + "median": 0.09124799817800522, + "min": 0.06431999802589417, + "std": 0.0205690155775906 + }, + { + "max": 0.1303039938211441, + "mean": 0.09712959825992584, + "median": 0.09161599725484848, + "min": 0.07648000121116638, + "std": 0.019072048129173236 + }, + { + "max": 0.13836799561977386, + "mean": 0.09821119979023933, + "median": 0.09148800000548363, + "min": 0.0727040022611618, + "std": 0.021314066545189116 + }, + { + "max": 0.12992000579833984, + "mean": 0.09272959977388381, + "median": 0.08931199833750725, + "min": 0.06406400352716446, + "std": 0.021360803830354765 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08580800145864487, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 32, + "payload_bytes": 131072, + "per_rank_time_ms": [ + { + "max": 0.13449600338935852, + "mean": 0.08960640132427215, + "median": 0.08299200236797333, + "min": 0.06796800345182419, + "std": 0.020633597898445998 + }, + { + "max": 0.14735999703407288, + "mean": 0.09248319901525974, + "median": 0.08580800145864487, + "min": 0.05913599953055382, + "std": 0.025960234928829564 + }, + { + "max": 0.13705599308013916, + "mean": 0.08947199806571007, + "median": 0.08460799977183342, + "min": 0.0634239986538887, + "std": 0.02122838946992339 + }, + { + "max": 0.13846400380134583, + "mean": 0.0859104000031948, + "median": 0.08308799937367439, + "min": 0.05974400043487549, + "std": 0.022373631424433445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09867199882864952, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 64, + "payload_bytes": 262144, + "per_rank_time_ms": [ + { + "max": 0.1279039978981018, + "mean": 0.09748799875378608, + "median": 0.09678399935364723, + "min": 0.0655359998345375, + "std": 0.02315719144614622 + }, + { + "max": 0.1356479972600937, + "mean": 0.10018239840865135, + "median": 0.09532799944281578, + "min": 0.06185600161552429, + "std": 0.02330700253650042 + }, + { + "max": 0.13142399489879608, + "mean": 0.09778879955410957, + "median": 0.09492799639701843, + "min": 0.06560000032186508, + "std": 0.02307579212430045 + }, + { + "max": 0.1276479959487915, + "mean": 0.09611519873142242, + "median": 0.09867199882864952, + "min": 0.0642239972949028, + "std": 0.02273612181973431 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.09646400064229965, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 128, + "payload_bytes": 524288, + "per_rank_time_ms": [ + { + "max": 0.14060799777507782, + "mean": 0.09770880043506622, + "median": 0.09115200117230415, + "min": 0.06950400024652481, + "std": 0.024752645089849798 + }, + { + "max": 0.14377599954605103, + "mean": 0.09824960008263588, + "median": 0.08999999985098839, + "min": 0.07103999704122543, + "std": 0.025403407389046027 + }, + { + "max": 0.13680000603199005, + "mean": 0.09993600100278854, + "median": 0.09646400064229965, + "min": 0.06790400296449661, + "std": 0.022297424273985882 + }, + { + "max": 0.1391039937734604, + "mean": 0.09769919961690902, + "median": 0.09601600095629692, + "min": 0.06835199892520905, + "std": 0.023899922100804445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08377600088715553, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 256, + "payload_bytes": 1048576, + "per_rank_time_ms": [ + { + "max": 0.1464959979057312, + "mean": 0.09276160076260567, + "median": 0.08377600088715553, + "min": 0.06777600198984146, + "std": 0.024751086465295658 + }, + { + "max": 0.14319999516010284, + "mean": 0.09080640003085136, + "median": 0.080400001257658, + "min": 0.06796800345182419, + "std": 0.023365718881708488 + }, + { + "max": 0.1382399946451187, + "mean": 0.09063360020518303, + "median": 0.08193599805235863, + "min": 0.06627199798822403, + "std": 0.02372618650854302 + }, + { + "max": 0.14313599467277527, + "mean": 0.09044799953699112, + "median": 0.08128000050783157, + "min": 0.06652799993753433, + "std": 0.02462486862033686 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.1128000020980835, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 512, + "payload_bytes": 2097152, + "per_rank_time_ms": [ + { + "max": 0.1600320041179657, + "mean": 0.11426239982247352, + "median": 0.1128000020980835, + "min": 0.07657600194215775, + "std": 0.030240087702350687 + }, + { + "max": 0.15881599485874176, + "mean": 0.11206399947404862, + "median": 0.10628800094127655, + "min": 0.0772159993648529, + "std": 0.029933135345483627 + }, + { + "max": 0.15612800419330597, + "mean": 0.10761600062251091, + "median": 0.09860799834132195, + "min": 0.07689599692821503, + "std": 0.02634237020678647 + }, + { + "max": 0.15865600109100342, + "mean": 0.11094079986214637, + "median": 0.10979199782013893, + "min": 0.07583999633789062, + "std": 0.028619217028542445 + } + ], + "selected_backend": "flashinfer_trtllm", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.08755199983716011, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 1024, + "payload_bytes": 4194304, + "per_rank_time_ms": [ + { + "max": 0.11999999731779099, + "mean": 0.0843871995806694, + "median": 0.08755199983716011, + "min": 0.06332799792289734, + "std": 0.016638543890716024 + }, + { + "max": 0.1218239963054657, + "mean": 0.08518079966306687, + "median": 0.08032000064849854, + "min": 0.06393600255250931, + "std": 0.019900899429956945 + }, + { + "max": 0.11849600076675415, + "mean": 0.0843968003988266, + "median": 0.08702399954199791, + "min": 0.06297600269317627, + "std": 0.017031908773433545 + }, + { + "max": 0.12300799787044525, + "mean": 0.0846304003149271, + "median": 0.08139199763536453, + "min": 0.06195199862122536, + "std": 0.020106523698622265 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.12361599877476692, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 2048, + "payload_bytes": 8388608, + "per_rank_time_ms": [ + { + "max": 0.17155200242996216, + "mean": 0.12168959975242614, + "median": 0.12055999785661697, + "min": 0.09609600156545639, + "std": 0.022565485532483928 + }, + { + "max": 0.9246399998664856, + "mean": 0.19978560134768486, + "median": 0.12361599877476692, + "min": 0.09715200215578079, + "std": 0.24230694305662265 + }, + { + "max": 0.9317439794540405, + "mean": 0.20037759989500045, + "median": 0.12327999994158745, + "min": 0.09603200107812881, + "std": 0.24450903278000383 + }, + { + "max": 0.9321280121803284, + "mean": 0.19875840097665787, + "median": 0.12230399996042252, + "min": 0.0950080007314682, + "std": 0.24519252220785093 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.20030399411916733, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 4096, + "payload_bytes": 16777216, + "per_rank_time_ms": [ + { + "max": 0.3261440098285675, + "mean": 0.20826880186796187, + "median": 0.19974400103092194, + "min": 0.1409280002117157, + "std": 0.051397264855718945 + }, + { + "max": 0.3248000144958496, + "mean": 0.20548800230026246, + "median": 0.1979840025305748, + "min": 0.141184002161026, + "std": 0.04980002399656717 + }, + { + "max": 0.32547199726104736, + "mean": 0.21280319690704347, + "median": 0.20030399411916733, + "min": 0.14127999544143677, + "std": 0.05304243085685509 + }, + { + "max": 0.26047998666763306, + "mean": 0.1969312012195587, + "median": 0.18079999834299088, + "min": 0.14057600498199463, + "std": 0.042174123638424224 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + }, + { + "communicator": "vllm.tensor_model_parallel_all_reduce", + "critical_path_median_ms": 0.2924960106611252, + "dtype": "bfloat16", + "hidden_dim": 2048, + "num_tokens": 8192, + "payload_bytes": 33554432, + "per_rank_time_ms": [ + { + "max": 0.45052799582481384, + "mean": 0.3013375997543335, + "median": 0.28273600339889526, + "min": 0.21334399282932281, + "std": 0.07214223764527236 + }, + { + "max": 0.4466240108013153, + "mean": 0.2968191936612129, + "median": 0.27796798944473267, + "min": 0.21241599321365356, + "std": 0.07400678240849741 + }, + { + "max": 0.3830080032348633, + "mean": 0.2946112036705017, + "median": 0.2924960106611252, + "min": 0.21084800362586975, + "std": 0.05323627615746332 + }, + { + "max": 0.4609600007534027, + "mean": 0.3054272010922432, + "median": 0.289792001247406, + "min": 0.21062399446964264, + "std": 0.07608482904865672 + } + ], + "selected_backend": "nccl_fallback", + "tensor_parallel_size": 4 + } + ], + "schema_version": "qwen30_vllm020_allreduce_frozen.v1" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention.csv new file mode 100644 index 0000000..3220b8b --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention.csv @@ -0,0 +1,133 @@ +time_stats.attn_input_reshape.min,time_stats.attn_input_reshape.max,time_stats.attn_input_reshape.mean,time_stats.attn_input_reshape.median,time_stats.attn_input_reshape.std,time_stats.attn_kv_cache_save.min,time_stats.attn_kv_cache_save.max,time_stats.attn_kv_cache_save.mean,time_stats.attn_kv_cache_save.median,time_stats.attn_kv_cache_save.std,time_stats.attn_prefill.min,time_stats.attn_prefill.max,time_stats.attn_prefill.mean,time_stats.attn_prefill.median,time_stats.attn_prefill.std,time_stats.attn_decode.min,time_stats.attn_decode.max,time_stats.attn_decode.mean,time_stats.attn_decode.median,time_stats.attn_decode.std,time_stats.attn_output_reshape.min,time_stats.attn_output_reshape.max,time_stats.attn_output_reshape.mean,time_stats.attn_output_reshape.median,time_stats.attn_output_reshape.std,n_embd,n_q_head,n_kv_head,block_size,num_tensor_parallel_workers,max_model_len,batch_size,prefill_chunk_size,kv_cache_size,is_prefill,attention_backend,is_mixed_batch,mode,seq_lens,total_tokens,max_seq_len,min_seq_len,avg_seq_len,equal_seq_len,seq_len_variance,seq_len_std,seq_len_cv,is_chunked_prefill_sample,chunk_start_token,chunk_end_token,total_prefill_tokens,profiling_precision,model_arch,quant_signature,measurement_type,is_true_mixed_batch,prefill_seq_lens,prefill_kv_cache_sizes,decode_kv_cache_sizes,num_prefill_seqs,num_decode_seqs,decode_batch_size,total_batch_size,total_decode_tokens,decode_avg_kv_cache_size,batch_composition_ratio,batch_spec,projection_policy +0.0,0.0,0.0,0.0,0.0,0.01414399966597557,0.028863999992609024,0.019705599918961526,0.01771199982613325,0.005157200849836681,0.047968000173568726,0.07046400010585785,0.05810240097343922,0.05810240097343922,0.007477463486041561,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04947200044989586,0.020412799902260303,0.01635199971497059,0.010107497379722417,0.046560000628232956,0.08323200047016144,0.05587520003318787,0.05587520003318787,0.011126758739503428,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.022207999601960182,0.017033600155264138,0.015312000177800655,0.002819235991970241,0.05104000121355057,0.07692799717187881,0.056396800279617305,0.056396800279617305,0.007481982178637539,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015072000212967396,0.022272000089287758,0.01706880023702979,0.01616000011563301,0.002460889579319197,0.06931199878454208,0.0838719978928566,0.07432000041007995,0.07432000041007995,0.004777766433175866,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.018592000007629395,0.028543999418616295,0.02095999978482723,0.019183999858796597,0.003198175496053494,0.12179200351238251,0.15408000349998474,0.1307712011039257,0.1307712011039257,0.00858807797538298,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.027775999158620834,0.03385600075125694,0.030131200328469276,0.029680000618100166,0.0021152558103575215,0.32678401470184326,0.3450239896774292,0.33396480381488797,0.33396480381488797,0.0045872424917606375,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04438399896025658,0.05084799975156784,0.046540799736976626,0.04531199857592583,0.002277905811237223,1.0959680080413818,1.1151360273361206,1.0999775886535645,1.0999775886535645,0.005694403246120485,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.078015998005867,0.08691199868917465,0.08114239946007729,0.08019199967384338,0.00292795706334475,4.070400238037109,4.113152027130127,4.087088012695312,4.087088012695312,0.013660567012509554,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.05167999863624573,0.022115200012922286,0.017583999782800674,0.010340094822340818,0.05196800082921982,0.09011200070381165,0.06328320093452933,0.06328320093452933,0.012557341255467452,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01583999954164028,0.026623999699950218,0.018927999772131443,0.017376000061631203,0.003514650316260619,0.06681600213050842,0.07993599772453308,0.0725280001759529,0.0725280001759529,0.004343502558613716,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01648000068962574,0.030880000442266464,0.01945280022919178,0.017967999912798405,0.004096211183007485,0.1311360001564026,0.1546880006790161,0.13908160030841826,0.13908160030841826,0.007511906874366178,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.03558399900794029,0.020851199887692927,0.018559999763965607,0.005235911594130716,0.32950401306152344,0.350271999835968,0.33912960588932034,0.33912960588932034,0.006027400986663648,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.019328000023961067,0.040608000010252,0.022790400311350822,0.020704000256955624,0.006113051965778337,1.1415679454803467,1.1518720388412476,1.144483208656311,1.144483208656311,0.0032332311374389127,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015807999297976494,0.030688000842928886,0.019596799835562707,0.01774400006979704,0.004343384771033462,0.0,0.0,0.0,0.0,0.0,0.049056001007556915,0.07580800354480743,0.05948160067200661,0.05948160067200661,0.009031541471446955,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.02486399933695793,0.01923839971423149,0.018079999834299088,0.0032282528537266424,0.0,0.0,0.0,0.0,0.0,0.05142400041222572,0.07353600114583969,0.059328000620007516,0.059328000620007516,0.0073307807735143084,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016543999314308167,0.03977600112557411,0.021379199624061585,0.018511999398469925,0.006593576176246171,0.0,0.0,0.0,0.0,0.0,0.0488319993019104,0.06435199826955795,0.05479039996862411,0.05479039996862411,0.005672522998491864,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01635199971497059,0.02844800055027008,0.019267200119793416,0.017952000722289085,0.0035068687666949577,0.0,0.0,0.0,0.0,0.0,0.049855999648571014,0.07798399776220322,0.05986879989504815,0.05986879989504815,0.01043914754878828,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016383999958634377,0.026079999282956123,0.01923519968986511,0.017791999503970146,0.0032161974331284568,0.0,0.0,0.0,0.0,0.0,0.058111999183893204,0.1045759990811348,0.06708480007946492,0.06708480007946492,0.013479022462646494,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.04057599976658821,0.019100800156593323,0.015455999877303839,0.007512281243011577,0.0,0.0,0.0,0.0,0.0,0.05363199859857559,0.07782399654388428,0.06090559959411622,0.06090559959411622,0.007544620176348091,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.02191999927163124,0.016684799920767546,0.01563199982047081,0.0024293118621811216,0.0,0.0,0.0,0.0,0.0,0.0629120022058487,0.07891199737787247,0.06891520097851753,0.06891520097851753,0.005472695665695425,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.038943998515605927,0.018313600029796363,0.01561600062996149,0.007127270260115769,0.0,0.0,0.0,0.0,0.0,0.08675199747085571,0.10662399977445602,0.09391999915242194,0.09391999915242194,0.006988099589086635,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.054687999188899994,0.021439999900758268,0.01539199985563755,0.012052764849597775,0.0,0.0,0.0,0.0,0.0,0.13488000631332397,0.1528639942407608,0.1431359991431236,0.1431359991431236,0.005436271464033599,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.041919998824596405,0.01899839974939823,0.015343999955803156,0.007989843526623287,0.0,0.0,0.0,0.0,0.0,0.06176000088453293,0.08374399691820145,0.06747519969940186,0.06747519969940186,0.0066067747128778,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016256000846624374,0.11353600025177002,0.042761600017547606,0.028960000723600388,0.029104301538020762,0.0,0.0,0.0,0.0,0.0,0.09734400361776352,0.14422400295734406,0.11392960175871848,0.11392960175871848,0.013198594600417867,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.034143999218940735,0.018918400071561335,0.01643200032413006,0.005500943993080684,0.0,0.0,0.0,0.0,0.0,0.12918399274349213,0.15087999403476715,0.13807999789714814,0.13807999789714814,0.007658330538677587,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.016063999384641647,0.03641600161790848,0.0198208000510931,0.01780799962580204,0.0057264128169845765,0.0,0.0,0.0,0.0,0.0,0.22099199891090393,0.23904000222682953,0.2293503984808922,0.2293503984808922,0.004861342907006028,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.035840000957250595,0.018908800091594458,0.015792000107467175,0.0064374817924757475,0.0,0.0,0.0,0.0,0.0,0.10134399682283401,0.12201599776744843,0.10896319895982742,0.10896319895982742,0.006336330809165179,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.03417599946260452,0.017846399918198586,0.014800000004470348,0.006495007539635255,0.0,0.0,0.0,0.0,0.0,0.13126400113105774,0.15561600029468536,0.1389280006289482,0.1389280006289482,0.008381472811075022,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03519999980926514,0.019168000388890504,0.01600000075995922,0.005995522477654695,0.0,0.0,0.0,0.0,0.0,0.21728000044822693,0.2343679964542389,0.2231455981731415,0.2231455981731415,0.004720730646739123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.03670400008559227,0.018441599886864425,0.015023999847471714,0.006596535162793127,0.0,0.0,0.0,0.0,0.0,0.39321601390838623,0.4524799883365631,0.4058080047369003,0.4058080047369003,0.01578349755088941,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.022143999114632607,0.016912000067532063,0.015168000012636185,0.0028156089295136347,0.0,0.0,0.0,0.0,0.0,0.15587200224399567,0.3079040050506592,0.17838079929351805,0.17838079929351805,0.04355865575265927,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.02112000063061714,0.015516800060868263,0.01488000014796853,0.001940870731593904,0.0,0.0,0.0,0.0,0.0,0.21587200462818146,0.23561599850654602,0.22250880002975468,0.22250880002975468,0.006181951170646666,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.039264000952243805,0.023609600123018028,0.02131200022995472,0.007236979625711548,0.0,0.0,0.0,0.0,0.0,0.408735990524292,0.470335990190506,0.4336863994598388,0.4336863994598388,0.01844662383160074,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.025407999753952026,0.016227199975401164,0.014960000291466713,0.0031617375441736185,0.0,0.0,0.0,0.0,0.0,0.7412800192832947,0.7627840042114258,0.7464000046253203,0.7464000046253203,0.006112167448837547,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.02812799997627735,0.020652799773961304,0.021359999664127827,0.004308706957613102,0.028383498565450627,0.039859687970646644,0.032492258074592426,0.032492258074592426,0.00453597266208597,0.029312501176103633,0.04116431058787463,0.03355574193327539,0.03355574193327539,0.004684436757701648,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.034143999218940735,0.023171199765056372,0.024255999363958836,0.005908614918096971,0.03333159243114438,0.038935341782478095,0.03580428402241854,0.03580428402241854,0.002082270297044095,0.03633240903369937,0.04244065945634484,0.03902771507087562,0.03902771507087562,0.0022697354261490147,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.02611199952661991,0.019635199941694735,0.018112000077962875,0.0038298050749257795,0.04189529417991216,0.057484239920526384,0.04744885718421094,0.04744885718421094,0.004779830748455743,0.051672703037266184,0.07089975418195164,0.05852234134479412,0.05852234134479412,0.005895334539786378,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.029311999678611755,0.018662399891763926,0.01673599984496832,0.0044017112162725355,0.04322973959325901,0.05049827064705393,0.04535414343408448,0.04535414343408448,0.0022944156000240697,0.08733025617719538,0.10201372836398578,0.09162185574241775,0.09162185574241775,0.004635047631846023,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.026367999613285065,0.016336000058799982,0.01515199989080429,0.003415353455946402,0.06031842775160765,0.06618323188375198,0.06274415549817247,0.06274415549817247,0.001985647289644255,0.14768157653992678,0.16204076249052324,0.15362064543185072,0.15362064543185072,0.004861590945216247,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.02454400062561035,0.017167999967932702,0.016191999427974224,0.0028685016454498436,0.09128700688359712,0.09689150775996329,0.09360396051475776,0.09360396051475776,0.0013477542744916764,0.2740889887523892,0.2909164873209846,0.2810456356995366,0.2810456356995366,0.004046628526808561,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01881599985063076,0.03097599931061268,0.024598400108516216,0.024848000146448612,0.0037539437391565975,0.1035249255866932,0.10603132147437412,0.10399797220840973,0.10399797220840973,0.0007025109429056814,0.5652750707893444,0.5789606938874117,0.5678580377517648,0.5678580377517648,0.003835906384194897,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018624000251293182,0.043487999588251114,0.024460799992084503,0.019600000232458115,0.008922081629781394,0.196169204945307,0.2076140047945799,0.2008444429250931,0.2008444429250931,0.00394350953292805,1.1196708009293268,1.1849940417370974,1.1463555573610091,1.1463555573610091,0.022508285530529783,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.027135999873280525,0.03667199984192848,0.02945920005440712,0.028256000019609928,0.0029446678408169553,0.37757279619664613,0.3898113624476372,0.38075520430942184,0.38075520430942184,0.0036600203831042254,0.2522831942132049,0.2604606493092597,0.25440958703617433,0.25440958703617433,0.0024455194930253126,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.0435199998319149,0.049695998430252075,0.04502719938755036,0.04395199939608574,0.002035785660169308,1.1048984388245497,1.1189053886476108,1.1112371236754128,1.1112371236754128,0.0050220696724624985,0.1395495077239122,0.14131859607071193,0.14035008841031085,0.14035008841031085,0.0006342911944856293,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,1,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.02364799939095974,0.01809599995613098,0.016352000646293163,0.0035481127058959038,0.04822399839758873,0.08566399663686752,0.05961279980838299,0.05961279980838299,0.011445665413968877,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.020864000543951988,0.01668160008266568,0.015664000064134598,0.0025769063833097584,0.049695998430252075,0.08057600259780884,0.05882879942655563,0.05882879942655563,0.009515126108519331,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.028704000636935234,0.017430400010198355,0.015056000091135502,0.004349294613335555,0.049855999648571014,0.07366400212049484,0.05459520071744919,0.05459520071744919,0.0069610920757925574,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.03161599859595299,0.017788799852132796,0.015711999963968992,0.005005385723318659,0.06102399900555611,0.08179199695587158,0.06650560013949873,0.06650560013949873,0.006947995595801105,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.04416000097990036,0.019200000166893005,0.015488000120967627,0.008561241323364038,0.08054400235414505,0.09196799993515015,0.08607039973139763,0.08607039973139763,0.004145329035483754,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017855999991297722,0.023391999304294586,0.019667199812829494,0.018463999964296818,0.0022577669687832585,0.18729600310325623,0.20585599541664124,0.19546559900045393,0.19546559900045393,0.006339068824303663,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.026240000501275063,0.03446400165557861,0.028297600522637367,0.027088000439107418,0.0025148441522922374,0.5754240155220032,0.5889919996261597,0.5800191938877105,0.5800191938877105,0.003858869273829596,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.04182400181889534,0.047200001776218414,0.043036799877882004,0.0423360001295805,0.0017073405772076728,2.063199996948242,2.0787200927734375,2.067151999473572,2.067151999473572,0.004959651271127835,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.05648000165820122,0.02270399993285537,0.017935999669134617,0.012377915150727689,0.049536000937223434,0.07196799665689468,0.056015999615192415,0.056015999615192415,0.0070476637552742884,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01408000010997057,0.021344000473618507,0.01611520005390048,0.014928000047802925,0.0025499884993961702,0.07072000205516815,0.2642880082130432,0.1588256008923054,0.1588256008923054,0.053220347086102376,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01360000018030405,0.03014400042593479,0.016975999902933837,0.015056000091135502,0.004715159697364111,0.08393599838018417,0.11036799848079681,0.09160000011324881,0.09160000011324881,0.007911669434472792,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014495999552309513,0.020767999812960625,0.016233599931001663,0.015392000321298838,0.002202615907995427,0.1998399943113327,0.22070400416851044,0.20855360180139543,0.20855360180139543,0.00689307200230667,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.033440001308918,0.018684800155460833,0.016048000194132328,0.00553991005639174,0.6043199896812439,0.635807991027832,0.6126143991947173,0.6126143991947173,0.008745933953408096,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.03359999880194664,0.017811199743300678,0.014944000169634819,0.005926140483565472,0.0,0.0,0.0,0.0,0.0,0.045471999794244766,0.07539200037717819,0.054758400097489356,0.054758400097489356,0.010253548506101549,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02409599907696247,0.01641279999166727,0.01473599998280406,0.003347158638713987,0.0,0.0,0.0,0.0,0.0,0.0461760014295578,0.07529599964618683,0.05460800044238568,0.05460800044238568,0.009748937798340135,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.028416000306606293,0.018611199874430894,0.01598400017246604,0.005209875093863647,0.0,0.0,0.0,0.0,0.0,0.048128001391887665,0.07897599786520004,0.061353600397706036,0.061353600397706036,0.010488153655157845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.025728000327944756,0.016092800162732603,0.01512000011280179,0.003300888123052605,0.0,0.0,0.0,0.0,0.0,0.04864000156521797,0.07648000121116638,0.05810560062527656,0.05810560062527656,0.009473544218404408,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014655999839305878,0.03551999852061272,0.018588799890130757,0.016064000315964222,0.006071057437992447,0.0,0.0,0.0,0.0,0.0,0.04822399839758873,0.07862400263547897,0.05660480037331582,0.05660480037331582,0.009565394213730401,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014303999952971935,0.028831999748945236,0.01699519995599985,0.015024000313133001,0.004285000744868188,0.0,0.0,0.0,0.0,0.0,0.04854400083422661,0.06719999760389328,0.05778240002691746,0.05778240002691746,0.006852125554679805,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.0144640002399683,0.024927999824285507,0.0161183999851346,0.01521599991247058,0.0029774808524673907,0.0,0.0,0.0,0.0,0.0,0.05379199981689453,0.08966399729251862,0.06270079985260964,0.06270079985260964,0.009983591277092696,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014431999996304512,0.03001599945127964,0.017648000083863736,0.01547200046479702,0.004585126309484848,0.0,0.0,0.0,0.0,0.0,0.061664000153541565,0.07692799717187881,0.06704320013523103,0.06704320013523103,0.005500191798490629,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.026367999613285065,0.016947199776768684,0.015039999969303608,0.0037951566103550205,0.0,0.0,0.0,0.0,0.0,0.08799999952316284,0.111455999314785,0.0964031994342804,0.0964031994342804,0.007397541615558088,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01369599997997284,0.021247999742627144,0.015359999984502793,0.014640000183135271,0.0020942770950814317,0.0,0.0,0.0,0.0,0.0,0.051711998879909515,0.07065600156784058,0.058054400235414506,0.058054400235414506,0.006633034815910025,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.0307839997112751,0.017148799914866685,0.015039999969303608,0.004882374686312284,0.0,0.0,0.0,0.0,0.0,0.061919998377561576,0.07843200117349625,0.06628479920327664,0.06628479920327664,0.004962852689801192,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013887999579310417,0.02969600073993206,0.017500799987465142,0.015232000034302473,0.00467273319705314,0.0,0.0,0.0,0.0,0.0,0.08819200098514557,0.11097600311040878,0.09493440166115762,0.09493440166115762,0.007509235042985577,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.022112000733613968,0.01751680001616478,0.016047999262809753,0.0030306621792915785,0.0,0.0,0.0,0.0,0.0,0.13065600395202637,0.15110400319099426,0.13857279866933822,0.13857279866933822,0.00750137841249771,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.04012800008058548,0.01892479993402958,0.01550400024279952,0.007459545267816961,0.0,0.0,0.0,0.0,0.0,0.06278400123119354,0.08259200304746628,0.0704512007534504,0.0704512007534504,0.005979055984382744,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.02236800082027912,0.016332800220698118,0.014928000047802925,0.002784044363186731,0.0,0.0,0.0,0.0,0.0,0.1003199964761734,0.1279360055923462,0.10921279862523078,0.10921279862523078,0.00862769617046716,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0225600004196167,0.016128000058233737,0.014800000004470348,0.002958953332547708,0.0,0.0,0.0,0.0,0.0,0.13116799294948578,0.14812800288200378,0.13783999979496003,0.13783999979496003,0.005696384361148053,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014208000153303146,0.029983999207615852,0.017468800116330386,0.01508800033479929,0.0047379550962483065,0.0,0.0,0.0,0.0,0.0,0.217631995677948,0.2447360008955002,0.22715839892625808,0.22715839892625808,0.008831828221847138,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014047999866306782,0.02304000034928322,0.016512000095099212,0.014512000139802694,0.0035026774828624254,0.0,0.0,0.0,0.0,0.0,0.11020799726247787,0.12307199835777283,0.11600959971547126,0.11600959971547126,0.004667637266950902,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014112000353634357,0.042080000042915344,0.017510399967432023,0.014607999939471483,0.00823117883530167,0.0,0.0,0.0,0.0,0.0,0.15702399611473083,0.17587199807167053,0.16399359852075576,0.16399359852075576,0.006588299074676393,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.013919999822974205,0.0208320003002882,0.015299199987202883,0.01462399959564209,0.001916768086505386,0.0,0.0,0.0,0.0,0.0,0.21779200434684753,0.2415360063314438,0.2260768011212349,0.2260768011212349,0.007251352080685236,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014015999622642994,0.021088000386953354,0.015619200188666582,0.01473599998280406,0.002013227452302141,0.0,0.0,0.0,0.0,0.0,0.3959999978542328,0.4152640104293823,0.40332479774951924,0.40332479774951924,0.006942401431914052,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014175999909639359,0.03379200026392937,0.020595200080424547,0.019600000232458115,0.006548754248881344,0.026192623739694512,0.040006882507168426,0.029155180178492036,0.029155180178492036,0.00408828748438241,0.024591375524545756,0.037561119537986146,0.027372820355088746,0.027372820355088746,0.0038383559348575944,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.020896000787615776,0.015318400040268899,0.014431999996304512,0.0019640312960926966,0.029349018208693862,0.06236262941356679,0.03714313592014963,0.03714313592014963,0.009618313791872569,0.028826981462526914,0.06125337308649041,0.036482463672774496,0.036482463672774496,0.009447230957011863,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.032127998769283295,0.017286399938166143,0.014479999896138906,0.005536495446636193,0.03273085874558354,0.04394578491937082,0.03563837490653034,0.03563837490653034,0.003429601730256567,0.03488514202593899,0.04683821345079978,0.037984025407085474,0.037984025407085474,0.0036553316361902684,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013856000266969204,0.021503999829292297,0.015078400075435639,0.01425600005313754,0.002238435975478849,0.042100813549974185,0.051784144690147756,0.045378692890289625,0.045378692890289625,0.003184109940650555,0.05111518843748309,0.06287185663927425,0.05509490773570219,0.05509490773570219,0.0038658725544299132,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.02687999978661537,0.017667199857532977,0.01566399959847331,0.003864811846387173,0.048475323773821306,0.05599956978723733,0.05123499252968345,0.05123499252968345,0.002427379820423941,0.08429268135885387,0.09737642835214408,0.0890914090616175,0.0890914090616175,0.0042209177332077005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01462399959564209,0.03359999880194664,0.019804799742996693,0.016032000072300434,0.006750519908145474,0.07121508474579985,0.07292308367136396,0.07194410845270183,0.07194410845270183,0.0005500065915943844,0.14760091249712767,0.15114092373010238,0.14911189243564577,0.14911189243564577,0.0011399477384397005,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.021856000646948814,0.016835200227797033,0.015263999812304974,0.00283854448975065,0.08146338272142935,0.08560865714384096,0.0834932625520734,0.0834932625520734,0.0012259635905347874,0.2782486219401307,0.29240733788179385,0.2851819365989658,0.2851819365989658,0.004187435731481665,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014399999752640724,0.04495999962091446,0.02143679987639189,0.015856000129133463,0.009709987872683342,0.1194459208702178,0.12302524755181463,0.12053885718696096,0.12053885718696096,0.001005118119341339,0.5597220650459199,0.5764947443228802,0.5648435507167102,0.5648435507167102,0.004709970715400788,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.018112000077962875,0.026496000587940216,0.019318400137126445,0.018432000651955605,0.0024249605112359905,0.19770253574610402,0.222811797868348,0.2054811520619412,0.2054811520619412,0.008000944254868043,0.13941746080159492,0.157124211777114,0.14490284788179197,0.14490284788179197,0.005642170080515992,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.02595200017094612,0.03155200183391571,0.02727359998971224,0.026575999334454536,0.0016260948764843166,0.6014684881116659,0.6095472988212,0.6046757700946376,0.6046757700946376,0.0023537435563177303,0.11325152264578045,0.11477269562841545,0.11385542721485654,0.11385542721485654,0.0004431903698023628,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,2,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014560000039637089,0.022784000262618065,0.016435200069099664,0.015519999898970127,0.0023688301421469523,0.04879999905824661,0.09139200299978256,0.06054079942405224,0.06054079942405224,0.012152608702448775,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,False,0.0,64.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014816000126302242,0.02844800055027008,0.017008000146597625,0.015696000307798386,0.003941466294662679,0.047807998955249786,0.07356800138950348,0.05626560002565384,0.05626560002565384,0.00842179125412236,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,False,0.0,128.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.053408000618219376,0.019353600218892097,0.01532800029963255,0.01138302400841626,0.048448000103235245,0.0785600021481514,0.0556256003677845,0.0556256003677845,0.009348274502616908,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,False,0.0,256.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.03139200061559677,0.01823679991066456,0.016159999649971724,0.004860071238302512,0.055424001067876816,0.08505599945783615,0.0640383992344141,0.0640383992344141,0.00921448636178623,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,False,0.0,512.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014751999638974667,0.026528000831604004,0.017324799951165915,0.01536000007763505,0.0036004822686428305,0.07660800218582153,0.0942080020904541,0.08209280073642732,0.08209280073642732,0.00515895587669752,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,False,0.0,1024.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.021247999742627144,0.016403199825435876,0.01532800029963255,0.001995897022647934,0.11395200341939926,0.15113599598407745,0.12431039959192276,0.12431039959192276,0.011164431123683732,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,2048,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[2048],2048,2048,2048,2048.0,True,0.0,0.0,0.0,False,0.0,2048.0,2048,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.017184000462293625,0.028672000393271446,0.019865600019693376,0.017823999747633934,0.0035798491315929977,0.3171840012073517,0.3341119885444641,0.3261695951223373,0.3261695951223373,0.005111046452878918,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,4096,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[4096],4096,4096,4096,4096.0,True,0.0,0.0,0.0,False,0.0,4096.0,4096,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.024639999493956566,0.030400000512599945,0.02656640000641346,0.02556800004094839,0.0020189846603237303,1.0648640394210815,1.0828479528427124,1.0715327858924866,1.0715327858924866,0.005558639263049851,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,8192,0.0,True,FLASH_ATTN,False,vllm020_batch_spec,[8192],8192,8192,8192,8192.0,True,0.0,0.0,0.0,False,0.0,8192.0,8192,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014976000413298607,0.021695999428629875,0.017305599898099898,0.01593599934130907,0.0025718046517268054,0.04956800118088722,0.07932800054550171,0.06228480041027069,0.06228480041027069,0.01027160349757827,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,64,448.0,True,FLASH_ATTN,False,vllm020_batch_spec,[64],64,64,64,64.0,True,0.0,0.0,0.0,True,448.0,512.0,64,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q64s512,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01539199985563755,0.03580800071358681,0.019686400331556796,0.017136000096797943,0.005918387811304003,0.05158400163054466,0.10713600367307663,0.06364160068333148,0.06364160068333148,0.015832957809696766,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,128,896.0,True,FLASH_ATTN,False,vllm020_batch_spec,[128],128,128,128,128.0,True,0.0,0.0,0.0,True,896.0,1024.0,128,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q128s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02223999984562397,0.016672000009566545,0.015887999907135963,0.0020934945946034563,0.06521599739789963,0.08902399986982346,0.07520959973335266,0.07520959973335266,0.007913840684102929,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,256,1792.0,True,FLASH_ATTN,False,vllm020_batch_spec,[256],256,256,256,256.0,True,0.0,0.0,0.0,True,1792.0,2048.0,256,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q256s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015231999568641186,0.04028800129890442,0.019971200078725816,0.01646399963647127,0.007351305886577286,0.14467200636863708,0.16844800114631653,0.15363519936800005,0.15363519936800005,0.008174435899956223,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,512,3584.0,True,FLASH_ATTN,False,vllm020_batch_spec,[512],512,512,512,512.0,True,0.0,0.0,0.0,True,3584.0,4096.0,512,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q512s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015519999898970127,0.02304000034928322,0.01775679988786578,0.016784000210464,0.0025077243712082584,0.33740800619125366,0.35343998670578003,0.3445120006799698,0.3445120006799698,0.00445648463590358,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,1024,7168.0,True,FLASH_ATTN,False,vllm020_batch_spec,[1024],1024,1024,1024,1024.0,True,0.0,0.0,0.0,True,7168.0,8192.0,1024,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1ks8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.02393599972128868,0.01809599995613098,0.01654400024563074,0.002998393102466254,0.0,0.0,0.0,0.0,0.0,0.0504320003092289,0.0843840017914772,0.060083200410008426,0.060083200410008426,0.00986959318572296,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,1,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,[1],1,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01603199914097786,0.030047999694943428,0.019510399922728537,0.017680000513792038,0.004065185265963332,0.0,0.0,0.0,0.0,0.0,0.05004800111055374,0.07036799937486649,0.059315200522542,0.059315200522542,0.006537768821329647,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01484800036996603,0.02393599972128868,0.018035199772566558,0.016512000001966953,0.0030667689116777724,0.0,0.0,0.0,0.0,0.0,0.05100800096988678,0.06735999882221222,0.058387200161814694,0.058387200161814694,0.005832787739241381,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.026208000257611275,0.017500799987465142,0.01646399963647127,0.0030666103306165714,0.0,0.0,0.0,0.0,0.0,0.05023999884724617,0.07254400104284286,0.057254400476813315,0.057254400476813315,0.006890853653068151,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.027327999472618103,0.0179776000790298,0.01648000068962574,0.003783417447531392,0.0,0.0,0.0,0.0,0.0,0.04819199815392494,0.07100799679756165,0.05621119923889638,0.05621119923889638,0.007824391484124725,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,128.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s128,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015647999942302704,0.036448001861572266,0.019136000238358975,0.016704000532627106,0.006076530095624803,0.0,0.0,0.0,0.0,0.0,0.047488000243902206,0.08441600203514099,0.0588383998721838,0.0588383998721838,0.011275940726332522,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015135999768972397,0.033952001482248306,0.02119360016658902,0.018000000156462193,0.007136646614305304,0.0,0.0,0.0,0.0,0.0,0.04931199923157692,0.06992000341415405,0.05755840018391609,0.05755840018391609,0.007297587694315226,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014879999682307243,0.034272000193595886,0.0204415999352932,0.017311999574303627,0.00695404701803013,0.0,0.0,0.0,0.0,0.0,0.05215999856591225,0.0735040009021759,0.06117440015077591,0.06117440015077591,0.007381118436894922,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03743999823927879,0.02012479966506362,0.01775999926030636,0.006181045509079057,0.0,0.0,0.0,0.0,0.0,0.06355199962854385,0.08508799970149994,0.07019200026988984,0.07019200026988984,0.0062246580442117845,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,1024.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s1k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015552000142633915,0.032607998698949814,0.01959999995306134,0.017487999983131886,0.004882475068460749,0.0,0.0,0.0,0.0,0.0,0.04918399825692177,0.0865280032157898,0.05973760038614274,0.05973760038614274,0.011546847942113974,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015168000012636185,0.02675200067460537,0.017430400010198355,0.016560000367462635,0.003243135094239819,0.0,0.0,0.0,0.0,0.0,0.052671998739242554,0.08367999643087387,0.06238719932734965,0.06238719932734965,0.011207824780630104,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.01500799972563982,0.03612799942493439,0.019327999837696553,0.01688000001013279,0.006058251425153085,0.0,0.0,0.0,0.0,0.0,0.06364800035953522,0.07878399640321732,0.06935679838061332,0.06935679838061332,0.004515588328677643,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.03711999952793121,0.019670399930328132,0.017152000218629837,0.006165994646522264,0.0,0.0,0.0,0.0,0.0,0.09071999788284302,0.11507199704647064,0.10252480059862136,0.10252480059862136,0.008782544051535657,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,2048.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s2k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015039999969303608,0.026528000831604004,0.018396800104528665,0.01601599995046854,0.004279788048986283,0.0,0.0,0.0,0.0,0.0,0.05331199988722801,0.07977599650621414,0.06076480001211167,0.06076480001211167,0.008761579122069606,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.02643200010061264,0.01726400004699826,0.015855999663472176,0.003210008417242029,0.0,0.0,0.0,0.0,0.0,0.062431998550891876,0.08246400207281113,0.06970879957079888,0.06970879957079888,0.007016341676068233,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.02223999984562397,0.0166015999391675,0.015887999907135963,0.00204913969129354,0.0,0.0,0.0,0.0,0.0,0.10127999633550644,0.1141119971871376,0.10621120035648347,0.10621120035648347,0.004029318311754605,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.03363199904561043,0.019305599946528675,0.016671999357640743,0.0055445582307981234,0.0,0.0,0.0,0.0,0.0,0.1319040060043335,0.15839999914169312,0.14040640145540234,0.14040640145540234,0.007998041073596942,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,4096.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s4k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.04447999969124794,0.018908800091594458,0.0157279996201396,0.0086814937461721,0.0,0.0,0.0,0.0,0.0,0.06428799778223038,0.08982399851083755,0.07349760085344315,0.07349760085344315,0.008605284512197258,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,8,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1]",8,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,8q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.015104000456631184,0.028672000393271446,0.01890560006722808,0.016080000437796116,0.004797584627815021,0.0,0.0,0.0,0.0,0.0,0.11123199760913849,0.14115199446678162,0.11942399889230729,0.11942399889230729,0.008513394706791027,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,16,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",16,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,16q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014271999709308147,0.024927999824285507,0.016915200091898442,0.015216000378131866,0.003374147499442635,0.0,0.0,0.0,0.0,0.0,0.15887999534606934,0.18111999332904816,0.16934399753808976,0.16934399753808976,0.007415181260761123,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,32,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",32,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,32q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014336000196635723,0.026944000273942947,0.017737600207328796,0.015199999790638685,0.00415598714375255,0.0,0.0,0.0,0.0,0.0,0.2192319929599762,0.23472000658512115,0.22809920012950893,0.22809920012950893,0.004730841376327335,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,64,0,8192.0,False,FLASH_ATTN,False,vllm020_batch_spec,"[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]",64,1,1,1.0,True,0.0,0.0,0.0,False,0,0,0,BF16,generic,none,CUDA_EVENT,False,,,,,,,,,,,64q1s8k,measured_FA3_core_plus_measured_KV;reshape_assumed_zero;mean_as_median +0.0,0.0,0.0,0.0,0.0,0.014688000082969666,0.05052800104022026,0.020320000313222408,0.015520000364631414,0.010604522177924678,0.024270629882498958,0.03340247625954076,0.028446344104128624,0.028446344104128624,0.0032330369099793834,0.023697370291069768,0.03261352722995356,0.027774456325453972,0.027774456325453972,0.0031566742680923386,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,72,,,,,,,,False,,,64,BF16,generic,none,CUDA_EVENT,True,[64],[0],"[512, 512, 512, 512, 512, 512, 512, 512]",1,8,8,9,8,512.0,0.1111111111111111,q64_8q1s512,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.015359999611973763,0.033663999289274216,0.019987199828028678,0.018240000121295452,0.005522929139253732,0.0259194055660947,0.03719755183990719,0.029916029687899703,0.029916029687899703,0.003966666590554318,0.027104595853449518,0.03889844644729374,0.03128396954022015,0.03128396954022015,0.004148046317967881,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,9,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,136,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,8,8,9,8,1024.0,0.1111111111111111,q128_8q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014592000283300877,0.03299200162291527,0.019840000104159115,0.016207999549806118,0.006546343008726814,0.02587869595769926,0.03763167265431482,0.030174939058162802,0.030174939058162802,0.0037303581782277364,0.026473304070195713,0.03849632587654989,0.03086826083864964,0.03086826083864964,0.003816069654529222,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,144,,,,,,,,False,,,128,BF16,generic,none,CUDA_EVENT,True,[128],[0],"[1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024, 1024]",1,16,16,17,16,1024.0,0.058823529411764705,q128_16q1s1k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.013824000023305416,0.021023999899625778,0.016304000187665223,0.015584000386297703,0.0023236165806545476,0.031810621525966996,0.04156949936878106,0.0346749350032807,0.0346749350032807,0.003130209181143985,0.035677378270900374,0.04662250161636451,0.038889864871740294,0.038889864871740294,0.0035107033960828727,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,272,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,16,16,17,16,2048.0,0.058823529411764705,q256_16q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.02175999991595745,0.01569600012153387,0.015008000191301107,0.0020882666168036863,0.038211712107062853,0.07212229256520057,0.04364509673334097,0.04364509673334097,0.009671953074025085,0.0476442859917874,0.08992570455184198,0.05441890342616107,0.05441890342616107,0.01205947791784038,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,288,,,,,,,,False,,,256,BF16,generic,none,CUDA_EVENT,True,[256],[0],"[2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048, 2048]",1,32,32,33,32,2048.0,0.030303030303030304,q256_32q1s2k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014368000440299511,0.033824000507593155,0.017628800217062236,0.014864000026136637,0.005791495403857738,0.05214261250030033,0.06266261508706669,0.05677791295527661,0.05677791295527661,0.0030298223079651514,0.08648138506878382,0.10392938682791132,0.09416928531647478,0.09416928531647478,0.005025126612204489,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,544,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q512_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014527999795973301,0.031199999153614044,0.017900799959897996,0.015343999955803156,0.004974475661316249,0.06411958891421111,0.07405276123263956,0.06793047918211377,0.06793047918211377,0.0033918658556700006,0.14058441263169497,0.16236323591492058,0.1489399211274489,0.1489399211274489,0.00743678300375353,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,576,,,,,,,,False,,,512,BF16,generic,none,CUDA_EVENT,True,[512],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,64,64,65,64,4096.0,0.015384615384615385,q512_64q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014720000326633453,0.03855999931693077,0.018495999928563833,0.015647999942302704,0.006918530056761627,0.09872985549401277,0.10683454583043753,0.10185316839884552,0.10185316839884552,0.0020802254037042074,0.2743261390166379,0.2968454508984596,0.2830044295482752,0.2830044295482752,0.005780016596065092,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,65,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,1088,,,,,,,,False,,,1024,BF16,generic,none,CUDA_EVENT,True,[1024],[0],"[8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192]",1,64,64,65,64,8192.0,0.015384615384615385,q1k_64q1s8k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.014911999925971031,0.030400000512599945,0.01977920001372695,0.01726400014013052,0.005350883464206169,0.13918872472233365,0.16279522855335207,0.1501912864839173,0.1501912864839173,0.008992185529011513,0.11892328861766266,0.13909276049083735,0.1283239123428725,0.1283239123428725,0.007682951884956939,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,33,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,2080,,,,,,,,False,,,2048,BF16,generic,none,CUDA_EVENT,True,[2048],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,32,32,33,32,4096.0,0.030303030303030304,q2k_32q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio +0.0,0.0,0.0,0.0,0.0,0.01727999933063984,0.02223999984562397,0.01819519978016615,0.017680000513792038,0.0014397298270620873,0.3728044181625443,0.38295504353701676,0.3761264483787333,0.3761264483787333,0.0033660272332490977,0.07967557017401883,0.08184495665371805,0.08038555277807365,0.08038555277807365,0.0007193856241088455,0.0,0.0,0.0,0.0,0.0,2048,32,4,16,4,40960,17,0,0,True,FLASH_ATTN,False,true_mixed_fused_projected,,4112,,,,,,,,False,,,4096,BF16,generic,none,CUDA_EVENT,True,[4096],[0],"[4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096, 4096]",1,16,16,17,16,4096.0,0.058823529411764705,q4k_16q1s4k,fused_total_conserving_projection_by_same_tp_pure_prefill_decode_reference_ratio diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention_true_mixed_fused.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention_true_mixed_fused.csv new file mode 100644 index 0000000..153bb52 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/attention_true_mixed_fused.csv @@ -0,0 +1,31 @@ +num_tensor_parallel_workers,batch_spec,num_prefill_seqs,num_decode_seqs,total_prefill_tokens,total_decode_tokens,decode_avg_kv_cache_size,attention_core_mean_ms,attention_core_mean_as_median_ms,kv_cache_update_median_ms,pure_prefill_reference_mean_ms,pure_decode_reference_mean_ms,projected_prefill_mean_ms,projected_decode_mean_ms,projection_sum_error_ms,representation +1,q64_8q1s512,1,8,64,8,512.0,0.06604800000786781,0.06604800000786781,0.021359999664127827,0.05810240097343922,0.0600041144660541,0.032492258074592426,0.03355574193327539,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_8q1s1k,1,8,128,8,1024.0,0.07483199909329416,0.07483199909329416,0.024255999363958836,0.05587520003318787,0.06090559959411622,0.03580428402241854,0.03902771507087562,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q128_16q1s1k,1,16,128,16,1024.0,0.10597119852900506,0.10597119852900506,0.018112000077962875,0.05587520003318787,0.06891520097851753,0.04744885718421094,0.05852234134479412,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_16q1s2k,1,16,256,16,2048.0,0.13697599917650224,0.13697599917650224,0.01673599984496832,0.056396800279617305,0.11392960175871848,0.04535414343408448,0.09162185574241775,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q256_32q1s2k,1,32,256,32,2048.0,0.2163648009300232,0.2163648009300232,0.01515199989080429,0.056396800279617305,0.13807999789714814,0.06274415549817247,0.15362064543185072,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_32q1s4k,1,32,512,32,4096.0,0.3746495962142944,0.3746495962142944,0.016191999427974224,0.07432000041007995,0.2231455981731415,0.09360396051475776,0.2810456356995366,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q512_64q1s4k,1,64,512,64,4096.0,0.6718560099601746,0.6718560099601746,0.024848000146448612,0.07432000041007995,0.4058080047369003,0.10399797220840973,0.5678580377517648,-1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q1k_64q1s8k,1,64,1024,64,8192.0,1.3472000002861022,1.3472000002861022,0.019600000232458115,0.1307712011039257,0.7464000046253203,0.2008444429250931,1.1463555573610091,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q2k_32q1s4k,1,32,2048,32,4096.0,0.6351647913455962,0.6351647913455962,0.028256000019609928,0.33396480381488797,0.2231455981731415,0.38075520430942184,0.25440958703617433,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +1,q4k_16q1s4k,1,16,4096,16,4096.0,1.2515872120857237,1.2515872120857237,0.04395199939608574,1.0999775886535645,0.1389280006289482,1.1112371236754128,0.14035008841031085,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q64_8q1s512,1,8,64,8,512.0,0.05652800053358078,0.05652800053358078,0.019600000232458115,0.05961279980838299,0.055968457407185014,0.029155180178492036,0.027372820355088746,6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_8q1s1k,1,8,128,8,1024.0,0.07362559959292413,0.07362559959292413,0.014431999996304512,0.05882879942655563,0.05778240002691746,0.03714313592014963,0.036482463672774496,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q128_16q1s1k,1,16,128,16,1024.0,0.0736224003136158,0.0736224003136158,0.014479999896138906,0.05882879942655563,0.06270079985260964,0.03563837490653034,0.037984025407085474,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_16q1s2k,1,16,256,16,2048.0,0.10047360062599181,0.10047360062599181,0.01425600005313754,0.05459520071744919,0.06628479920327664,0.045378692890289625,0.05509490773570219,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q256_32q1s2k,1,32,256,32,2048.0,0.14032640159130094,0.14032640159130094,0.01566399959847331,0.05459520071744919,0.09493440166115762,0.05123499252968345,0.0890914090616175,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_32q1s4k,1,32,512,32,4096.0,0.22105600088834762,0.22105600088834762,0.016032000072300434,0.06650560013949873,0.13783999979496003,0.07194410845270183,0.14911189243564577,-2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q512_64q1s4k,1,64,512,64,4096.0,0.36867519915103913,0.36867519915103913,0.015263999812304974,0.06650560013949873,0.22715839892625808,0.0834932625520734,0.2851819365989658,5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q1k_64q1s8k,1,64,1024,64,8192.0,0.6853824079036711,0.6853824079036711,0.015856000129133463,0.08607039973139763,0.40332479774951924,0.12053885718696096,0.5648435507167102,1.1102230246251565e-16,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q2k_32q1s4k,1,32,2048,32,4096.0,0.3503839999437332,0.3503839999437332,0.018432000651955605,0.19546559900045393,0.13783999979496003,0.2054811520619412,0.14490284788179197,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +2,q4k_16q1s4k,1,16,4096,16,4096.0,0.7185311973094941,0.7185311973094941,0.026575999334454536,0.5800191938877105,0.10921279862523078,0.6046757700946376,0.11385542721485654,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q64_8q1s512,1,8,64,8,512.0,0.0562208004295826,0.0562208004295826,0.015520000364631414,0.06054079942405224,0.0591108573866742,0.028446344104128624,0.027774456325453972,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_8q1s1k,1,8,128,8,1024.0,0.061199999228119854,0.061199999228119854,0.018240000121295452,0.05626560002565384,0.0588383998721838,0.029916029687899703,0.03128396954022015,-6.938893903907228e-18,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q128_16q1s1k,1,16,128,16,1024.0,0.06104319989681244,0.06104319989681244,0.016207999549806118,0.05626560002565384,0.05755840018391609,0.030174939058162802,0.03086826083864964,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_16q1s2k,1,16,256,16,2048.0,0.07356479987502099,0.07356479987502099,0.015584000386297703,0.0556256003677845,0.06238719932734965,0.0346749350032807,0.038889864871740294,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q256_32q1s2k,1,32,256,32,2048.0,0.09806400015950203,0.09806400015950203,0.015008000191301107,0.0556256003677845,0.06935679838061332,0.04364509673334097,0.05441890342616107,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_32q1s4k,1,32,512,32,4096.0,0.1509471982717514,0.1509471982717514,0.014864000026136637,0.0640383992344141,0.10621120035648347,0.05677791295527661,0.09416928531647478,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q512_64q1s4k,1,64,512,64,4096.0,0.21687040030956264,0.21687040030956264,0.015343999955803156,0.0640383992344141,0.14040640145540234,0.06793047918211377,0.1489399211274489,2.7755575615628914e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q1k_64q1s8k,1,64,1024,64,8192.0,0.3848575979471207,0.3848575979471207,0.015647999942302704,0.08209280073642732,0.22809920012950893,0.10185316839884552,0.2830044295482752,0.0,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q2k_32q1s4k,1,32,2048,32,4096.0,0.2785151988267898,0.2785151988267898,0.01726400014013052,0.12431039959192276,0.10621120035648347,0.1501912864839173,0.1283239123428725,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation +4,q4k_16q1s4k,1,16,4096,16,4096.0,0.456512001156807,0.456512001156807,0.017680000513792038,0.3261695951223373,0.06970879957079888,0.3761264483787333,0.08038555277807365,-5.551115123125783e-17,one_fused_FA3_call_projected_for_Frontier_with_total_conservation diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/linear_op.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/linear_op.csv new file mode 100644 index 0000000..2803f02 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/linear_op.csv @@ -0,0 +1,37 @@ +time_stats.emb.min,time_stats.emb.max,time_stats.emb.mean,time_stats.emb.median,time_stats.emb.std,time_stats.input_layernorm.min,time_stats.input_layernorm.max,time_stats.input_layernorm.mean,time_stats.input_layernorm.median,time_stats.input_layernorm.std,time_stats.attn_pre_proj.min,time_stats.attn_pre_proj.max,time_stats.attn_pre_proj.mean,time_stats.attn_pre_proj.median,time_stats.attn_pre_proj.std,time_stats.attn_rope.min,time_stats.attn_rope.max,time_stats.attn_rope.mean,time_stats.attn_rope.median,time_stats.attn_rope.std,time_stats.attn_post_proj.min,time_stats.attn_post_proj.max,time_stats.attn_post_proj.mean,time_stats.attn_post_proj.median,time_stats.attn_post_proj.std,time_stats.post_attention_layernorm.min,time_stats.post_attention_layernorm.max,time_stats.post_attention_layernorm.mean,time_stats.post_attention_layernorm.median,time_stats.post_attention_layernorm.std,n_head,n_kv_head,n_embd,n_expanded_embd,vocab_size,use_gated_mlp,use_qk_norm,attn_output_gate,num_tokens,num_tensor_parallel_workers,padded_n_embd,padded_n_expanded_embd,model_arch,is_step2_mini,share_expert_dim,share_q_dim,measurement_type,profiling_precision,quant_signature +0.029184000566601753,0.06780800223350525,0.03157280012965202,0.030736000277101994,0.005874173435341216,0.033215999603271484,0.04825599864125252,0.03443359974771738,0.0337119996547699,0.0031825678429048183,1.438431978225708,1.505568027496338,1.446228802204132,1.4429279565811157,0.014128607642643025,0.538752019405365,0.5440319776535034,0.5416463971138,0.5420799851417542,0.0016099306164432847,1.0073280334472656,1.0163840055465698,1.0098415970802308,1.0081279873847961,0.0032980457197329728,0.04022400081157684,0.041728001087903976,0.04091359991580248,0.04081599973142147,0.0004728505416767937,32,4,2048,768,151936,True,True,False,8192,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.01360000018030405,0.06784000247716904,0.017755200061947106,0.0179840000346303,0.00837681421401443,0.01836800016462803,0.03651199862360954,0.019606399815529585,0.018719999119639397,0.0038821958848767424,0.7512000203132629,0.8208960294723511,0.7566704005002975,0.75382399559021,0.014793092586577971,0.28995200991630554,0.29337599873542786,0.29135999977588656,0.29150401055812836,0.000998381071258815,0.5149760246276855,0.5169600248336792,0.5160208016633987,0.5158880054950714,0.0005038652783291977,0.02051199972629547,0.021503999829292297,0.021067200042307378,0.021104000508785248,0.00023542855306902367,32,4,2048,768,151936,True,True,False,4096,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.0080960001796484,0.04281599819660187,0.011092799971811474,0.011039999779313803,0.005489135752949302,0.012223999947309494,0.026335999369621277,0.013187199970707298,0.01247999956831336,0.0030236510562153375,0.3928639888763428,0.45372799038887024,0.3976895987987518,0.39528000354766846,0.012905176716136006,0.1547199934720993,0.15884800255298615,0.15712319910526276,0.1573439985513687,0.001165121945135037,0.26633599400520325,0.268095999956131,0.26719200164079665,0.2671840041875839,0.0004242740199415328,0.013024000450968742,0.013887999579310417,0.013489600038155913,0.013520000036805868,0.0002382817554057738,32,4,2048,768,151936,True,True,False,2048,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.00825599953532219,0.04755200073122978,0.02398160002194345,0.01961600035429001,0.00874683840888523,0.018880000337958336,0.03868800029158592,0.021590400114655496,0.0208320003002882,0.004057015751746236,0.24316799640655518,0.2710399925708771,0.2521967992186546,0.25065599381923676,0.007998418528894075,0.09644799679517746,0.19120000302791595,0.10407840013504029,0.09963199868798256,0.020043722414992166,0.13600000739097595,0.18892799317836761,0.15760480016469955,0.15760000050067902,0.0112223677907762,0.009664000011980534,0.010015999898314476,0.009836799977347255,0.009824000298976898,9.016971952948177e-05,32,4,2048,768,151936,True,True,False,1024,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017376000061631203,0.044704001396894455,0.026318399980664254,0.027312000282108784,0.007565344034680866,0.01836800016462803,0.03014400042593479,0.021276800055056812,0.020655999891459942,0.002632977855097951,0.1438719928264618,0.17132799327373505,0.152497598528862,0.15012799948453903,0.007947150319625347,0.1430719941854477,0.19305600225925446,0.16630879789590836,0.1685439944267273,0.014628024163894684,0.08899199962615967,0.10467199981212616,0.0943599995225668,0.09374399855732918,0.003472669494074113,0.007615999784320593,0.007935999892652035,0.007769599952735007,0.0077760000713169575,7.680004540222077e-05,32,4,2048,768,151936,True,True,False,512,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016992000862956047,0.0544000007212162,0.02573199989274144,0.026016000658273697,0.00803323401720434,0.018432000651955605,0.02348800003528595,0.020648000109940768,0.02062400057911873,0.0013939985047930988,0.10220800340175629,0.1361600011587143,0.11850560046732425,0.11684799939393997,0.010637754898360304,0.16710400581359863,0.21110400557518005,0.19078560024499894,0.19409599900245667,0.013714352646558832,0.06265600025653839,0.0740479975938797,0.06842879951000214,0.0690080001950264,0.0031292240446560557,0.00979200005531311,0.033440001308918,0.01864320016466081,0.017280000261962414,0.006957998188876383,32,4,2048,768,151936,True,True,False,256,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017152000218629837,0.04396799951791763,0.025907999789342284,0.02598400041460991,0.007714554737915267,0.018400000408291817,0.03667199984192848,0.024132800102233887,0.02112000063061714,0.006120348733354326,0.10678400099277496,0.1363839954137802,0.1193264003843069,0.11583999916911125,0.009838647443214228,0.17017599940299988,0.22748799622058868,0.18853759989142418,0.18433599919080734,0.015728078443174653,0.04569600149989128,0.06652799993753433,0.05192639995366335,0.05151999928057194,0.004516605694976449,0.021856000646948814,0.026335999369621277,0.02384479995816946,0.023599999956786633,0.0011301153013314744,32,4,2048,768,151936,True,True,False,128,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.017343999817967415,1.0683200359344482,0.05748240072280168,0.029504000209271908,0.16366824814254827,0.018688000738620758,0.3317759931087494,0.03685439983382821,0.021151999942958355,0.06767422466731164,0.10255999863147736,0.9434880018234253,0.16412640027701855,0.11956800147891045,0.17964606281728834,0.1714559942483902,2.1306240558624268,0.3011296011507511,0.1926399990916252,0.42310127734378766,0.03574400022625923,0.6859520077705383,0.08389280084520578,0.04279999993741512,0.14321647071615612,0.020479999482631683,0.1831360012292862,0.033024000097066165,0.023856000043451786,0.03470987082429836,32,4,2048,768,151936,True,True,False,64,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.016095999628305435,0.05142400041222572,0.026363200135529043,0.02676799986511469,0.008637725852473854,0.01849599927663803,0.03577600046992302,0.022193600237369538,0.020848000422120094,0.004615266181181815,0.10540799796581268,0.15014399588108063,0.12211520001292228,0.11896000057458878,0.012772013396624768,0.17315199971199036,0.21478399634361267,0.1881632000207901,0.1873439997434616,0.011761164657572015,0.03481600061058998,0.058079998940229416,0.042200000025331974,0.03969600051641464,0.006461281521769989,0.02143999934196472,0.038816001266241074,0.02466559996828437,0.023856000043451786,0.0035418033748569927,32,4,2048,768,151936,True,True,False,32,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015904000028967857,0.03753599897027016,0.023127999808639287,0.02527999971061945,0.0054230027890305385,0.018400000408291817,0.03001599945127964,0.020488000102341176,0.020096000283956528,0.002585688394137697,0.10355199873447418,0.1438400000333786,0.11536479964852334,0.11124800145626068,0.011136028294919255,0.16502399742603302,0.2072959989309311,0.18646399974822997,0.19075199961662292,0.013296437761247597,0.03142400085926056,0.05215999856591225,0.03888959977775812,0.03750399872660637,0.0057399302067536314,0.020128000527620316,0.04064000025391579,0.023937600292265417,0.023648000322282314,0.004144197400898248,32,4,2048,768,151936,True,True,False,16,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.015200000256299973,0.037151999771595,0.02144480012357235,0.02195199951529503,0.005576364091933697,0.017952000722289085,0.0226879995316267,0.019934400077909233,0.019952000118792057,0.0011812499470458758,0.10063999891281128,0.13468800485134125,0.11595199964940547,0.1207519993185997,0.011974301621092394,0.16332800686359406,0.20748800039291382,0.18162400051951408,0.1767839938402176,0.01474120743720334,0.03222399950027466,0.043455999344587326,0.03829439990222454,0.03859200142323971,0.0031378131240041122,0.020959999412298203,0.03587200120091438,0.023795200139284135,0.023423999547958374,0.0031339489695198443,32,4,2048,768,151936,True,True,False,8,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +0.014944000169634819,0.04022400081157684,0.02162720002233982,0.022463999688625336,0.006017219317026815,0.018464000895619392,0.026528000831604004,0.021264000236988066,0.020896000787615776,0.001934095017586082,0.10156799852848053,0.1703999936580658,0.12565439902245998,0.1244799979031086,0.01641776722785409,0.1653759926557541,0.23865599930286407,0.1969360001385212,0.19223999977111816,0.02058903050274278,0.03254399821162224,0.06752000004053116,0.0443536002188921,0.041519999504089355,0.00984829071098989,0.020096000283956528,0.040031999349594116,0.026934400014579297,0.02478400059044361,0.00562071702648593,32,4,2048,768,151936,True,True,False,1,1,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.7531200051307678,0.8461440205574036,0.7618160009384155,0.7576479911804199,0.019464233617982506,0.2922559976577759,0.2985599935054779,0.2953856036067009,0.29576000571250916,0.001575901077689139,0.510047972202301,0.5140479803085327,0.5121696025133133,0.5123839974403381,0.0012263137313476844,,,,,,32,4,2048,768,151936,True,True,False,8192,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.39529600739479065,0.47279998660087585,0.40216960161924364,0.39825600385665894,0.0162749796240819,0.15625600516796112,0.16211199760437012,0.15959519892930984,0.15988799929618835,0.0011493418942396922,0.2635200023651123,0.2642880082130432,0.2639120012521744,0.26392000913619995,0.0001903593401384135,,,,,,32,4,2048,768,151936,True,True,False,4096,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2531839907169342,0.32419198751449585,0.26924319565296173,0.26049599051475525,0.01657194262368116,0.10051199793815613,0.2375359982252121,0.12411200068891048,0.10311999917030334,0.039466165428540506,0.15881599485874176,0.19289599359035492,0.16896959990262986,0.1640480011701584,0.009162416086418127,,,,,,32,4,2048,768,151936,True,True,False,2048,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15302400290966034,0.1780800074338913,0.16284480094909667,0.16113600134849548,0.00740355661356144,0.14521600306034088,0.20233599841594696,0.17807039842009545,0.1796799972653389,0.013908448560094403,0.0907519981265068,0.09750399738550186,0.09460479989647866,0.09478399902582169,0.0017965487667361475,,,,,,32,4,2048,768,151936,True,True,False,1024,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11740799993276596,0.17958399653434753,0.13262080028653145,0.12878400087356567,0.014146060532423056,0.17468799650669098,0.21161599457263947,0.1910431995987892,0.18966399878263474,0.01270903647700971,0.05926400050520897,0.07577600330114365,0.06530559975653887,0.06404799968004227,0.00479458462514625,,,,,,32,4,2048,768,151936,True,True,False,512,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11260800063610077,0.17209599912166595,0.13610880002379416,0.13809599727392197,0.017675922458993677,0.18729600310325623,0.26822400093078613,0.20815680101513861,0.2078079953789711,0.01700718593658771,0.04499199986457825,0.06537599861621857,0.0507551996037364,0.04787199944257736,0.005727123232692158,,,,,,32,4,2048,768,151936,True,True,False,256,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11257600039243698,0.23452800512313843,0.13866880126297473,0.13964799791574478,0.026590047697062115,0.1828799992799759,0.24751999974250793,0.20735519900918006,0.20670399814844131,0.01500438131586336,0.03654399886727333,0.05593600124120712,0.0408239996060729,0.03892800025641918,0.004864409450710354,,,,,,32,4,2048,768,151936,True,True,False,128,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11734399944543839,0.18726399540901184,0.13890240006148816,0.13308800011873245,0.019086167340006257,0.16991999745368958,0.2443840056657791,0.19185120090842248,0.19257599860429764,0.018017536159219673,0.03033600002527237,0.04956800118088722,0.038387199863791466,0.03750400058925152,0.00500897018702887,,,,,,32,4,2048,768,151936,True,True,False,64,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11420799791812897,0.18115200102329254,0.13938880078494548,0.1393439993262291,0.018141313962922536,0.1693439930677414,0.221343994140625,0.19187839925289155,0.19438399374485016,0.01430752121272052,0.03017600066959858,0.06019200012087822,0.03866560012102127,0.0364960003644228,0.0074199790031205266,,,,,,32,4,2048,768,151936,True,True,False,32,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11276800185441971,0.16223999857902527,0.1331360016018152,0.13305599987506866,0.01432493740801146,0.17187200486660004,0.23625600337982178,0.19287680014967917,0.1913280040025711,0.01754628831589704,0.029823999851942062,0.07574400305747986,0.03900959976017475,0.036927999928593636,0.009277465083962879,,,,,,32,4,2048,768,151936,True,True,False,16,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1130559965968132,0.19814400374889374,0.13379519879817964,0.12531199678778648,0.020074427302248836,0.16841599345207214,0.21139200031757355,0.1893615983426571,0.19092799723148346,0.012418720676762711,0.029503999277949333,0.07558400183916092,0.040144000016152856,0.03728000074625015,0.010223239196498205,,,,,,32,4,2048,768,151936,True,True,False,8,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.115167997777462,0.2640640139579773,0.13805920109152794,0.13118399679660797,0.03132849683515479,0.17103999853134155,0.2977280020713806,0.19899839907884598,0.1976960003376007,0.028928046763067948,0.0297279991209507,0.05990400165319443,0.03834720011800528,0.0363520011305809,0.0070885330713495905,,,,,,32,4,2048,768,151936,True,True,False,1,2,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.3940800130367279,0.46483200788497925,0.39923040121793746,0.3957759886980057,0.015108903906233569,0.16022400557994843,0.1634880006313324,0.16203359961509706,0.16228799521923065,0.0010220720912414007,0.26073598861694336,0.2627840042114258,0.26138080209493636,0.2613760083913803,0.00041640363494172775,,,,,,32,4,2048,768,151936,True,True,False,8192,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.2536959946155548,0.27161601185798645,0.25960480123758317,0.2577280104160309,0.005142017404245015,0.09849599748849869,0.10281600058078766,0.10057279989123344,0.10063999891281128,0.0008836232237268523,0.13913600146770477,0.17606399953365326,0.15875840038061143,0.15988799929618835,0.010058952112389172,,,,,,32,4,2048,768,151936,True,True,False,4096,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.15142400562763214,0.1780479997396469,0.16139679849147798,0.1602879986166954,0.0070656055731385115,0.14601600170135498,0.23343999683856964,0.17913119941949845,0.179967999458313,0.023065274309176566,0.09388799965381622,0.12108799815177917,0.10317599996924401,0.10073599964380264,0.007364345618470178,,,,,,32,4,2048,768,151936,True,True,False,2048,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12198399752378464,0.18406400084495544,0.13770400024950505,0.13232000172138214,0.01623164885687898,0.1773120015859604,0.20688000321388245,0.1870912007987499,0.18535999953746796,0.00825628058448234,0.05766399949789047,0.06739199906587601,0.06187200043350458,0.061216000467538834,0.003004312467037863,,,,,,32,4,2048,768,151936,True,True,False,1024,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11737599968910217,0.18614399433135986,0.13109439946711063,0.12771200388669968,0.01372168725934724,0.17587199807167053,0.2699519991874695,0.19926720038056372,0.19075199961662292,0.02327478982490538,0.041728001087903976,0.06224000081419945,0.05000480003654957,0.049375999718904495,0.006185850944273961,,,,,,32,4,2048,768,151936,True,True,False,512,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1159679964184761,0.1610880047082901,0.12988320142030715,0.12494400516152382,0.011865375783184065,0.1767680048942566,0.27379199862480164,0.21035519987344742,0.21275199949741364,0.023206964017550125,0.037087999284267426,0.062144000083208084,0.04439679980278015,0.04283200018107891,0.006155226392511492,,,,,,32,4,2048,768,151936,True,True,False,256,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12015999853610992,0.158720001578331,0.13573280088603495,0.1343199983239174,0.010666830836014414,0.1737920045852661,0.22127999365329742,0.20261440128087999,0.20321600139141083,0.011153965849067301,0.03587200120091438,0.04944000020623207,0.038265600241720675,0.037328001111745834,0.0030431580113575636,,,,,,32,4,2048,768,151936,True,True,False,128,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.1212799996137619,0.15139199793338776,0.13183839991688728,0.13014400005340576,0.00851207547022104,0.17257599532604218,0.2250880002975464,0.1872655987739563,0.18193599581718445,0.013638284975248818,0.02969600073993206,0.0525440014898777,0.03840640028938651,0.03444799967110157,0.007598511754254174,,,,,,32,4,2048,768,151936,True,True,False,64,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11270400136709213,0.1624639928340912,0.13508000001311302,0.13180799782276154,0.014243564451606775,0.17052799463272095,0.23715199530124664,0.1953311987221241,0.1966560035943985,0.01500670718978398,0.030239999294281006,0.05270399898290634,0.03807039987295866,0.03742399998009205,0.004906165602021684,,,,,,32,4,2048,768,151936,True,True,False,32,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.12319999933242798,0.17919999361038208,0.14059039913117885,0.14156799763441086,0.016217662982107223,0.1701119989156723,0.21987199783325195,0.18903039917349815,0.1870879977941513,0.014117854507841239,0.030368000268936157,0.06406400352716446,0.04045119984075427,0.03710399940609932,0.00878942205983628,,,,,,32,4,2048,768,151936,True,True,False,16,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11296000331640244,0.16761599481105804,0.13843199908733367,0.13814399391412735,0.014925286935582404,0.1711679995059967,0.21561600267887115,0.1906527981162071,0.1876479983329773,0.011803992622137974,0.0306560005992651,0.09216000139713287,0.041129599791020155,0.036847999319434166,0.013739721468154687,,,,,,32,4,2048,768,151936,True,True,False,8,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none +,,,,,,,,,,0.11929599940776825,0.2008959949016571,0.14535359852015972,0.13971199840307236,0.024769473061576282,0.16803200542926788,0.25123199820518494,0.19573760256171227,0.19366399943828583,0.023196011081705884,0.03049599938094616,0.08089599758386612,0.04739360017701984,0.0453919991850853,0.01286389846889463,,,,,,32,4,2048,768,151936,True,True,False,1,4,2048,768,generic,False,,,CUDA_EVENT,BF16,none diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/manifest.json b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/manifest.json new file mode 100644 index 0000000..3abcf44 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/manifest.json @@ -0,0 +1,53 @@ +{ + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" +} diff --git a/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/moe.csv b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/moe.csv new file mode 100644 index 0000000..e5bf507 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2/moe.csv @@ -0,0 +1,73 @@ +time_stats.moe_gating_linear.min,time_stats.moe_gating_linear.max,time_stats.moe_gating_linear.mean,time_stats.moe_gating_linear.median,time_stats.moe_gating_linear.std,time_stats.moe_gating_routing_topk.min,time_stats.moe_gating_routing_topk.max,time_stats.moe_gating_routing_topk.mean,time_stats.moe_gating_routing_topk.median,time_stats.moe_gating_routing_topk.std,time_stats.moe_shuffling.min,time_stats.moe_shuffling.max,time_stats.moe_shuffling.mean,time_stats.moe_shuffling.median,time_stats.moe_shuffling.std,time_stats.moe_grouped_gemm.min,time_stats.moe_grouped_gemm.max,time_stats.moe_grouped_gemm.mean,time_stats.moe_grouped_gemm.median,time_stats.moe_grouped_gemm.std,num_tokens,num_experts,num_experts_per_device,expert_parallel_size,routing_runtime_path,routing_assignment_policy,routing_weight_policy,routing_uses_router_logits,gating_runtime_context,gating_runtime_context_impl,router_topk,hidden_dim,expert_hidden_dim,use_gated,num_tensor_parallel_workers,total_routed_tokens,model_expansion_ratio,tokens_per_expert_avg,tokens_to_experts_ratio,expert_utilization,min_load_ratio,load_imbalance_cv,max_load_ratio,load_entropy,load_gini_coefficient,load_distribution,seed,moe_grouped_gemm_backend,measurement_type,profiling_precision,model_arch,quant_signature,router_median_nonadditivity_ratio,projection_policy +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.33926400542259216,0.405023992061615,0.36780479848384856,0.36507199704647064,0.01690507644474779,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,1.1943039894104004,1.286784052848816,1.228384006023407,1.2273280024528503,0.02832547242381263,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.3984375,0.0,1.346291201783626,4.0,5.59375,0.661865234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,1.679744005203247,1.766144037246704,1.7095808148384095,1.7015680074691772,0.02438921262998535,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.625,0.0,1.015504800579495,5.0,6.15516433212955,0.529052734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,2.1576640605926514,2.2921600341796875,2.2097824096679686,2.188944101333618,0.045572321842012986,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.875,0.0,0.6343057228182637,2.5,6.64370748444639,0.35369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,2.422368049621582,2.5130879878997803,2.4516672134399413,2.434159994125366,0.03278900287381846,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.984375,0.0,0.4921254921257382,2.25,6.817190042344769,0.272369384765625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,2.2217600345611572,2.289599895477295,2.2571327924728393,2.263375997543335,0.021660416089449488,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,1.0,0.125,0.3486861500690843,1.875,6.908192310183997,0.197662353515625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,2.18668794631958,2.318079948425293,2.2218016147613526,2.211087942123413,0.035380897213135316,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,1.0,0.4375,0.2525504668006971,1.875,6.953347743053017,0.1410369873046875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,2.2291839122772217,2.3929600715637207,2.2908096313476562,2.2804640531539917,0.04479348924786221,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,1.0,0.65625,0.15765965680164504,1.5625,6.98229848728205,0.08779525756835938,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,2.0678720474243164,2.1297600269317627,2.0837119817733765,2.0779199600219727,0.017880044357986735,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,1.0,0.625,0.12169081635504074,1.3125,6.9892029662356325,0.06879425048828125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,2.916032075881958,3.0819520950317383,2.9805248022079467,2.9656319618225098,0.05482195799019572,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,1.0,0.796875,0.07935434147688751,1.1796875,6.9954297964750305,0.044734954833984375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,4.421599864959717,4.535359859466553,4.486294317245483,4.497056007385254,0.036990243787549344,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,1.0,0.8203125,0.060849326483103046,1.17578125,6.9973188375859685,0.033740997314453125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,7.302591800689697,7.402751922607422,7.354758310317993,7.3464319705963135,0.032142662400335566,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,1.0,0.890625,0.0412323087266341,1.08984375,6.998772433185578,0.02334284782409668,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.35280001163482666,0.39692801237106323,0.37662720382213594,0.37196800112724304,0.013401318050665304,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.4692479968070984,0.5523840188980103,0.5134752035140991,0.5100640058517456,0.02291433464135784,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.34652799367904663,0.4119040071964264,0.3789471983909607,0.38550400733947754,0.02073945105803335,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.31462401151657104,0.7456960082054138,0.38617280423641204,0.34545600414276123,0.12230201266253077,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.32521599531173706,0.419871985912323,0.36325119733810424,0.34968000650405884,0.03161798848223672,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.289792001247406,0.4663360118865967,0.4091839998960495,0.41655999422073364,0.0446001986506615,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.3761279881000519,0.4416320025920868,0.40686399936676027,0.40540799498558044,0.02257778769899645,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.7172480225563049,0.8663039803504944,0.7723807990550995,0.7591840028762817,0.04164772379451242,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.0195519924163818,1.2216639518737793,1.1253888130187988,1.1453600525856018,0.06548005322243594,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7490559816360474,1.9644800424575806,1.8529024004936219,1.814303994178772,0.08042565617327288,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,3.2479360103607178,3.385279893875122,3.296070408821106,3.2800960540771484,0.04525529026442046,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,6.344799995422363,6.517856121063232,6.464438438415527,6.478623867034912,0.05116674443145098,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,1,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.2648000121116638,0.325439989566803,0.28852800130844114,0.28390398621559143,0.01933778377635077,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.7347840070724487,0.862496018409729,0.7769344031810761,0.769216001033783,0.03290485328796285,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.421875,0.0,1.346291201783626,6.0,5.652114648336087,0.636962890625,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.9198399782180786,0.9646080136299133,0.9412063956260681,0.9411839842796326,0.014939365085478117,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.5703125,0.0,1.118033988749895,5.0,6.008641773518898,0.580810546875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,1.2796800136566162,1.3484159708023071,1.3006976008415223,1.2929120063781738,0.020807998177176254,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.8828125,0.0,0.6959705453537527,3.0,6.60872850615583,0.38055419921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,1.3630399703979492,1.4430400133132935,1.3909215927124023,1.3892319798469543,0.022335744492366926,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.984375,0.0,0.5201036555341637,3.0,6.798826509158851,0.28302001953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,1.27948796749115,1.3904000520706177,1.3176063895225525,1.309440016746521,0.038060887827312775,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,1.0,0.25,0.3511282039725661,1.875,6.91002266305238,0.1970977783203125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,1.264415979385376,1.3145920038223267,1.2791999936103822,1.2753440141677856,0.014130605249568332,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,1.0,0.375,0.24692938483248605,1.6875,6.955481130775285,0.13909912109375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,1.3081920146942139,1.347648024559021,1.3292255997657776,1.329967975616455,0.014558863679016933,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,1.0,0.625,0.17143053326165383,1.5625,6.9786675275754035,0.09520339965820312,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,1.242751955986023,1.3112000226974487,1.2747935891151427,1.266207993030548,0.021093073517695057,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,1.0,0.78125,0.11000099875256815,1.296875,6.991308871213679,0.062183380126953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.7388160228729248,1.8077759742736816,1.772764801979065,1.772704005241394,0.021056644077284283,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,1.0,0.78125,0.0864630150197678,1.1796875,6.994552526394139,0.048796653747558594,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,2.6563520431518555,2.7063679695129395,2.6785055875778196,2.6791679859161377,0.01639963463052944,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,1.0,0.8671875,0.06127686514721937,1.16015625,6.997291583027146,0.03497934341430664,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,4.386879920959473,4.452256202697754,4.4108480453491214,4.406303882598877,0.019768161791937636,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,1.0,0.884765625,0.041723768525324195,1.1171875,6.998746434318934,0.02298593521118164,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.24208000302314758,0.4028480052947998,0.3041536003351212,0.277103990316391,0.05660721484881584,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.2447039932012558,0.30502399802207947,0.26446720361709597,0.26265600323677063,0.016744548364435372,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.2337920069694519,0.2881599962711334,0.26074880361557007,0.264384001493454,0.016469850143940968,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.23369599878787994,0.28591999411582947,0.25465920120477675,0.25385600328445435,0.01593204652619194,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.2295999974012375,0.26556798815727234,0.24674240052700042,0.2497600018978119,0.010345732066199003,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.21721599996089935,0.29020801186561584,0.2394208014011383,0.2346400022506714,0.018747330357768585,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2717759907245636,0.305184006690979,0.28813759982585907,0.28809599578380585,0.01183422600545854,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.3917759954929352,0.43772798776626587,0.41130879521369934,0.4131519943475723,0.012992473640805227,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.6176639795303345,0.7009919881820679,0.642767995595932,0.6330719888210297,0.024074084919938756,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.0820800065994263,1.1674879789352417,1.1034304022789,1.0977439880371094,0.0233067292981566,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.9809919595718384,2.0415360927581787,2.003715181350708,1.992751955986023,0.022066076434645737,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,3.790112018585205,3.8651199340820312,3.829139161109924,3.8230879306793213,0.025177464160110564,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,2,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.212351992726326,0.24383999407291412,0.22760000079870224,0.22723200172185898,0.01050568575837594,1,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.47494399547576904,0.5184000134468079,0.4920704007148743,0.49169600009918213,0.011991064701471855,8,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.3984375,0.0,1.3919410907075054,6.0,5.570159765557392,0.667236328125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.6360960006713867,0.7004479765892029,0.6608384013175964,0.6572319865226746,0.020416877242438597,16,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.625,0.0,1.0307764064044151,4.0,6.138251855282827,0.5382080078125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.780896008014679,0.8301439881324768,0.80346559882164,0.8030399978160858,0.016801230312128875,32,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.859375,0.0,0.6903350635742038,3.0,6.5943747091218174,0.38067626953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.8607040047645569,0.9195200204849243,0.8783008038997651,0.8751039803028107,0.01719059253115595,64,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.9765625,0.0,0.49410588440130926,2.75,6.814452474347134,0.271270751953125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.833952009677887,0.894752025604248,0.8619967997074127,0.863215982913971,0.018716378851797198,128,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,1.0,0.25,0.33693529145074724,2.125,6.9186075263155535,0.1867218017578125,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.834879994392395,0.8871039748191833,0.8651552021503448,0.8638879954814911,0.015262430894400969,256,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,1.0,0.375,0.25567294018677456,1.8125,6.952441049154937,0.14349365234375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.8518080115318298,0.9097599983215332,0.8810272097587586,0.8751040101051331,0.017005819633271906,512,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,1.0,0.625,0.1747801353218523,1.53125,6.978069554482723,0.09820938110351562,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.8470079898834229,0.9010239839553833,0.8694015920162201,0.8716959953308105,0.016138881171190216,1024,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,1.0,0.65625,0.1158122428154187,1.3125,6.9901908183358845,0.06445503234863281,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,1.1698240041732788,1.2311359643936157,1.1888479948043824,1.1890720129013062,0.017978797794492758,2048,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,1.0,0.78125,0.08347181893108634,1.1796875,6.994921772573154,0.046871185302734375,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.7702720165252686,1.8097599744796753,1.7919103980064393,1.7956640124320984,0.012641295676021557,4096,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,1.0,0.78125,0.06866734477822484,1.20703125,6.996602562938728,0.03801727294921875,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.968672037124634,3.0278079509735107,2.9899007797241213,2.9824799299240112,0.01816414122631667,8192,128,128,1,standard_fused_topk,logit_topk,softmax_renorm,True,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,1.0,0.8984375,0.04399546833977376,1.126953125,6.998607314922362,0.024699926376342773,uniform_random_logits,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02502400055527687,0.06092799827456474,0.033839999698102474,0.028672000393271446,0.010315255343709818,0.019360000267624855,0.0352960005402565,0.023424000293016434,0.022064000368118286,0.0038898102289194572,0.0,0.0,0.0,0.0,0.0,0.19574399292469025,0.2512960135936737,0.21939200013875962,0.2199999988079071,0.017532156418212565,1,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8,0.375,0.0625,0.0625,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0233364439829928,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025280000641942024,0.05132799968123436,0.030641599837690593,0.027343999594449997,0.007562409232763304,0.020160000771284103,0.052960000932216644,0.024145600199699403,0.021424000151455402,0.007450198645599985,0.0,0.0,0.0,0.0,0.0,0.20483200252056122,0.24006399512290955,0.22215040028095245,0.22433599829673767,0.00969639786132892,8,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,64,0.375,0.5,0.5,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9806430689981738,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024831999093294144,0.046560000628232956,0.030371200107038022,0.02763199992477894,0.005878487205028602,0.020320000126957893,0.04560000076889992,0.02601920012384653,0.02270400058478117,0.00751129965906799,0.0,0.0,0.0,0.0,0.0,0.20559999346733093,0.24726399779319763,0.22126719802618028,0.22207999974489212,0.01328093478485499,16,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,128,0.375,1.0,1.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9103623678483975,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024288000538945198,0.049375999718904495,0.03086080001667142,0.0267359996214509,0.0070864531384997225,0.020479999482631683,0.030912000685930252,0.02274719988927245,0.021359999664127827,0.0029966813650672505,0.0,0.0,0.0,0.0,0.0,0.20003199577331543,0.2301120012998581,0.21453119963407516,0.21598400175571442,0.010402239855151332,32,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,256,0.375,2.0,2.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9610778571819444,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02393599972128868,0.04342399910092354,0.029380799923092126,0.026688000187277794,0.0057374782000526574,0.020031999796628952,0.036607999354600906,0.022487999964505435,0.020911999978125095,0.0038135203062103235,0.0,0.0,0.0,0.0,0.0,0.19551999866962433,0.22972799837589264,0.21238719969987868,0.21488000452518463,0.010706095835489097,64,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,512,0.375,4.0,4.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9952941013961014,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.025407999753952026,0.05510399863123894,0.031430399790406224,0.02798399981111288,0.007335050273982725,0.020479999482631683,0.03561599925160408,0.02275839988142252,0.021551999263465405,0.003545718365165811,0.0,0.0,0.0,0.0,0.0,0.19420799612998962,0.2903999984264374,0.2211231991648674,0.21476799994707108,0.025955008886196004,128,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,1024,0.375,8.0,8.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9273256282883522,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.0244159996509552,0.05395200103521347,0.03188959984108806,0.026031999848783016,0.00943995927387483,0.02051199972629547,0.036607999354600906,0.023247999791055917,0.02147199958562851,0.003910623837069648,0.0,0.0,0.0,0.0,0.0,0.2290560007095337,0.289247989654541,0.2543327987194061,0.24939200282096863,0.01663236753503115,256,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,2048,0.375,16.0,16.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,1.0380599882396606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.024512000381946564,0.04620800167322159,0.02884640023112297,0.026320000179111958,0.005516557583355925,0.02054399996995926,0.03846399858593941,0.023401600029319524,0.021263999864459038,0.0044742944198265374,0.0,0.0,0.0,0.0,0.0,0.30831998586654663,0.36953601241111755,0.3324000000953674,0.3288639932870865,0.018617999572156707,512,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,4096,0.375,32.0,32.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9569603278386984,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02412799932062626,0.06940799951553345,0.030817600432783365,0.026800000108778477,0.010047085187652939,0.020128000527620316,0.044096000492572784,0.024606400076299904,0.022304000332951546,0.00622857166823714,0.0,0.0,0.0,0.0,0.0,0.462911993265152,0.5497919917106628,0.4893856018781662,0.4816960096359253,0.023348152887178286,1024,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,8192,0.375,64.0,64.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.9524274993623046,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.02831999957561493,0.043487999588251114,0.031744000129401685,0.02991999965161085,0.003973415836428909,0.02070399932563305,0.029343999922275543,0.022886400017887353,0.021743999794125557,0.0026873065045088873,0.0,0.0,0.0,0.0,0.0,0.7662079930305481,0.8717759847640991,0.788454395532608,0.7744799852371216,0.03174746482604812,2048,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,16384,0.375,128.0,128.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8971818172100244,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.03830400109291077,0.06268800050020218,0.043337599746882914,0.040511999279260635,0.005823016247946116,0.023135999217629433,0.03747199848294258,0.025206399988383053,0.02393599972128868,0.003271381256231161,0.0,0.0,0.0,0.0,0.0,1.363935947418213,1.4143040180206299,1.3812703967094422,1.3798720240592957,0.014770450075530007,4096,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,32768,0.375,256.0,256.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.8113207890716184,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize +0.05660799890756607,0.07932800054550171,0.06249920018017292,0.06039999984204769,0.005636461691480845,0.02956799976527691,0.03747199848294258,0.031126399897038935,0.030287999659776688,0.002157571006631541,0.0,0.0,0.0,0.0,0.0,2.5507519245147705,2.680704116821289,2.579859209060669,2.566223978996277,0.03673283558365955,8192,128,128,1,standard_fused_topk,fixed_hotset8,softmax_renorm,False,standalone_legacy,vllm020_replicated_linear,8,2048,768,True,4,65536,0.375,512.0,512.0,0.0625,0.0,3.872983346207417,16.0,3.0,0.9375,hotset8,20260716,FlashInfer CUTLASS,CUDA_EVENT,BF16,generic,none,0.883909666885606,measured_gate+topk+modular_expert;shuffling_zero_because_expert_measurement_includes_prepare_finalize diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp2.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp2.toml new file mode 100644 index 0000000..9debff3 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp2.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1020 bash run_allreduce_profile.sh" +artifacts = ["artifacts/allreduce-full-tp2-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +NUM_TOKENS = "1 8 16 32 64 128 256 512 1024 2048 4096 8192" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/allreduce-full-tp2-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp4.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp4.toml new file mode 100644 index 0000000..28194da --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_full_tp4.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1020 bash run_allreduce_profile.sh" +artifacts = ["artifacts/allreduce-full-tp4-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +NUM_TOKENS = "1 8 16 32 64 128 256 512 1024 2048 4096 8192" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/allreduce-full-tp4-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_smoke_tp2.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_smoke_tp2.toml new file mode 100644 index 0000000..72721af --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_smoke_tp2.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-flashinfer-allreduce-smoke-tp2-20260716-v3-model-context" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 840 bash run_allreduce_profile.sh" +artifacts = ["artifacts/allreduce-smoke-tp2-v3"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +NUM_TOKENS = "8" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/allreduce-smoke-tp2-v3" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_flashattn_full.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_flashattn_full.toml new file mode 100644 index 0000000..5af0e6c --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_flashattn_full.toml @@ -0,0 +1,55 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh" +artifacts = ["artifacts/flashattn-kv-full-v2-tp1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-kv-full-v2-tp1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh" +artifacts = ["artifacts/flashattn-kv-full-v2-tp2"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-kv-full-v2-tp2" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh" +artifacts = ["artifacts/flashattn-kv-full-v2-tp4"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-kv-full-v2-tp4" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_full.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_full.toml new file mode 100644 index 0000000..c24a55c --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_full.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1620 bash run_frontier_linear_full.sh" +artifacts = ["artifacts/frontier-linear-full-v2"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/frontier-linear-full-v2" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +FRONTIER_ROOT = "/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_smoke.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_smoke.toml new file mode 100644 index 0000000..2208aa7 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_frontier_linear_smoke.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-frontier-linear-smoke-20260716-v4-rmsnorm-api" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1020 bash run_frontier_linear_smoke.sh" +artifacts = ["artifacts/frontier-linear-smoke-v4"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/frontier-linear-smoke-v4" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +FRONTIER_ROOT = "/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_full.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_full.toml new file mode 100644 index 0000000..aaeae66 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_full.toml @@ -0,0 +1,18 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-moe-full-20260716-v1-local-shard" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1620 bash run_moe_full.sh" +artifacts = ["artifacts/moe-full-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/moe-full-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_smoke.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_smoke.toml new file mode 100644 index 0000000..f8403b3 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_moe_smoke.toml @@ -0,0 +1,18 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-moe-smoke-20260716-v2-local-shard" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1020 bash run_moe_smoke.sh" +artifacts = ["artifacts/moe-smoke-v2"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/moe-smoke-v2" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_router_full.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_router_full.toml new file mode 100644 index 0000000..4dd0908 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_router_full.toml @@ -0,0 +1,18 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-router-full-20260716-v3-tp-context" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1020 bash run_router_full.sh" +artifacts = ["artifacts/router-full-v3"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/router-full-v3" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_smoke.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_smoke.toml new file mode 100644 index 0000000..650fc76 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_smoke.toml @@ -0,0 +1,18 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-flashattn-kv-smoke-20260716-v3-cu129" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 900 bash run_flashattn_smoke.sh" +artifacts = ["artifacts/flashattn-kv-smoke-v3"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-kv-smoke-v3" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/jobs_trace_routing.toml b/runs/frontier-qwen30-vllm020-profile-v1/jobs_trace_routing.toml new file mode 100644 index 0000000..5b7d00e --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/jobs_trace_routing.toml @@ -0,0 +1,19 @@ +version = 1 + +[[jobs]] +name = "qwen30-vllm020-trace-routing-20260716-v1" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 3720 bash run_trace_routing.sh" +artifacts = ["artifacts/trace-routing-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/trace-routing-v1" +FIXTURE = "/tmp/wjh/qwen30-routing-fixture-20260716.jsonl" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build" +MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/mock-profile-ablation.png b/runs/frontier-qwen30-vllm020-profile-v1/mock-profile-ablation.png new file mode 100644 index 0000000..0457cd2 Binary files /dev/null and b/runs/frontier-qwen30-vllm020-profile-v1/mock-profile-ablation.png differ diff --git a/runs/frontier-qwen30-vllm020-profile-v1/plot_mock_ablation.py b/runs/frontier-qwen30-vllm020-profile-v1/plot_mock_ablation.py new file mode 100644 index 0000000..c61deac --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/plot_mock_ablation.py @@ -0,0 +1,41 @@ +#!/usr/bin/env python3 +"""Render the preregistered profile-ablation figure with schematic data.""" + +from pathlib import Path + +import matplotlib.pyplot as plt + + +CONFIGS = [f"TP{tp}\nMNS{mns}" for tp in (1, 2, 4) for mns in (8, 16, 32, 64)] + +# Schematic only. These values are deliberately not derived from experiment data. +MOCK = { + "Real (mock)": [0.42, 0.50, 0.58, 0.57, 0.60, 0.76, 1.00, 0.99, 0.55, 0.70, 0.74, 0.73], + "Old profile-only (mock)": [0.38, 0.44, 0.51, 0.50, 0.47, 0.58, 0.70, 0.69, 0.66, 0.84, 0.95, 0.96], + "New P-020 profile-only (H1 mock)": [0.40, 0.49, 0.57, 0.56, 0.58, 0.74, 0.98, 0.97, 0.53, 0.69, 0.75, 0.74], + "Per-TP calibrated upper bound (mock)": [0.41, 0.50, 0.58, 0.57, 0.59, 0.75, 0.99, 0.98, 0.54, 0.69, 0.75, 0.74], +} + + +def main() -> None: + output = Path(__file__).with_name("mock-profile-ablation.png") + fig, ax = plt.subplots(figsize=(13.5, 5.8), constrained_layout=True) + x = list(range(len(CONFIGS))) + styles = ["o-", "s--", "^-", "D:"] + for (label, values), style in zip(MOCK.items(), styles, strict=True): + ax.plot(x, values, style, linewidth=2, markersize=5, label=label) + ax.axvline(3.5, color="0.75", linewidth=1) + ax.axvline(7.5, color="0.75", linewidth=1) + ax.set_xticks(x, CONFIGS) + ax.set_ylabel("Normalized SLO-feasible throughput (schematic)") + ax.set_xlabel("Configuration") + ax.set_ylim(0.3, 1.08) + ax.grid(axis="y", alpha=0.25) + ax.legend(ncol=2, frameon=False, loc="upper left") + ax.set_title("MOCK / SCHEMATIC — expected discriminative result, not experiment data") + fig.savefig(output, dpi=180) + print(output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/plot_profile_ablation.py b/runs/frontier-qwen30-vllm020-profile-v1/plot_profile_ablation.py new file mode 100644 index 0000000..f2a831a --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/plot_profile_ablation.py @@ -0,0 +1,142 @@ +#!/usr/bin/env python3 +"""Render the profile ablation and execution-context diagnostics.""" + +from __future__ import annotations + +import argparse +import json +from collections import Counter +from pathlib import Path + +import matplotlib + +matplotlib.use("Agg") +import matplotlib.pyplot as plt +import numpy as np + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--s2", type=Path, required=True) + parser.add_argument("--routing", type=Path, required=True) + parser.add_argument("--opprof", type=Path, required=True) + parser.add_argument("--p1", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + s2 = json.loads(args.s2.read_text()) + routing = json.loads(args.routing.read_text()) + opprof = json.loads(args.opprof.read_text()) + p1 = json.loads(args.p1.read_text()) + + plt.rcParams.update({"font.size": 9, "axes.titlesize": 10, "axes.labelsize": 9}) + fig, axes = plt.subplots(2, 2, figsize=(13.2, 8.2), constrained_layout=True) + + real = s2["real_scores"] + cells = sorted(real, key=lambda cell: (-real[cell], cell)) + x = np.arange(len(cells)) + series = ( + ("Real", real, "#111111", "o"), + ( + "Old profile-only", + s2["historical_modes"]["historical-profile-only"]["simulated_scores"], + "#d95f02", + "s", + ), + ( + "vLLM 0.20 profile-only", + s2["vllm020_profile_only"]["SLO-gated"]["simulated_scores"], + "#7570b3", + "^", + ), + ( + "Frozen per-TP calibration", + s2["historical_modes"]["historical-per-tp-calibration"]["simulated_scores"], + "#1b9e77", + "D", + ), + ) + ax = axes[0, 0] + for label, values, color, marker in series: + ax.plot(x, [values[cell] for cell in cells], label=label, color=color, marker=marker, lw=1.7) + ax.set_xticks(x, [cell.replace("_", "\n") for cell in cells]) + ax.set_ylabel("SLO-feasible req/s/GPU") + ax.set_title("(a) Full 92-probe config ranking") + ax.grid(axis="y", alpha=0.25) + ax.legend(fontsize=8, ncol=2, loc="upper right") + + ax = axes[0, 1] + categories = ("Actual prefill", "Actual decode", "Frontier prior") + cv = [ + routing["phase_summary"]["prefill"]["actual"]["load_cv"]["median"], + routing["phase_summary"]["decode"]["actual"]["load_cv"]["median"], + routing["phase_summary"]["prefill"]["frontier_simulation"]["load_cv"]["median"], + ] + max_ratio = [ + routing["phase_summary"]["prefill"]["actual"]["max_load_ratio"]["median"], + routing["phase_summary"]["decode"]["actual"]["max_load_ratio"]["median"], + routing["phase_summary"]["prefill"]["frontier_simulation"]["max_load_ratio"]["median"], + ] + bx = np.arange(len(categories)) + width = 0.34 + bars1 = ax.bar(bx - width / 2, cv, width, label="Load CV", color="#66c2a5") + bars2 = ax.bar(bx + width / 2, max_ratio, width, label="Max/mean load", color="#fc8d62") + ax.bar_label(bars1, fmt="%.2f", fontsize=8) + ax.bar_label(bars2, fmt="%.2f", fontsize=8) + ax.set_xticks(bx, categories) + ax.set_ylim(0, max(max_ratio) * 1.2) + ax.set_title("(b) Per-layer MoE routing skew") + ax.legend(fontsize=8) + ax.grid(axis="y", alpha=0.25) + + graph_counts = {phase: Counter() for phase in ("pure_decode", "pure_prefill", "true_mixed")} + for cell in opprof["cells"]: + for group in cell["groups"]: + graph_counts[group["phase"]][group["cudagraph_runtime_mode"]] += int(group["steps"]) + ax = axes[1, 0] + phases = tuple(graph_counts) + bottoms = np.zeros(len(phases)) + for mode, color in (("FULL", "#1b9e77"), ("PIECEWISE", "#e6ab02"), ("NONE", "#d95f02")): + values = np.asarray( + [100 * graph_counts[phase][mode] / sum(graph_counts[phase].values()) for phase in phases] + ) + ax.bar(np.arange(len(phases)), values, bottom=bottoms, label=mode, color=color) + bottoms += values + ax.set_xticks(np.arange(len(phases)), [phase.replace("_", " ") for phase in phases]) + ax.set_ylabel("Observed scheduler steps (%)") + ax.set_ylim(0, 100) + ax.set_title("(c) Real vLLM execution mode is phase-dependent") + ax.legend(fontsize=8, ncol=3, loc="lower left") + + ax = axes[1, 1] + mode_order = ("historical-calibrated", "historical-profile-only", "vllm020-profile-only") + labels = ("Per-TP\ncalibration", "Old\nprofile-only", "vLLM 0.20\nprofile-only") + accuracy = [100 * p1["summaries"][mode]["probe_classification"]["accuracy"] for mode in mode_order] + mae = [100 * p1["summaries"][mode]["pass_rate_mae"] for mode in mode_order] + px = np.arange(len(labels)) + bars1 = ax.bar(px - width / 2, accuracy, width, label="Label accuracy (higher better)", color="#1b9e77") + bars2 = ax.bar(px + width / 2, mae, width, label="Pass-rate MAE (lower better)", color="#d95f02") + ax.bar_label(bars1, fmt="%.1f", fontsize=8) + ax.bar_label(bars2, fmt="%.1f", fontsize=8) + ax.set_xticks(px, labels) + ax.set_ylabel("Percent") + ax.set_ylim(0, 105) + ax.set_title("(d) Held-out P1 boundary probes (12 labels)") + ax.legend(fontsize=8, loc="upper right") + ax.grid(axis="y", alpha=0.25) + + fig.suptitle( + "Qwen3-30B-A3B / vLLM 0.20 / BF16 / dash0 H20: operator provenance is not execution-context fidelity", + fontsize=12, + ) + args.output.parent.mkdir(parents=True, exist_ok=True) + fig.savefig(args.output, dpi=180) + fig.savefig(args.output.with_suffix(".svg")) + print(args.output) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/prepare_profile_comparison.py b/runs/frontier-qwen30-vllm020-profile-v1/prepare_profile_comparison.py new file mode 100644 index 0000000..d154272 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/prepare_profile_comparison.py @@ -0,0 +1,104 @@ +#!/usr/bin/env python3 +"""Prepare old/new profile-only Frontier manifests from the frozen P1 probes.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +from pathlib import Path +from typing import Any + + +PROFILE_KEYS = { + "linear_op_input_file": "linear_op.csv", + "atten_input_file": "attention.csv", + "moe_input_file": "moe.csv", +} + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--source", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--mode", choices=("old-profile-only", "new-profile-only"), required=True) + parser.add_argument("--profile-root", type=Path) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + source = json.loads(args.source.read_text()) + if source.get("status") != "PASS" or len(source.get("entries", [])) != 12: + raise SystemExit("source manifest must contain 12 passing P1 probes") + if args.mode == "new-profile-only" and args.profile_root is None: + raise SystemExit("--profile-root is required for new-profile-only") + + output = args.output.resolve() + config_root = output / "configs" + cache_root = output / "prediction-cache" + entries: list[dict[str, Any]] = [] + profile_hashes: dict[str, str] = {} + if args.profile_root is not None: + profile_root = args.profile_root.resolve() + for filename in PROFILE_KEYS.values(): + path = profile_root / filename + if not path.is_file(): + raise SystemExit(f"missing frozen profile: {path}") + profile_hashes[str(path)] = sha256(path) + + for entry in source["entries"]: + config_path = Path(entry["config"]) + config = json.loads(config_path.read_text()) + config["mode"] = args.mode + config["config_id"] = f"{config['cell_id']}__{args.mode}" + config["calibration"]["a_tp"] = 1.0 + knobs = config["frontier"]["knobs"] + knobs["cache_dir"] = str(cache_root) + knobs["no_cache"] = False + if args.mode == "new-profile-only": + for key, filename in PROFILE_KEYS.items(): + knobs[key] = str((args.profile_root.resolve() / filename)) + + target_config = config_root / f"{entry['fixture_id']}.json" + write_json(target_config, config) + updated_entry = dict(entry) + updated_entry["config"] = str(target_config) + updated_entry["calibration_scale"] = 1.0 + entries.append(updated_entry) + + prepared = { + "schema": "frontier-qwen30-profile-comparison-prepared.v1", + "status": "PASS", + "mode": args.mode, + "source": { + "manifest": str(args.source.resolve()), + "sha256": sha256(args.source), + }, + "profile_hashes": profile_hashes, + "isolation": { + "calibration_a_tp": 1.0, + "prediction_cache": str(cache_root), + "all_non_profile_knobs_inherited": True, + }, + "entries": entries, + } + write_json(output / "prepared-manifest.json", prepared) + print(output / "prepared-manifest.json") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/prepare_s2_profile_comparison.py b/runs/frontier-qwen30-vllm020-profile-v1/prepare_s2_profile_comparison.py new file mode 100644 index 0000000..962952a --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/prepare_s2_profile_comparison.py @@ -0,0 +1,133 @@ +#!/usr/bin/env python3 +"""Prepare the 92 frozen S2-R-b probes with a replacement profile bundle.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +from pathlib import Path +from typing import Any + + +PROFILE_KEYS = { + "linear_op_input_file": "linear_op.csv", + "atten_input_file": "attention.csv", + "moe_input_file": "moe.csv", +} + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def write_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--resolved-plan", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--profile-root", type=Path, required=True) + parser.add_argument("--cache-root", type=Path) + parser.add_argument("--shards", type=int, default=1) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + if args.shards < 1: + raise SystemExit("--shards must be positive") + source = json.loads(args.resolved_plan.read_text()) + source_rows = [row for row in source["runs"] if row["mode"] == "uncalibrated"] + if len(source_rows) != 92: + raise SystemExit(f"expected 92 uncalibrated probes, found {len(source_rows)}") + + output = args.output.resolve() + profile_root = args.profile_root.resolve() + cache_root = (args.cache_root or (output / "prediction-cache")).resolve() + profile_hashes: dict[str, str] = {} + for filename in PROFILE_KEYS.values(): + path = profile_root / filename + if not path.is_file(): + raise SystemExit(f"missing frozen profile: {path}") + profile_hashes[str(path)] = sha256(path) + + entries: list[dict[str, Any]] = [] + for row in source_rows: + fixture_dir = Path(row["fixture_dir"]).resolve() + fixture_manifest_path = fixture_dir / "fixture_manifest.json" + fixture = json.loads(fixture_manifest_path.read_text()) + config = json.loads(Path(row["config_path"]).read_text()) + config["mode"] = "new-profile-only" + config["config_id"] = f"{row['cell_id']}__new-profile-only" + config["calibration"]["a_tp"] = 1.0 + knobs = config["frontier"]["knobs"] + knobs["cache_dir"] = str(cache_root) + knobs["no_cache"] = False + for key, filename in PROFILE_KEYS.items(): + knobs[key] = str(profile_root / filename) + + target_config = output / "configs" / f"{row['fixture_id']}.json" + write_json(target_config, config) + entries.append( + { + "fixture_id": row["fixture_id"], + "cell": row["cell_id"], + "role": f"probe-{int(row['probe_index']):02d}", + "anchor": float(row["sampling_u"]), + "selected_count": int(row["request_count"]), + "probe_index": int(row["probe_index"]), + "sampling_u": float(row["sampling_u"]), + "tensor_parallel_size": int(row["tensor_parallel_size"]), + "config": str(target_config), + "fixture_manifest": str(fixture_manifest_path), + "frontier_csv": str(fixture_dir / "frontier.csv"), + "sidecar": str(fixture_dir / "sidecar.jsonl"), + "calibration_scale": 1.0, + "request_count": int(fixture["request_count"]), + } + ) + + base = { + "schema": "frontier-qwen30-s2-profile-comparison-prepared.v1", + "status": "PASS", + "mode": "new-profile-only", + "source": { + "resolved_plan": str(args.resolved_plan.resolve()), + "sha256": sha256(args.resolved_plan), + }, + "profile_hashes": profile_hashes, + "isolation": { + "calibration_a_tp": 1.0, + "prediction_cache": str(cache_root), + "all_non_profile_knobs_inherited": True, + }, + "suite_total_runs": len(entries), + } + write_json( + output / "prepared-manifest.json", + {**base, "expected_runs": len(entries), "entries": entries}, + ) + for shard in range(args.shards): + shard_entries = entries[shard :: args.shards] + write_json( + output / f"prepared-manifest-shard-{shard:02d}-of-{args.shards:02d}.json", + { + **base, + "shard": {"index": shard, "count": args.shards}, + "expected_runs": len(shard_entries), + "entries": shard_entries, + }, + ) + print(output / "prepared-manifest.json") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_allreduce.py b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_allreduce.py new file mode 100644 index 0000000..09b3752 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_allreduce.py @@ -0,0 +1,192 @@ +#!/usr/bin/env python3 +"""Profile vLLM 0.20 TP all-reduce and assert FlashInfer TRTLLM dispatch.""" + +from __future__ import annotations + +import argparse +import json +import os +import statistics +import subprocess +from pathlib import Path +from typing import Any + +import torch +import torch.distributed as dist +import vllm + + +VLLM_VERSION = "0.20.0" +VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--vllm-source", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--num-tokens", type=int, nargs="+", default=[8]) + parser.add_argument("--hidden-dim", type=int, default=2048) + parser.add_argument("--warmup-iters", type=int, default=3) + parser.add_argument("--repeats", type=int, default=10) + return parser.parse_args() + + +def git_head(repo: Path) -> str: + return subprocess.check_output( + ["git", "-C", str(repo), "rev-parse", "HEAD"], text=True + ).strip() + + +def stats_ms(samples: list[float]) -> dict[str, float]: + return { + "min": min(samples), + "max": max(samples), + "mean": statistics.fmean(samples), + "median": statistics.median(samples), + "std": statistics.pstdev(samples), + } + + +def main() -> None: + args = parse_args() + if vllm.__version__ != VLLM_VERSION: + raise SystemExit(f"expected vLLM {VLLM_VERSION}, got {vllm.__version__}") + source_head = git_head(args.vllm_source) + if source_head != VLLM_COMMIT: + raise SystemExit(f"expected vLLM source {VLLM_COMMIT}, got {source_head}") + if os.getenv("VLLM_ALLREDUCE_USE_FLASHINFER") != "1": + raise SystemExit("VLLM_ALLREDUCE_USE_FLASHINFER must equal 1") + if os.getenv("VLLM_FLASHINFER_ALLREDUCE_BACKEND") != "trtllm": + raise SystemExit("VLLM_FLASHINFER_ALLREDUCE_BACKEND must equal trtllm") + if "RANK" not in os.environ or "WORLD_SIZE" not in os.environ: + raise SystemExit("launch with torchrun") + + from vllm.distributed import ( + destroy_distributed_environment, + destroy_model_parallel, + init_distributed_environment, + initialize_model_parallel, + tensor_model_parallel_all_reduce, + ) + from vllm.distributed.parallel_state import get_tp_group + from vllm.config import ( + ModelConfig, + ParallelConfig, + VllmConfig, + set_current_vllm_config, + ) + + rank = int(os.environ["RANK"]) + local_rank = int(os.environ["LOCAL_RANK"]) + world_size = int(os.environ["WORLD_SIZE"]) + if world_size not in (2, 4): + raise SystemExit(f"expected TP world size 2 or 4, got {world_size}") + device = torch.device(f"cuda:{local_rank}") + torch.accelerator.set_device_index(device) + torch.set_default_device(device) + init_distributed_environment() + model_config = ModelConfig( + model=str(args.model), + dtype="bfloat16", + max_model_len=8192, + skip_tokenizer_init=True, + generation_config="vllm", + ) + vllm_config = VllmConfig( + model_config=model_config, + parallel_config=ParallelConfig(tensor_parallel_size=world_size) + ) + with set_current_vllm_config(vllm_config): + initialize_model_parallel(tensor_model_parallel_size=world_size) + + rows: list[dict[str, Any]] = [] + expected_sum = world_size * (world_size + 1) / 2 + try: + for num_tokens in args.num_tokens: + input_tensor = torch.full( + (num_tokens, args.hidden_dim), + float(rank + 1), + dtype=torch.bfloat16, + device=device, + ) + for _ in range(args.warmup_iters): + output = tensor_model_parallel_all_reduce(input_tensor) + torch.accelerator.synchronize() + torch.testing.assert_close( + output, + torch.full_like(output, expected_sum), + atol=0.0, + rtol=0.0, + ) + + communicator = get_tp_group().device_communicator + fi_comm = communicator.fi_ar_comm + if fi_comm is None or fi_comm.disabled: + raise SystemExit( + f"FlashInfer all-reduce was not selected at TP={world_size}, " + f"tokens={num_tokens}" + ) + uses_flashinfer = fi_comm.should_use_fi_ar(input_tensor) + + samples: list[float] = [] + for _ in range(args.repeats): + dist.barrier() + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + output = tensor_model_parallel_all_reduce(input_tensor) + end.record() + torch.accelerator.synchronize() + samples.append(float(start.elapsed_time(end))) + + gathered: list[list[float] | None] = [None] * world_size + dist.all_gather_object(gathered, samples) + if rank == 0: + per_rank = [stats_ms(item) for item in gathered if item is not None] + row = { + "tensor_parallel_size": world_size, + "num_tokens": num_tokens, + "hidden_dim": args.hidden_dim, + "payload_bytes": num_tokens + * args.hidden_dim + * torch.tensor([], dtype=torch.bfloat16).element_size(), + "dtype": "bfloat16", + "communicator": "vllm.tensor_model_parallel_all_reduce", + "selected_backend": ( + "flashinfer_trtllm" if uses_flashinfer else "nccl_fallback" + ), + "per_rank_time_ms": per_rank, + "critical_path_median_ms": max( + rank_stats["median"] for rank_stats in per_rank + ), + } + rows.append(row) + print(json.dumps(row, sort_keys=True), flush=True) + finally: + destroy_model_parallel() + destroy_distributed_environment() + + if rank == 0: + payload = { + "schema_version": "qwen30_vllm020_allreduce_raw.v1", + "environment": { + "vllm_version": vllm.__version__, + "vllm_source_commit": source_head, + "torch_version": torch.__version__, + "torch_cuda": torch.version.cuda, + "gpu": torch.cuda.get_device_name(device), + "model": str(args.model), + "backend_env": { + "VLLM_ALLREDUCE_USE_FLASHINFER": "1", + "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm", + }, + }, + "rows": rows, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_flashattn.py b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_flashattn.py new file mode 100644 index 0000000..200762e --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_flashattn.py @@ -0,0 +1,444 @@ +#!/usr/bin/env python3 +"""Profile the exact vLLM 0.20 FlashAttention backend at TP-local shapes. + +This deliberately uses vLLM's own v0.20.0 attention benchmark runner instead +of Frontier's FlashInfer-only attention wrapper. The output is raw evidence; +projection into Frontier's split attention CSV schema is a separate step. +""" + +from __future__ import annotations + +import argparse +import json +import statistics +import subprocess +import sys +import types +from pathlib import Path + +import torch +import vllm + + +VLLM_VERSION = "0.20.0" +VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--vllm-source", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--tp", type=int, choices=(1, 2, 4), nargs="+", default=[1, 2, 4]) + parser.add_argument( + "--batch-specs", + nargs="+", + default=["q128", "4q1s128", "q128_4q1s128"], + ) + parser.add_argument("--warmup-iters", type=int, default=3) + parser.add_argument("--repeats", type=int, default=5) + parser.add_argument("--device", default="cuda:0") + parser.add_argument("--profile-kv-update", action="store_true") + parser.add_argument( + "--profile-method", + choices=("cuda_event", "record_function"), + default="cuda_event", + ) + parser.add_argument("--frontier-source", type=Path) + return parser.parse_args() + + +def git_head(repo: Path) -> str: + return subprocess.check_output( + ["git", "-C", str(repo), "rev-parse", "HEAD"], text=True + ).strip() + + +def json_default(value: object) -> object: + if isinstance(value, (Path, torch.dtype, torch.device)): + return str(value) + item = getattr(value, "item", None) + if callable(item): + return item() + raise TypeError(f"cannot serialize {type(value).__name__}") + + +def main() -> None: + args = parse_args() + if vllm.__version__ != VLLM_VERSION: + raise SystemExit(f"expected vLLM {VLLM_VERSION}, got {vllm.__version__}") + source_head = git_head(args.vllm_source) + if source_head != VLLM_COMMIT: + raise SystemExit(f"expected vLLM source {VLLM_COMMIT}, got {source_head}") + if not args.model.joinpath("config.json").is_file(): + raise SystemExit(f"missing model config: {args.model / 'config.json'}") + if args.profile_method == "record_function" and args.frontier_source is None: + raise SystemExit("--frontier-source is required for --profile-method record_function") + + bench_dir = args.vllm_source / "benchmarks" / "attention_benchmarks" + sys.path.insert(0, str(bench_dir)) + import runner # type: ignore[import-not-found] # noqa: PLC0415 + from batch_spec import parse_batch_spec # type: ignore[import-not-found] # noqa: PLC0415 + from common import ( # type: ignore[import-not-found] # noqa: PLC0415 + BenchmarkConfig, + BenchmarkResult, + ) + from vllm.config import ( # noqa: PLC0415 + CacheConfig, + CompilationConfig, + DeviceConfig, + LoadConfig, + ModelConfig, + ParallelConfig, + SchedulerConfig, + VllmConfig, + set_current_vllm_config, + ) + from vllm.v1.attention.backends.utils import ( # noqa: PLC0415 + get_kv_cache_layout, + set_kv_cache_layout, + ) + from vllm.v1.kv_cache_interface import FullAttentionSpec # noqa: PLC0415 + from vllm.v1.worker.workspace import init_workspace_manager # noqa: PLC0415 + + record_function_tracer = None + if args.profile_method == "record_function": + sys.path.insert(0, str(args.frontier_source.resolve())) + from frontier.profiling.utils.record_function_tracer import RecordFunctionTracer + + record_function_tracer = RecordFunctionTracer + + def create_vllm_config(config: BenchmarkConfig, max_num_blocks: int) -> VllmConfig: + model_config = ModelConfig( + model=str(args.model), + tokenizer=str(args.model), + trust_remote_code=False, + dtype="bfloat16", + seed=0, + max_model_len=40960, + ) + cache_config = CacheConfig(block_size=config.block_size, cache_dtype="auto") + cache_config.num_gpu_blocks = max_num_blocks + cache_config.num_cpu_blocks = 0 + parallel_config = ParallelConfig(tensor_parallel_size=1) + scheduler_config = SchedulerConfig( + max_num_seqs=256, + max_num_batched_tokens=8192, + max_model_len=40960, + is_encoder_decoder=False, + enable_chunked_prefill=True, + ) + model_config.get_num_layers = types.MethodType( + lambda self: config.num_layers, model_config + ) + model_config.get_sliding_window_for_layer = types.MethodType( + lambda self, i: None, model_config + ) + model_config.get_logits_soft_cap_for_layer = types.MethodType( + lambda self, i: 0.0, model_config + ) + model_config.get_sm_scale_for_layer = types.MethodType( + lambda self, i: 1.0 / config.head_dim**0.5, model_config + ) + model_config.get_num_attention_heads = types.MethodType( + lambda self, parallel_config=None: config.num_q_heads, model_config + ) + model_config.get_num_kv_heads = types.MethodType( + lambda self, parallel_config=None: config.num_kv_heads, model_config + ) + model_config.get_head_size = types.MethodType( + lambda self: config.head_dim, model_config + ) + model_config.get_sliding_window = types.MethodType( + lambda self: None, model_config + ) + return VllmConfig( + model_config=model_config, + cache_config=cache_config, + parallel_config=parallel_config, + scheduler_config=scheduler_config, + device_config=DeviceConfig(), + load_config=LoadConfig(), + compilation_config=CompilationConfig(), + ) + + runner._create_vllm_config = create_vllm_config + init_workspace_manager(args.device) + args.output.parent.mkdir(parents=True, exist_ok=True) + if args.profile_method == "record_function": + (args.output.parent / "profiler_traces").mkdir(exist_ok=True) + + def profile_kv_cache_update(config: BenchmarkConfig) -> dict[str, float]: + device = torch.device(config.device) + requests = parse_batch_spec(config.batch_spec) + total_q = sum(request.q_len for request in requests) + max_kv = max(request.kv_len for request in requests) + max_blocks_per_request = (max_kv + config.block_size - 1) // config.block_size + max_num_blocks = len(requests) * max_blocks_per_request + vllm_config = create_vllm_config(config, max_num_blocks) + dtype = vllm_config.model_config.dtype + with set_current_vllm_config(vllm_config): + backend_config = runner._get_backend_config(config.backend) + backend_class, impl, layer = runner._create_backend_impl( + backend_config, config, device, dtype + ) + required_layout = backend_class.get_required_kv_cache_layout() + if required_layout is not None: + set_kv_cache_layout(required_layout) + get_kv_cache_layout.cache_clear() + common_metadata = runner._build_common_attn_metadata( + [request.q_len for request in requests], + [request.kv_len for request in requests], + config.block_size, + device, + ) + kv_cache_spec = FullAttentionSpec( + block_size=config.block_size, + num_kv_heads=config.num_kv_heads, + head_size=config.head_dim, + dtype=dtype, + ) + layer._kv_cache_spec = kv_cache_spec + _, key_list, value_list = runner._create_input_tensors( + config, total_q, device, dtype + ) + cache_list = runner._create_kv_cache( + config, max_num_blocks, backend_class, device, dtype + ) + for _ in range(config.warmup_iters): + for layer_index in range(config.num_layers): + impl.do_kv_cache_update( + layer, + key_list[layer_index], + value_list[layer_index], + cache_list[layer_index], + common_metadata.slot_mapping, + ) + torch.accelerator.synchronize() + samples: list[float] = [] + for _ in range(config.repeats): + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + for layer_index in range(config.num_layers): + impl.do_kv_cache_update( + layer, + key_list[layer_index], + value_list[layer_index], + cache_list[layer_index], + common_metadata.slot_mapping, + ) + end.record() + torch.accelerator.synchronize() + samples.append(float(start.elapsed_time(end)) / config.num_layers) + return { + "min_ms": min(samples), + "max_ms": max(samples), + "mean_ms": statistics.fmean(samples), + "median_ms": statistics.median(samples), + "std_ms": statistics.pstdev(samples), + } + + def profile_kernel_only( + config: BenchmarkConfig, + ) -> tuple[BenchmarkResult, dict[str, float] | None]: + """Trace exactly one vLLM FA3 forward/KV-update per annotation. + + `RecordFunctionTracer` is Frontier's actual KERNEL_ONLY collector: it + follows CUDA launch correlations and sums kernels under the annotation. + The profiling loop therefore contains no CUDA-event value relabeling. + """ + device = torch.device(config.device) + torch.accelerator.set_device_index(device) + backend_config = runner._get_backend_config(config.backend) + requests = parse_batch_spec(config.batch_spec) + q_lens = [request.q_len for request in requests] + kv_lens = [request.kv_len for request in requests] + total_q = sum(q_lens) + max_kv = max(kv_lens) + max_blocks_per_request = (max_kv + config.block_size - 1) // config.block_size + max_num_blocks = len(requests) * max_blocks_per_request + + with runner.log_warnings_and_errors_only(): + vllm_config = create_vllm_config(config, max_num_blocks) + dtype = vllm_config.model_config.dtype + with set_current_vllm_config(vllm_config): + backend_class, impl, layer = runner._create_backend_impl( + backend_config, config, device, dtype + ) + required_layout = backend_class.get_required_kv_cache_layout() + if required_layout is not None: + set_kv_cache_layout(required_layout) + get_kv_cache_layout.cache_clear() + common_metadata = runner._build_common_attn_metadata( + q_lens, kv_lens, config.block_size, device + ) + kv_cache_spec = FullAttentionSpec( + block_size=config.block_size, + num_kv_heads=config.num_kv_heads, + head_size=config.head_dim, + dtype=dtype, + ) + builder = runner._create_metadata_builder( + backend_class, kv_cache_spec, vllm_config, device, config.backend + ) + attn_metadata = builder.build( + common_prefix_len=0, common_attn_metadata=common_metadata + ) + quantize_query = config.kv_cache_dtype.startswith("fp8") and getattr( + impl, "supports_quant_query_input", False + ) + q_list, k_list, v_list = runner._create_input_tensors( + config, total_q, device, dtype, quantize_query=quantize_query + ) + cache_list = runner._create_kv_cache( + config, max_num_blocks, backend_class, device, dtype + ) + output = torch.empty( + total_q, + config.num_q_heads, + config.head_dim, + device=device, + dtype=dtype, + ) + + def run_core() -> None: + for layer_index in range(config.num_layers): + impl.forward( + layer, + q_list[layer_index], + k_list[layer_index], + v_list[layer_index], + cache_list[layer_index], + attn_metadata, + output=output, + ) + + for _ in range(config.warmup_iters): + run_core() + torch.accelerator.synchronize() + core_tracer = record_function_tracer(str(args.output.parent)) + with core_tracer: + for _ in range(config.repeats): + with torch.profiler.record_function("vidur_attention_core"): + run_core() + core_stats = core_tracer.get_operation_time_stats() + if "attention_core" not in core_stats: + raise RuntimeError("missing KERNEL_ONLY FlashAttention core stats") + core = { + name: float(value) / config.num_layers + for name, value in core_stats["attention_core"].items() + } + + kv_stats = None + if args.profile_kv_update: + def run_kv_cache_update() -> None: + for layer_index in range(config.num_layers): + impl.do_kv_cache_update( + layer, + k_list[layer_index], + v_list[layer_index], + cache_list[layer_index], + common_metadata.slot_mapping, + ) + + for _ in range(config.warmup_iters): + run_kv_cache_update() + torch.accelerator.synchronize() + kv_tracer = record_function_tracer(str(args.output.parent)) + with kv_tracer: + for _ in range(config.repeats): + with torch.profiler.record_function("vidur_attn_kv_cache_save"): + run_kv_cache_update() + kv_time_stats = kv_tracer.get_operation_time_stats() + if "attn_kv_cache_save" not in kv_time_stats: + raise RuntimeError("missing KERNEL_ONLY KV-update stats") + kv_stats = { + f"{name}_ms": float(value) / config.num_layers + for name, value in kv_time_stats["attn_kv_cache_save"].items() + } + + result = BenchmarkResult( + config=config, + mean_time=core["mean"] / 1000.0, + std_time=core["std"] / 1000.0, + min_time=core["min"] / 1000.0, + max_time=core["max"] / 1000.0, + throughput_tokens_per_sec=( + total_q / (core["mean"] / 1000.0) if core["mean"] > 0 else 0.0 + ), + ) + return result, kv_stats + + rows: list[dict[str, object]] = [] + for tp in args.tp: + for batch_spec in args.batch_specs: + config = BenchmarkConfig( + backend="FLASH_ATTN", + batch_spec=batch_spec, + num_layers=1, + head_dim=128, + num_q_heads=32 // tp, + num_kv_heads=4 // tp, + block_size=16, + device=args.device, + dtype=torch.bfloat16, + repeats=args.repeats, + warmup_iters=args.warmup_iters, + profile_memory=True, + kv_cache_dtype="auto", + use_cuda_graphs=False, + ) + if args.profile_method == "record_function": + result, kv_stats = profile_kernel_only(config) + else: + result = runner.run_attention_benchmark(config) + kv_stats = ( + profile_kv_cache_update(config) if args.profile_kv_update else None + ) + row = result.to_dict() + row["tensor_parallel_size"] = tp + row["attention_core_excludes_kv_cache_update"] = True + if kv_stats is not None: + row["kv_cache_update_time"] = kv_stats + rows.append(row) + print( + json.dumps( + { + "tp": tp, + "batch_spec": batch_spec, + "mean_time_s": result.mean_time, + "error": result.error, + }, + sort_keys=True, + default=json_default, + ), + flush=True, + ) + if not result.success: + raise SystemExit(f"attention profile failed: {row}") + + payload = { + "schema_version": "qwen30_vllm020_flashattn_raw.v1", + "environment": { + "vllm_version": vllm.__version__, + "vllm_source_commit": source_head, + "torch_version": torch.__version__, + "torch_cuda": torch.version.cuda, + "gpu": torch.cuda.get_device_name(torch.device(args.device)), + "model": str(args.model), + "dtype": "bfloat16", + "attention_backend": "FLASH_ATTN", + "block_size": 16, + "profile_kv_update": args.profile_kv_update, + "profile_method": args.profile_method, + }, + "rows": rows, + } + args.output.write_text( + json.dumps(payload, indent=2, sort_keys=True, default=json_default) + "\n" + ) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_moe.py b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_moe.py new file mode 100644 index 0000000..2eecef2 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_moe.py @@ -0,0 +1,416 @@ +#!/usr/bin/env python3 +"""Profile vLLM 0.20's Qwen3-30B unquantized MoE kernel at TP-local shapes.""" + +from __future__ import annotations + +import argparse +import json +import math +import statistics +import subprocess +from pathlib import Path +from typing import Any, Callable + +import torch +import vllm + + +VLLM_VERSION = "0.20.0" +VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +HIDDEN_DIM = 2048 +INTERMEDIATE_DIM = 768 +NUM_EXPERTS = 128 +TOP_K = 8 + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--vllm-source", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--tp", type=int, choices=(1, 2, 4), nargs="+", default=[1, 2, 4]) + parser.add_argument("--num-tokens", type=int, nargs="+", default=[8]) + parser.add_argument( + "--routing-modes", + choices=("uniform_random_logits", "hotset8"), + nargs="+", + default=["uniform_random_logits"], + ) + parser.add_argument("--warmup-iters", type=int, default=3) + parser.add_argument("--repeats", type=int, default=5) + parser.add_argument("--device", default="cuda:0") + parser.add_argument("--check-reference", action="store_true") + parser.add_argument( + "--profile-method", + choices=("cuda_event", "record_function"), + default="cuda_event", + ) + parser.add_argument("--frontier-source", type=Path) + return parser.parse_args() + + +def git_head(repo: Path) -> str: + return subprocess.check_output( + ["git", "-C", str(repo), "rev-parse", "HEAD"], text=True + ).strip() + + +def stats_ms(samples: list[float]) -> dict[str, float]: + return { + "min": min(samples), + "max": max(samples), + "mean": statistics.fmean(samples), + "median": statistics.median(samples), + "std": statistics.pstdev(samples), + } + + +def measure_kernel_only_ms( + fn: Callable[[], torch.Tensor], + *, + warmup_iters: int, + repeats: int, + trace_root: Path, + operation_name: str, + record_function_tracer: type, +) -> tuple[torch.Tensor, dict[str, float]]: + """Use Frontier's KERNEL_ONLY contract, not a CUDA-event relabel.""" + result = None + for _ in range(warmup_iters): + result = fn() + torch.accelerator.synchronize() + + tracer = record_function_tracer(str(trace_root)) + with tracer: + for _ in range(repeats): + with torch.profiler.record_function(f"vidur_{operation_name}"): + result = fn() + stats = tracer.get_operation_time_stats() + if operation_name not in stats: + raise RuntimeError(f"missing RecordFunctionTracer stats for {operation_name}") + if result is None: + raise RuntimeError("kernel-only profiler executed no MoE step") + return result, {name: float(value) for name, value in stats[operation_name].items()} + + +def routing_inputs( + mode: str, num_tokens: int, device: torch.device +) -> tuple[torch.Tensor, torch.Tensor, dict[str, Any]]: + from vllm.model_executor.layers.fused_moe import fused_topk + + if mode == "uniform_random_logits": + logits = torch.randn( + (num_tokens, NUM_EXPERTS), device=device, dtype=torch.bfloat16 + ) + hidden_for_topk = torch.empty( + (num_tokens, HIDDEN_DIM), device=device, dtype=torch.bfloat16 + ) + weights, ids, _ = fused_topk( + hidden_for_topk, + logits, + TOP_K, + renormalize=True, + ) + elif mode == "hotset8": + ids = torch.arange(TOP_K, device=device, dtype=torch.int32).repeat( + num_tokens, 1 + ) + weights = torch.full( + (num_tokens, TOP_K), + 1.0 / TOP_K, + device=device, + dtype=torch.float32, + ) + else: + raise ValueError(mode) + + counts = torch.bincount(ids.flatten().to(torch.int64), minlength=NUM_EXPERTS) + counts_cpu = counts.cpu().tolist() + mean_load = num_tokens * TOP_K / NUM_EXPERTS + variance = sum((count - mean_load) ** 2 for count in counts_cpu) / NUM_EXPERTS + return weights, ids, { + "active_experts": sum(count > 0 for count in counts_cpu), + "min_tokens_per_expert": min(counts_cpu), + "max_tokens_per_expert": max(counts_cpu), + "load_cv": math.sqrt(variance) / mean_load if mean_load else 0.0, + "counts": counts_cpu, + } + + +def reference_partial_output( + hidden: torch.Tensor, + w13_original: torch.Tensor, + w2: torch.Tensor, + topk_weights: torch.Tensor, + topk_ids: torch.Tensor, +) -> torch.Tensor: + output = torch.zeros_like(hidden) + for token in range(hidden.shape[0]): + for route in range(TOP_K): + expert = int(topk_ids[token, route]) + gate_up = torch.mv(w13_original[expert], hidden[token]) + gate, up = gate_up.chunk(2) + activated = torch.nn.functional.silu(gate) * up + expert_output = torch.mv(w2[expert], activated) + output[token].add_( + expert_output * topk_weights[token, route].to(expert_output.dtype) + ) + return output + + +def main() -> None: + args = parse_args() + if vllm.__version__ != VLLM_VERSION: + raise SystemExit(f"expected vLLM {VLLM_VERSION}, got {vllm.__version__}") + source_head = git_head(args.vllm_source) + if source_head != VLLM_COMMIT: + raise SystemExit(f"expected vLLM source {VLLM_COMMIT}, got {source_head}") + model_config = json.loads(args.model.joinpath("config.json").read_text()) + expected_model = { + "hidden_size": HIDDEN_DIM, + "moe_intermediate_size": INTERMEDIATE_DIM, + "num_experts": NUM_EXPERTS, + "num_experts_per_tok": TOP_K, + "norm_topk_prob": True, + "torch_dtype": "bfloat16", + } + observed_model = {key: model_config.get(key) for key in expected_model} + if observed_model != expected_model: + raise SystemExit( + f"model contract mismatch: expected {expected_model}, got {observed_model}" + ) + if args.profile_method == "record_function" and args.frontier_source is None: + raise SystemExit("--frontier-source is required for --profile-method record_function") + + from vllm.config import ParallelConfig, VllmConfig, set_current_vllm_config + from vllm.model_executor.layers.fused_moe.activation import MoEActivation + from vllm.model_executor.layers.fused_moe.config import ( + FUSED_MOE_UNQUANTIZED_CONFIG, + FusedMoEConfig, + FusedMoEParallelConfig, + RoutingMethodType, + ) + from vllm.model_executor.layers.fused_moe.oracle.unquantized import ( + UnquantizedMoeBackend, + convert_to_unquantized_kernel_format, + make_unquantized_moe_kernel, + select_unquantized_moe_backend, + ) + from vllm.utils.math_utils import next_power_of_2 + from vllm.v1.worker.workspace import init_workspace_manager + + record_function_tracer = None + if args.profile_method == "record_function": + import sys + + sys.path.insert(0, str(args.frontier_source.resolve())) + from frontier.profiling.utils.record_function_tracer import RecordFunctionTracer + + record_function_tracer = RecordFunctionTracer + + device = torch.device(args.device) + torch.accelerator.set_device_index(device) + torch.manual_seed(20260716) + init_workspace_manager(args.device) + args.output.parent.mkdir(parents=True, exist_ok=True) + if args.profile_method == "record_function": + (args.output.parent / "profiler_traces").mkdir(exist_ok=True) + max_num_tokens = next_power_of_2(max(args.num_tokens)) + + rows: list[dict[str, Any]] = [] + for tp in args.tp: + parallel = FusedMoEParallelConfig( + tp_size=tp, + tp_rank=0, + pcp_size=1, + pcp_rank=0, + dp_size=1, + dp_rank=0, + ep_size=1, + ep_rank=0, + sp_size=1, + use_ep=False, + all2all_backend="allgather_reducescatter", + enable_eplb=False, + ) + moe_config = FusedMoEConfig( + num_experts=NUM_EXPERTS, + experts_per_token=TOP_K, + hidden_dim=HIDDEN_DIM, + intermediate_size_per_partition=INTERMEDIATE_DIM // tp, + num_local_experts=NUM_EXPERTS, + num_logical_experts=NUM_EXPERTS, + activation=MoEActivation.SILU, + device=device, + routing_method=RoutingMethodType.Renormalize, + moe_parallel_config=parallel, + in_dtype=torch.bfloat16, + max_num_tokens=max_num_tokens, + ) + # This process profiles one TP-local weight shard. Keep the global + # runtime context single-rank so vLLM does not initialize a collective; + # the action-conditioned shard size remains explicit in moe_config and + # the real TP2/TP4 all-reduce is profiled in a separate multi-GPU run. + vllm_config = VllmConfig( + parallel_config=ParallelConfig(tensor_parallel_size=1) + ) + with set_current_vllm_config(vllm_config): + backend, experts_cls = select_unquantized_moe_backend(moe_config) + if backend != UnquantizedMoeBackend.FLASHINFER_CUTLASS: + raise SystemExit( + "runtime backend mismatch: expected FlashInfer CUTLASS, " + f"got {backend.value} at TP={tp}" + ) + if experts_cls is None: + raise SystemExit(f"missing experts class for {backend.value}") + + w13_original = torch.empty( + (NUM_EXPERTS, 2 * (INTERMEDIATE_DIM // tp), HIDDEN_DIM), + device=device, + dtype=torch.bfloat16, + ).uniform_(-0.01, 0.01) + w2 = torch.empty( + (NUM_EXPERTS, HIDDEN_DIM, INTERMEDIATE_DIM // tp), + device=device, + dtype=torch.bfloat16, + ).uniform_(-0.01, 0.01) + + class Layer: + pass + + layer = Layer() + layer.moe_config = moe_config + w13_kernel, w2_kernel = convert_to_unquantized_kernel_format( + backend, + layer=layer, + w13_weight=w13_original, + w2_weight=w2, + ) + kernel = make_unquantized_moe_kernel( + quant_config=FUSED_MOE_UNQUANTIZED_CONFIG, + moe_config=moe_config, + backend=backend, + experts_cls=experts_cls, + ) + + reference_checked = False + for routing_mode in args.routing_modes: + for num_tokens in args.num_tokens: + hidden = torch.empty( + (num_tokens, HIDDEN_DIM), + device=device, + dtype=torch.bfloat16, + ).uniform_(-0.1, 0.1) + topk_weights, topk_ids, load = routing_inputs( + routing_mode, num_tokens, device + ) + + def run_kernel() -> torch.Tensor: + return kernel.apply( + hidden_states=hidden, + w1=w13_kernel, + w2=w2_kernel, + topk_weights=topk_weights, + topk_ids=topk_ids, + activation=MoEActivation.SILU, + global_num_experts=NUM_EXPERTS, + expert_map=None, + apply_router_weight_on_input=False, + ) + + if args.profile_method == "record_function": + output, time_ms = measure_kernel_only_ms( + run_kernel, + warmup_iters=args.warmup_iters, + repeats=args.repeats, + trace_root=args.output.parent, + operation_name="moe_grouped_gemm", + record_function_tracer=record_function_tracer, + ) + else: + for _ in range(args.warmup_iters): + output = run_kernel() + torch.accelerator.synchronize() + samples: list[float] = [] + for _ in range(args.repeats): + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + output = run_kernel() + end.record() + torch.accelerator.synchronize() + samples.append(float(start.elapsed_time(end))) + time_ms = stats_ms(samples) + + if output.shape != hidden.shape or not torch.isfinite(output).all(): + raise SystemExit( + f"invalid MoE output TP={tp} M={num_tokens} mode={routing_mode}" + ) + if args.check_reference and not reference_checked: + check_tokens = min(2, num_tokens) + reference = reference_partial_output( + hidden[:check_tokens], + w13_original, + w2, + topk_weights[:check_tokens], + topk_ids[:check_tokens], + ) + torch.testing.assert_close( + output[:check_tokens], reference, atol=0.03, rtol=0.03 + ) + reference_checked = True + + row = { + "tensor_parallel_size": tp, + "num_tokens": num_tokens, + "routing_mode": routing_mode, + "backend": backend.value, + "intermediate_size_per_partition": INTERMEDIATE_DIM // tp, + "output_is_reduced": kernel.output_is_reduced(), + "time_ms": time_ms, + "routing_load": load, + } + rows.append(row) + print( + json.dumps( + { + "tp": tp, + "num_tokens": num_tokens, + "routing_mode": routing_mode, + "backend": backend.value, + "median_ms": row["time_ms"]["median"], + }, + sort_keys=True, + ), + flush=True, + ) + + del kernel, w13_kernel, w2_kernel, w13_original, w2 + torch.accelerator.empty_cache() + + payload = { + "schema_version": "qwen30_vllm020_moe_raw.v1", + "environment": { + "vllm_version": vllm.__version__, + "vllm_source_commit": source_head, + "torch_version": torch.__version__, + "torch_cuda": torch.version.cuda, + "gpu": torch.cuda.get_device_name(device), + "model": str(args.model), + "dtype": "bfloat16", + "weight_quantization": "none", + "top_k": TOP_K, + "norm_topk_prob": True, + "profile_method": args.profile_method, + }, + "measurement_scope": ( + "one TP-local weight shard: vLLM modular MoE prepare+FlashInfer " + "CUTLASS experts+finalize; router linear/top-k and TP all-reduce excluded" + ), + "rows": rows, + } + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_router.py b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_router.py new file mode 100644 index 0000000..eac341f --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_router.py @@ -0,0 +1,280 @@ +#!/usr/bin/env python3 +"""Profile Qwen3's replicated MoE gate and fused top-k in vLLM 0.20.""" + +from __future__ import annotations + +import argparse +import json +import socket +import statistics +import subprocess +from pathlib import Path +from typing import Any, Callable + +import torch +import vllm + + +VLLM_VERSION = "0.20.0" +VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +HIDDEN_DIM = 2048 +NUM_EXPERTS = 128 +TOP_K = 8 + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--vllm-source", type=Path, required=True) + parser.add_argument("--model", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + parser.add_argument("--num-tokens", type=int, nargs="+", required=True) + parser.add_argument("--warmup-iters", type=int, default=5) + parser.add_argument("--repeats", type=int, default=20) + parser.add_argument("--device", default="cuda:0") + parser.add_argument( + "--profile-method", + choices=("cuda_event", "record_function"), + default="cuda_event", + ) + parser.add_argument("--frontier-source", type=Path) + return parser.parse_args() + + +def git_head(repo: Path) -> str: + return subprocess.check_output( + ["git", "-C", str(repo), "rev-parse", "HEAD"], text=True + ).strip() + + +def stats_ms(samples: list[float]) -> dict[str, float]: + return { + "min": min(samples), + "max": max(samples), + "mean": statistics.fmean(samples), + "median": statistics.median(samples), + "std": statistics.pstdev(samples), + } + + +def measure_ms( + fn: Callable[[], Any], + warmup_iters: int, + repeats: int, + *, + profile_method: str, + trace_root: Path | None = None, + operation_name: str | None = None, + record_function_tracer: type | None = None, +) -> tuple[Any, dict[str, float]]: + result = None + for _ in range(warmup_iters): + result = fn() + torch.accelerator.synchronize() + if profile_method == "record_function": + if trace_root is None or operation_name is None or record_function_tracer is None: + raise ValueError("record_function profiling requires tracer metadata") + tracer = record_function_tracer(str(trace_root)) + with tracer: + for _ in range(repeats): + with torch.profiler.record_function(f"vidur_{operation_name}"): + result = fn() + stats = tracer.get_operation_time_stats() + if operation_name not in stats: + raise RuntimeError(f"missing RecordFunctionTracer stats for {operation_name}") + return result, { + name: float(value) for name, value in stats[operation_name].items() + } + + samples: list[float] = [] + for _ in range(repeats): + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + start.record() + result = fn() + end.record() + torch.accelerator.synchronize() + samples.append(float(start.elapsed_time(end))) + return result, stats_ms(samples) + + +def main() -> None: + args = parse_args() + if vllm.__version__ != VLLM_VERSION: + raise SystemExit(f"expected vLLM {VLLM_VERSION}, got {vllm.__version__}") + source_head = git_head(args.vllm_source) + if source_head != VLLM_COMMIT: + raise SystemExit(f"expected vLLM source {VLLM_COMMIT}, got {source_head}") + if args.profile_method == "record_function" and args.frontier_source is None: + raise SystemExit("--frontier-source is required for --profile-method record_function") + + raw_model_config = json.loads(args.model.joinpath("config.json").read_text()) + observed = { + "hidden_size": raw_model_config.get("hidden_size"), + "num_experts": raw_model_config.get("num_experts"), + "num_experts_per_tok": raw_model_config.get("num_experts_per_tok"), + "norm_topk_prob": raw_model_config.get("norm_topk_prob"), + } + expected = { + "hidden_size": HIDDEN_DIM, + "num_experts": NUM_EXPERTS, + "num_experts_per_tok": TOP_K, + "norm_topk_prob": True, + } + if observed != expected: + raise SystemExit(f"model contract mismatch: expected {expected}, got {observed}") + + from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config + from vllm.distributed import ( + destroy_distributed_environment, + destroy_model_parallel, + init_distributed_environment, + initialize_model_parallel, + ) + from vllm.model_executor.layers.fused_moe import fused_topk + from vllm.model_executor.layers.linear import ReplicatedLinear + + record_function_tracer = None + if args.profile_method == "record_function": + import sys + + sys.path.insert(0, str(args.frontier_source.resolve())) + from frontier.profiling.utils.record_function_tracer import RecordFunctionTracer + + record_function_tracer = RecordFunctionTracer + + device = torch.device(args.device) + torch.accelerator.set_device_index(device) + torch.manual_seed(20260716) + model_config = ModelConfig( + model=str(args.model), + dtype="bfloat16", + max_model_len=8192, + skip_tokenizer_init=True, + generation_config="vllm", + ) + args.output.parent.mkdir(parents=True, exist_ok=True) + if args.profile_method == "record_function": + (args.output.parent / "profiler_traces").mkdir(exist_ok=True) + + rows: list[dict[str, Any]] = [] + with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as listener: + listener.bind(("127.0.0.1", 0)) + distributed_init_method = f"tcp://127.0.0.1:{listener.getsockname()[1]}" + init_distributed_environment( + world_size=1, + rank=0, + local_rank=0, + distributed_init_method=distributed_init_method, + ) + try: + with set_current_vllm_config(VllmConfig(model_config=model_config)): + initialize_model_parallel(tensor_model_parallel_size=1) + gate = ReplicatedLinear( + HIDDEN_DIM, + NUM_EXPERTS, + bias=False, + quant_config=None, + prefix="model.layers.0.mlp.gate", + ).to(device=device, dtype=torch.bfloat16) + gate.weight.data.uniform_(-0.01, 0.01) + + for num_tokens in args.num_tokens: + hidden = torch.empty( + (num_tokens, HIDDEN_DIM), device=device, dtype=torch.bfloat16 + ).uniform_(-0.1, 0.1) + logits, gate_time = measure_ms( + lambda: gate(hidden)[0], + args.warmup_iters, + args.repeats, + profile_method=args.profile_method, + trace_root=args.output.parent, + operation_name="moe_gating_linear", + record_function_tracer=record_function_tracer, + ) + topk_result, topk_time = measure_ms( + lambda: fused_topk(hidden, logits, TOP_K, renormalize=True), + args.warmup_iters, + args.repeats, + profile_method=args.profile_method, + trace_root=args.output.parent, + operation_name="moe_gating_routing_topk", + record_function_tracer=record_function_tracer, + ) + + def gate_and_topk() -> tuple[ + torch.Tensor, torch.Tensor, torch.Tensor + ]: + current_logits, _ = gate(hidden) + return fused_topk( + hidden, current_logits, TOP_K, renormalize=True + ) + + combined_result, combined_time = measure_ms( + gate_and_topk, + args.warmup_iters, + args.repeats, + profile_method=args.profile_method, + trace_root=args.output.parent, + operation_name="moe_gating_linear_and_routing_topk", + record_function_tracer=record_function_tracer, + ) + topk_weights, topk_ids, _ = topk_result + combined_weights, combined_ids, _ = combined_result + if logits.shape != (num_tokens, NUM_EXPERTS): + raise SystemExit( + f"invalid gate output shape: {tuple(logits.shape)}" + ) + if topk_ids.shape != (num_tokens, TOP_K): + raise SystemExit(f"invalid top-k shape: {tuple(topk_ids.shape)}") + torch.testing.assert_close( + topk_weights.sum(dim=-1), + torch.ones(num_tokens, device=device), + atol=1e-5, + rtol=1e-5, + ) + torch.testing.assert_close(combined_weights, topk_weights) + torch.testing.assert_close(combined_ids, topk_ids) + additive_median = gate_time["median"] + topk_time["median"] + row = { + "num_tokens": num_tokens, + "gate_linear_time_ms": gate_time, + "routing_topk_time_ms": topk_time, + "gate_plus_topk_time_ms": combined_time, + "median_nonadditivity_ratio": ( + combined_time["median"] / additive_median + if additive_median > 0 + else 1.0 + ), + } + rows.append(row) + print(json.dumps(row, sort_keys=True), flush=True) + finally: + destroy_model_parallel() + destroy_distributed_environment() + + payload = { + "schema_version": "qwen30_vllm020_router_raw.v1", + "environment": { + "vllm_version": vllm.__version__, + "vllm_source_commit": source_head, + "torch_version": torch.__version__, + "torch_cuda": torch.version.cuda, + "gpu": torch.cuda.get_device_name(device), + "model": str(args.model), + "dtype": "bfloat16", + "gate_replication": "replicated_across_tp", + "top_k": TOP_K, + "norm_topk_prob": True, + "profile_method": args.profile_method, + }, + "measurement_scope": ( + "vLLM ReplicatedLinear gate and fused_topk; measured separately and " + "as the actual sequential router path" + ), + "rows": rows, + } + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_allreduce_profile.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_allreduce_profile.sh new file mode 100644 index 0000000..0cf4565 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_allreduce_profile.sh @@ -0,0 +1,56 @@ +#!/usr/bin/env bash + +set -euo pipefail + +TP="${TP:?TP must be set to 2 or 4}" +case "${TP}" in + 2) HARD_GPU_CAP="0.40_H20h" ;; + 4) HARD_GPU_CAP="0.80_H20h" ;; + *) echo "ERROR: invalid TP=${TP}" >&2; exit 1 ;; +esac +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +NUM_TOKENS="${NUM_TOKENS:-8}" +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/profile.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPUs are required}" +if [[ "${#GPU_IDS[@]}" -ne "${TP}" ]]; then + echo "ERROR: TP=${TP} requires ${TP} GPUs, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +export VLLM_ALLREDUCE_USE_FLASHINFER=1 +export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=tensor_model_parallel_all_reduce backend=FlashInfer-TRTLLM tp=${TP} tokens=${NUM_TOKENS} hidden=2048 dtype=BF16 output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=720s hard_gpu_cap=${HARD_GPU_CAP}" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum "${MODEL}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256" +sha256sum profile_vllm020_allreduce.py run_allreduce_profile.sh \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" + +read -r -a TOKEN_ARGS <<< "${NUM_TOKENS}" +timeout --signal=TERM --kill-after=30s 600 \ + "${VENV_ROOT}/bin/torchrun" --standalone --nproc_per_node="${TP}" \ + profile_vllm020_allreduce.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/allreduce-tp${TP}.json" \ + --num-tokens "${TOKEN_ARGS[@]}" \ + --warmup-iters 3 \ + --repeats 10 + +test -s "${OUTPUT_ROOT}/raw/allreduce-tp${TP}.json" +sha256sum "${OUTPUT_ROOT}/raw/allreduce-tp${TP}.json" \ + "${OUTPUT_ROOT}/provenance"/* > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "ALLREDUCE_PROFILE_COMPLETE tp=${TP} tokens=${NUM_TOKENS}" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_full.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_full.sh new file mode 100644 index 0000000..5d84abc --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_full.sh @@ -0,0 +1,89 @@ +#!/usr/bin/env bash + +set -euo pipefail + +TP="${TP:?TP must be set to 1, 2, or 4}" +case "${TP}" in + 1|2|4) ;; + *) echo "ERROR: invalid TP=${TP}" >&2; exit 1 ;; +esac + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +LOG_DIR="${OUTPUT_ROOT}/logs" +PROVENANCE_DIR="${OUTPUT_ROOT}/provenance" + +PREFILL_SPECS=(q64 q128 q256 q512 q1k q2k q4k q8k) +PREFIX_EXTEND_SPECS=(q64s512 q128s1k q256s2k q512s4k q1ks8k) +DECODE_SPECS=( + q1s128 8q1s128 16q1s128 32q1s128 64q1s128 + 8q1s1k 16q1s1k 32q1s1k 64q1s1k + 8q1s2k 16q1s2k 32q1s2k 64q1s2k + 8q1s4k 16q1s4k 32q1s4k 64q1s4k + 8q1s8k 16q1s8k 32q1s8k 64q1s8k +) +MIXED_SPECS=( + q64_8q1s512 + q128_8q1s1k q128_16q1s1k + q256_16q1s2k q256_32q1s2k + q512_32q1s4k q512_64q1s4k + q1k_64q1s8k q2k_32q1s4k q4k_16q1s4k +) +BATCH_SPECS=( + "${PREFILL_SPECS[@]}" + "${PREFIX_EXTEND_SPECS[@]}" + "${DECODE_SPECS[@]}" + "${MIXED_SPECS[@]}" +) + +mkdir -p "${LOG_DIR}" "${PROVENANCE_DIR}" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${LOG_DIR}/full.log") 2>&1 + +if [[ -z "${CUDA_VISIBLE_DEVICES:-}" ]]; then + echo "ERROR: CUDA_VISIBLE_DEVICES must contain the fleet-allocated GPU" >&2 + exit 1 +fi +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operators=FlashAttention3,reshape_and_cache_flash tp=${TP} cases=${#BATCH_SPECS[@]} regimes=prefill,prefix_extend,decode,true_mixed dtype=BF16 block=16 output=${OUTPUT_ROOT} expected_wall=4-10m hard_wall=1200s hard_gpu_cap=0.33_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test -x "${VENV_ROOT}/bin/python" +test -f "${VLLM_SOURCE}/benchmarks/attention_benchmarks/runner.py" +test -f "${MODEL}/config.json" +test -f profile_vllm020_flashattn.py + +git rev-parse HEAD > "${PROVENANCE_DIR}/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${PROVENANCE_DIR}/vllm-source.commit" +sha256sum profile_vllm020_flashattn.py run_flashattn_full.sh \ + > "${PROVENANCE_DIR}/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${PROVENANCE_DIR}/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${PROVENANCE_DIR}/gpus.csv" +printf '%s\n' "${BATCH_SPECS[@]}" > "${PROVENANCE_DIR}/batch-specs.txt" + +timeout --signal=TERM --kill-after=30s 1080 \ + "${VENV_ROOT}/bin/python" profile_vllm020_flashattn.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" \ + --tp "${TP}" \ + --batch-specs "${BATCH_SPECS[@]}" \ + --warmup-iters 5 \ + --repeats 10 \ + --profile-kv-update + +test -s "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" +sha256sum "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" \ + "${PROVENANCE_DIR}"/* > "${OUTPUT_ROOT}/artifacts.sha256" +nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FLASHATTN_FULL_COMPLETE tp=${TP} cases=${#BATCH_SPECS[@]}" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_smoke.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_smoke.sh new file mode 100644 index 0000000..327a261 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_smoke.sh @@ -0,0 +1,59 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:-$(pwd)/artifacts/flashattn-smoke}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +LOG_DIR="${OUTPUT_ROOT}/logs" +PROVENANCE_DIR="${OUTPUT_ROOT}/provenance" + +mkdir -p "${LOG_DIR}" "${PROVENANCE_DIR}" +exec > >(tee -a "${LOG_DIR}/smoke.log") 2>&1 + +if [[ -z "${CUDA_VISIBLE_DEVICES:-}" ]]; then + echo "ERROR: CUDA_VISIBLE_DEVICES must contain the fleet-allocated GPU" >&2 + exit 1 +fi +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operators=FlashAttention3,reshape_and_cache_flash tp_local_shapes=1,2,4 specs=q128,4q1s128,q128_4q1s128 dtype=BF16 block=16 output=${OUTPUT_ROOT} expected_wall=3-8m hard_wall=900s hard_gpu_cap=0.15_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test -x "${VENV_ROOT}/bin/python" +test -f "${VLLM_SOURCE}/benchmarks/attention_benchmarks/runner.py" +test -f "${MODEL}/config.json" +test -f profile_vllm020_flashattn.py + +git rev-parse HEAD > "${PROVENANCE_DIR}/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${PROVENANCE_DIR}/vllm-source.commit" +sha256sum profile_vllm020_flashattn.py run_flashattn_smoke.sh \ + > "${PROVENANCE_DIR}/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${PROVENANCE_DIR}/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${PROVENANCE_DIR}/gpus.csv" + +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" profile_vllm020_flashattn.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/flashattn-smoke.json" \ + --tp 1 2 4 \ + --batch-specs q128 4q1s128 q128_4q1s128 \ + --warmup-iters 3 \ + --repeats 5 \ + --profile-kv-update + +test -s "${OUTPUT_ROOT}/raw/flashattn-smoke.json" +sha256sum "${OUTPUT_ROOT}/raw/flashattn-smoke.json" \ + "${PROVENANCE_DIR}"/* > "${OUTPUT_ROOT}/artifacts.sha256" +nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FLASHATTN_SMOKE_COMPLETE" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_full.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_full.sh new file mode 100644 index 0000000..e68cf88 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_full.sh @@ -0,0 +1,63 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +FRONTIER_ROOT="${FRONTIER_ROOT:-/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +export MODEL_ROOT +RUN_DIR="$(pwd -P)" +TOKENS=(1 8 16 32 64 128 256 512 1024 2048 4096 8192) +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/profiles" +exec > >(tee -a "${OUTPUT_ROOT}/logs/full.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=qwen3-a3b-30b-moe runtime=vLLM-0.20.0+cu129 profiler=Frontier@d9cfeb6 operator=linear,norm,rope,embedding tp=1,2,4 tokens=${TOKENS[*]} precision=BF16 method=cuda_event output=${OUTPUT_ROOT} expected_wall=3-10m hard_wall=1500s hard_gpu_cap=0.42_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test "$(git -C "${FRONTIER_ROOT}" rev-parse HEAD)" = "d9cfeb6d8791fbf2f295dd9744c56a666171776e" +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +test -s "${MODEL_ROOT}/config.json" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${FRONTIER_ROOT}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256" +sha256sum run_frontier_linear_full.sh frontier_vllm020_compat.py \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" +printf '%s\n' "${TOKENS[@]}" > "${OUTPUT_ROOT}/provenance/tokens.txt" + +cd "${FRONTIER_ROOT}" +timeout --signal=TERM --kill-after=30s 1380 \ + "${VENV_ROOT}/bin/python" "${RUN_DIR}/frontier_vllm020_compat.py" \ + --disable_ray \ + --num_gpus 1 \ + --output_dir "${OUTPUT_ROOT}/profiles" \ + --device h20 \ + --models qwen3-a3b-30b-moe \ + --num_tensor_parallel_workers 1 2 4 \ + --max_tokens 8192 \ + --num_tokens_list "${TOKENS[@]}" \ + --profile_method cuda_event \ + --precision BF16 \ + --is_moe \ + --yes + +find "${OUTPUT_ROOT}/profiles" -type f -size +0c -print \ + > "${OUTPUT_ROOT}/provenance/profile-files.txt" +test -s "${OUTPUT_ROOT}/provenance/profile-files.txt" +find "${OUTPUT_ROOT}/profiles" -type f -print0 | sort -z | xargs -0 sha256sum \ + > "${OUTPUT_ROOT}/profiles.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FRONTIER_LINEAR_FULL_COMPLETE" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_smoke.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_smoke.sh new file mode 100644 index 0000000..eaa2f7e --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_frontier_linear_smoke.sh @@ -0,0 +1,60 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +FRONTIER_ROOT="${FRONTIER_ROOT:-/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +export MODEL_ROOT +RUN_DIR="$(pwd -P)" +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/profiles" +exec > >(tee -a "${OUTPUT_ROOT}/logs/smoke.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=qwen3-a3b-30b-moe runtime=vLLM-0.20.0+cu129 profiler=Frontier@d9cfeb6 operator=linear,norm,rope,embedding tp=1,2,4 tokens=8 precision=BF16 method=cuda_event output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=900s hard_gpu_cap=0.25_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test "$(git -C "${FRONTIER_ROOT}" rev-parse HEAD)" = "d9cfeb6d8791fbf2f295dd9744c56a666171776e" +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +test -s "${MODEL_ROOT}/config.json" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${FRONTIER_ROOT}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256" +sha256sum run_frontier_linear_smoke.sh frontier_vllm020_compat.py \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" + +cd "${FRONTIER_ROOT}" +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" "${RUN_DIR}/frontier_vllm020_compat.py" \ + --disable_ray \ + --num_gpus 1 \ + --output_dir "${OUTPUT_ROOT}/profiles" \ + --device h20 \ + --models qwen3-a3b-30b-moe \ + --num_tensor_parallel_workers 1 2 4 \ + --num_tokens_list 8 \ + --profile_method cuda_event \ + --precision BF16 \ + --is_moe \ + --yes + +find "${OUTPUT_ROOT}/profiles" -type f -size +0c -print \ + > "${OUTPUT_ROOT}/provenance/profile-files.txt" +test -s "${OUTPUT_ROOT}/provenance/profile-files.txt" +find "${OUTPUT_ROOT}/profiles" -type f -print0 | sort -z | xargs -0 sha256sum \ + > "${OUTPUT_ROOT}/profiles.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "FRONTIER_LINEAR_SMOKE_COMPLETE" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_moe_full.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_moe_full.sh new file mode 100644 index 0000000..374a78b --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_moe_full.sh @@ -0,0 +1,51 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +TOKENS=(1 8 16 32 64 128 256 512 1024 2048 4096 8192) +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/full.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=FlashInfer-CUTLASS-unquantized-MoE tp_local_shapes=1,2,4 tokens=${TOKENS[*]} routing=uniform_random_logits,hotset8 reference_check=on dtype=BF16 output=${OUTPUT_ROOT} expected_wall=4-12m hard_wall=1500s hard_gpu_cap=0.42_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test -x "${VENV_ROOT}/bin/python" +test -f "${MODEL}/config.json" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum profile_vllm020_moe.py run_moe_full.sh \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" +printf '%s\n' "${TOKENS[@]}" > "${OUTPUT_ROOT}/provenance/tokens.txt" + +timeout --signal=TERM --kill-after=30s 1380 \ + "${VENV_ROOT}/bin/python" profile_vllm020_moe.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/moe-full.json" \ + --tp 1 2 4 \ + --num-tokens "${TOKENS[@]}" \ + --routing-modes uniform_random_logits hotset8 \ + --warmup-iters 5 \ + --repeats 10 \ + --check-reference + +test -s "${OUTPUT_ROOT}/raw/moe-full.json" +sha256sum "${OUTPUT_ROOT}/raw/moe-full.json" "${OUTPUT_ROOT}/provenance"/* \ + > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "MOE_FULL_COMPLETE rows=72" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_moe_smoke.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_moe_smoke.sh new file mode 100644 index 0000000..af7b784 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_moe_smoke.sh @@ -0,0 +1,47 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/smoke.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=FlashInfer-CUTLASS-unquantized-MoE tp_local_shapes=1,2,4 tokens=8 routing=uniform_random_logits reference_check=on dtype=BF16 output=${OUTPUT_ROOT} expected_wall=3-8m hard_wall=900s hard_gpu_cap=0.25_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum profile_vllm020_moe.py run_moe_smoke.sh \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" + +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" profile_vllm020_moe.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/moe-smoke.json" \ + --tp 1 2 4 \ + --num-tokens 8 \ + --routing-modes uniform_random_logits \ + --warmup-iters 3 \ + --repeats 5 \ + --check-reference + +test -s "${OUTPUT_ROOT}/raw/moe-smoke.json" +sha256sum "${OUTPUT_ROOT}/raw/moe-smoke.json" "${OUTPUT_ROOT}/provenance"/* \ + > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "MOE_SMOKE_COMPLETE" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_router_full.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_router_full.sh new file mode 100644 index 0000000..b4fd7a5 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_router_full.sh @@ -0,0 +1,49 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +TOKENS=(1 8 16 32 64 128 256 512 1024 2048 4096 8192) +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/full.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operators=ReplicatedLinear,fused_topk tokens=${TOKENS[*]} dtype=BF16 output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=900s hard_gpu_cap=0.25_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +test -s "${MODEL}/config.json" +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum profile_vllm020_router.py run_router_full.sh \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +sha256sum "${MODEL}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" +printf '%s\n' "${TOKENS[@]}" > "${OUTPUT_ROOT}/provenance/tokens.txt" + +timeout --signal=TERM --kill-after=30s 780 \ + "${VENV_ROOT}/bin/python" profile_vllm020_router.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --output "${OUTPUT_ROOT}/raw/router.json" \ + --num-tokens "${TOKENS[@]}" \ + --warmup-iters 5 \ + --repeats 20 + +test -s "${OUTPUT_ROOT}/raw/router.json" +sha256sum "${OUTPUT_ROOT}/raw/router.json" "${OUTPUT_ROOT}/provenance"/* \ + > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "ROUTER_FULL_COMPLETE cases=${#TOKENS[@]}" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/run_trace_routing.sh b/runs/frontier-qwen30-vllm020-profile-v1/run_trace_routing.sh new file mode 100644 index 0000000..16838de --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/run_trace_routing.sh @@ -0,0 +1,50 @@ +#!/usr/bin/env bash + +set -euo pipefail + +OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}" +FIXTURE="${FIXTURE:?FIXTURE must be set}" +VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}" +VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}" +MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}" +mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw" +exec > >(tee -a "${OUTPUT_ROOT}/logs/full.log") 2>&1 + +IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}" +if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then + echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2 + exit 1 +fi + +echo "ROUTING_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 trace_fixture=${FIXTURE} fixture_sha256=e9e7f5b4e0d3a788dcd99d432f939d9e36bff2a64e412649b407b0609f0e39bb requests=8 input_tokens_trace=23673 output_tokens_trace=6502 TP=1 MBT=8192 MNS=64 prefix_cache=true chunked_prefill=true dtype=BF16 output=${OUTPUT_ROOT} expected_wall=15-40m hard_wall=3600s hard_gpu_cap=1.0_H20h" +date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ" +nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader + +test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1" +test -s "${MODEL}/config.json" +echo "e9e7f5b4e0d3a788dcd99d432f939d9e36bff2a64e412649b407b0609f0e39bb ${FIXTURE}" | sha256sum -c - +git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit" +git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit" +sha256sum capture_trace_routing.py run_trace_routing.sh \ + > "${OUTPUT_ROOT}/provenance/source.sha256" +sha256sum "${MODEL}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256" +sha256sum "${FIXTURE}" > "${OUTPUT_ROOT}/provenance/fixture.sha256" +uv pip freeze --python "${VENV_ROOT}/bin/python" \ + > "${OUTPUT_ROOT}/provenance/pip-freeze.txt" +nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \ + --format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv" + +timeout --signal=TERM --kill-after=30s 3300 \ + "${VENV_ROOT}/bin/python" capture_trace_routing.py \ + --vllm-source "${VLLM_SOURCE}" \ + --model "${MODEL}" \ + --fixture "${FIXTURE}" \ + --output "${OUTPUT_ROOT}/raw/routing.json" \ + --routes "${OUTPUT_ROOT}/raw/routes.npz" + +test -s "${OUTPUT_ROOT}/raw/routing.json" +test -s "${OUTPUT_ROOT}/raw/routes.npz" +sha256sum "${OUTPUT_ROOT}/raw/routing.json" "${OUTPUT_ROOT}/raw/routes.npz" \ + "${OUTPUT_ROOT}/provenance"/* > "${OUTPUT_ROOT}/artifacts.sha256" +date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ" +echo "TRACE_ROUTING_COMPLETE requests=8" diff --git a/runs/frontier-qwen30-vllm020-profile-v1/summarize_opprof.py b/runs/frontier-qwen30-vllm020-profile-v1/summarize_opprof.py new file mode 100644 index 0000000..4b67f57 --- /dev/null +++ b/runs/frontier-qwen30-vllm020-profile-v1/summarize_opprof.py @@ -0,0 +1,144 @@ +#!/usr/bin/env python3 +"""Summarize prompt-free vLLM scheduler/operator-profiler step records.""" + +from __future__ import annotations + +import argparse +import json +import math +import statistics +from collections import Counter, defaultdict +from pathlib import Path +from typing import Any + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--cell-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + return parser.parse_args() + + +def percentile(values: list[float], fraction: float) -> float: + ordered = sorted(values) + index = min(len(ordered) - 1, max(0, math.ceil(fraction * len(ordered)) - 1)) + return ordered[index] + + +def stats(values: list[float]) -> dict[str, float]: + return { + "min": min(values), + "mean": statistics.fmean(values), + "p50": percentile(values, 0.50), + "p95": percentile(values, 0.95), + "p99": percentile(values, 0.99), + "max": max(values), + } + + +def summarize_cell(path: Path) -> dict[str, Any]: + streams = sorted(path.glob("opprof/*.jsonl")) + if len(streams) != 1: + raise ValueError(f"expected one opprof stream in {path}, got {len(streams)}") + groups: dict[tuple[str, str], dict[str, list[float]]] = defaultdict( + lambda: defaultdict(list) + ) + phase_counts: Counter[str] = Counter() + graph_counts: Counter[str] = Counter() + prefix_queries = 0 + prefix_hits = 0 + total_records = 0 + dropped_records_max = 0 + with streams[0].open() as handle: + for line in handle: + record = json.loads(line) + if not record.get("model_executed", False): + continue + prefill_tokens = int(record.get("prefill_tokens", 0)) + decode_tokens = int(record.get("decode_tokens", 0)) + if prefill_tokens and decode_tokens: + phase = "true_mixed" + elif prefill_tokens: + phase = "pure_prefill" + elif decode_tokens: + phase = "pure_decode" + else: + phase = "empty" + graph = str((record.get("cudagraph") or {}).get("runtime_mode", "UNKNOWN")) + duration_ms = ( + int(record["complete_mono_ns"]) - int(record["submit_mono_ns"]) + ) / 1e6 + if duration_ms < 0: + raise ValueError(f"negative duration in {streams[0]}") + group = groups[(phase, graph)] + group["submit_to_complete_ms"].append(duration_ms) + group["scheduled_requests"].append(int(record["scheduled_requests"])) + group["prefill_tokens"].append(prefill_tokens) + group["decode_tokens"].append(decode_tokens) + group["total_tokens"].append(prefill_tokens + decode_tokens) + group["queue_waiting"].append(int(record["queues"]["waiting"])) + group["kv_usage"].append(float(record["kv"]["usage"])) + local_prefix = (record.get("prefix") or {}).get("local") or {} + prefix_queries += int(local_prefix.get("queries", 0)) + prefix_hits += int(local_prefix.get("hits", 0)) + dropped_records_max = max( + dropped_records_max, int(record.get("dropped_records_before", 0)) + ) + phase_counts[phase] += 1 + graph_counts[graph] += 1 + total_records += 1 + if total_records == 0: + raise ValueError(f"empty opprof stream: {streams[0]}") + summarized_groups = [] + for (phase, graph), metrics in sorted(groups.items()): + summarized_groups.append( + { + "phase": phase, + "cudagraph_runtime_mode": graph, + "steps": len(metrics["submit_to_complete_ms"]), + **{name: stats(values) for name, values in metrics.items()}, + } + ) + cell = path.name + tp_text, mns_text = cell.split("_") + return { + "cell": cell, + "tensor_parallel_size": int(tp_text.removeprefix("tp")), + "max_num_seqs": int(mns_text.removeprefix("mns")), + "stream": str(streams[0]), + "records": total_records, + "phase_counts": dict(phase_counts), + "cudagraph_runtime_mode_counts": dict(graph_counts), + "prefix": { + "queries": prefix_queries, + "hits": prefix_hits, + "hit_rate": prefix_hits / prefix_queries if prefix_queries else 0.0, + }, + "dropped_records_before_max": dropped_records_max, + "groups": summarized_groups, + } + + +def main() -> None: + args = parse_args() + cells = [summarize_cell(path) for path in sorted(args.cell_root.glob("tp*_mns*"))] + if not cells: + raise SystemExit(f"no cells found in {args.cell_root}") + payload = { + "schema_version": "qwen30_vllm020_opprof_summary.v1", + "measurement_semantics": ( + "complete_mono_ns - submit_mono_ns from the existing vLLM opprof record; " + "this includes the instrumented submit-to-completion interval and is not " + "claimed to be a single-kernel CUDA-event duration" + ), + "contains_prompt_text": False, + "cell_root": str(args.cell_root), + "cells": cells, + } + args.output.parent.mkdir(parents=True, exist_ok=True) + args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + print(json.dumps({"cells": len(cells), "records": sum(c["records"] for c in cells)})) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v0/analyze_p1.py b/runs/intervention-response-v0/analyze_p1.py new file mode 100644 index 0000000..90f44ab --- /dev/null +++ b/runs/intervention-response-v0/analyze_p1.py @@ -0,0 +1,691 @@ +#!/usr/bin/env python3 +"""Prospective-repeat confirmation of the intervention-response hypothesis. + +P1 contains three pre-arranged, disjoint request bands per cell/load. TP1 and +TP4 use matched offered loads and request sequences across their MNS endpoints. +This script asks both whether the MNS response exceeds prospective repeat noise +and whether an early telemetry delta predicts full-run action efficacy beyond +the corresponding external-outcome delta. +""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +import re +import sys +from collections import defaultdict +from pathlib import Path +from statistics import fmean +from typing import Any, Iterable, Mapping + + +HERE = Path(__file__).resolve().parent +COMMON_STATE_DIR = HERE.parent / "telemetry-residual" +sys.path.insert(0, str(COMMON_STATE_DIR)) + +from common_state import load_jsonl, summarize_engine # noqa: E402 + + +def _load_v0(): + spec = importlib.util.spec_from_file_location( + "intervention_response_phase6_v0", HERE / "analyze_phase6.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +V0 = _load_v0() +SCHEMA = "intervention-response-p1-confirmation-v1" +HORIZONS_S = V0.HORIZONS_S +EXPECTED_ACTION_PAIRS = 12 +EXPECTED_REPEAT_PAIRS = 24 +MIN_EFFICACY_CLASS = 4 +MIN_EFFICACY_BALANCED_ACCURACY = 0.75 +MIN_EFFICACY_DELTA_OVER_OUTCOME = 0.15 +OUTCOME_FEATURES = ( + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "ttft_max_over_slo_max", + "ttft_mean_over_slo_max", + "tpot_max_over_slo", + "tpot_mean_over_slo", + "admitted_input_tokens_mean_over_limit", +) +RUN_PATTERN = re.compile(r"^(low|high)-rep([123])$") + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def _prefix_outcome( + result: Mapping[str, Any], + requests: list[dict[str, Any]], + horizon_s: float, +) -> dict[str, float]: + admitted = [request for request in requests if float(request["arrival_s"]) <= horizon_s] + completed = [ + request + for request in requests + if request.get("completed_elapsed_s") is not None + and float(request["completed_elapsed_s"]) <= horizon_s + ] + if not admitted: + raise ValueError("prefix contains no admitted request") + admitted_ids = {str(request["request_id"]) for request in admitted} + if any(str(request["request_id"]) not in admitted_ids for request in completed): + raise ValueError("completed request was not admitted in the prefix") + passed = sum(bool(request["slo_pass"]) for request in completed) + ttft = [float(request["ttft_ms"]) for request in completed] + tpot = [float(request["tpot_ms"]) for request in completed] + total = int(result["selection"]["count"]) + if total != len(requests): + raise ValueError("request JSONL count does not match the result") + return { + "admitted_fraction": len(admitted) / total, + "completed_over_admitted": len(completed) / len(admitted), + "completed_pass_rate": passed / max(1, len(completed)), + "completed_fail_fraction_of_total": (len(completed) - passed) / total, + "outstanding_over_admitted": (len(admitted) - len(completed)) / len(admitted), + "ttft_max_over_slo_max": max(ttft, default=0.0) / 6000.0, + "ttft_mean_over_slo_max": fmean(ttft) / 6000.0 if ttft else 0.0, + "tpot_max_over_slo": max(tpot, default=0.0) / 50.0, + "tpot_mean_over_slo": fmean(tpot) / 50.0 if tpot else 0.0, + "admitted_input_tokens_mean_over_limit": fmean( + float(request["raw_input_tokens"]) for request in admitted + ) + / 8192.0, + } + + +def load_trials( + run_root: Path, + *, + horizons_s: tuple[float, ...] = HORIZONS_S, +) -> tuple[dict[float, list[dict[str, Any]]], list[dict[str, Any]]]: + by_horizon = {horizon: [] for horizon in horizons_s} + streams = [] + for cell_dir in sorted((run_root / "cells").iterdir()): + if not cell_dir.is_dir(): + continue + stream_paths = sorted((cell_dir / "opprof").glob("*.jsonl")) + if len(stream_paths) != 1: + raise ValueError(f"{cell_dir}: expected one Layer-1 stream") + stream_path = stream_paths[0] + stream = load_jsonl(stream_path) + streams.append( + { + "path": str(stream_path.resolve()), + "sha256": sha256_file(stream_path), + "bytes": stream_path.stat().st_size, + } + ) + for run_dir in sorted(cell_dir.iterdir()): + match = RUN_PATTERN.match(run_dir.name) + if match is None: + continue + level, replicate_text = match.groups() + replicate = int(replicate_text) + result_path = run_dir / "result.json" + requests_path = run_dir / "requests.jsonl" + result = json.loads(result_path.read_text(encoding="utf-8")) + requests = load_jsonl(requests_path) + elapsed_s = float(result["interval"]["elapsed_s"]) + start_ns = int(result["interval"]["start_mono_ns"]) + for horizon_s in horizons_s: + if elapsed_s < horizon_s: + raise ValueError( + f"{result_path}: elapsed {elapsed_s} shorter than {horizon_s}s" + ) + state = V0.flatten_state( + summarize_engine( + stream, + start_ns=start_ns, + end_ns=start_ns + int(horizon_s * 1e9), + request_count=int(result["selection"]["count"]), + ) + ) + by_horizon[horizon_s].append( + { + "trial_id": str(result_path.relative_to(run_root)), + "cell": str(result["cell"]), + "tp": int(result["tp"]), + "mns": int(result["mns"]), + "level": level, + "replicate": replicate, + "offered_rate_per_gpu": float( + result["selection"]["offered_req_s_per_gpu"] + ), + "request_hash": str( + result["selection"]["request_id_order_sha256"] + ), + "request_count": int(result["selection"]["count"]), + "result_sha256": sha256_file(result_path), + "requests_sha256": sha256_file(requests_path), + "full_pass_rate": float(result["pass_rate"]), + "full_feasible": bool(result["feasible"]), + "early_stopped": bool(result["early_stopped"]), + "state": state, + "outcome": _prefix_outcome(result, requests, horizon_s), + } + ) + return by_horizon, streams + + +def validate_manifest( + trials: list[dict[str, Any]], manifest_path: Path +) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + if manifest.get("schema") != "fidelity-prefix-pilot-manifest-v1": + raise ValueError("unexpected P1 manifest schema") + cells = manifest.get("cells") + if not isinstance(cells, dict): + raise ValueError("P1 manifest has no cell mapping") + seen = set() + for trial in trials: + key = (trial["cell"], trial["level"], trial["replicate"]) + if key in seen: + raise ValueError(f"duplicate P1 trial identity: {key}") + seen.add(key) + try: + cell = cells[trial["cell"]] + selection = cell["targets"][trial["level"]]["selections"][ + f"{trial['level']}{trial['replicate']}" + ] + except (KeyError, TypeError) as error: + raise ValueError(f"trial is absent from P1 manifest: {key}") from error + if int(cell["tp"]) != trial["tp"] or int(cell["mns"]) != trial["mns"]: + raise ValueError(f"trial config disagrees with P1 manifest: {key}") + if str(selection["request_id_order_sha256"]) != trial["request_hash"]: + raise ValueError(f"trial request hash disagrees with P1 manifest: {key}") + if int(selection["selected_count"]) != trial["request_count"]: + raise ValueError(f"trial request count disagrees with P1 manifest: {key}") + if not math.isclose( + float(selection["offered_req_s_per_gpu"]), + trial["offered_rate_per_gpu"], + rel_tol=0.0, + abs_tol=1e-12, + ): + raise ValueError(f"trial offered load disagrees with P1 manifest: {key}") + expected = { + (cell_name, level, replicate) + for cell_name in cells + for level in ("low", "high") + for replicate in (1, 2, 3) + } + if seen != expected: + missing = sorted(expected - seen) + unexpected = sorted(seen - expected) + raise ValueError( + f"P1 trial/manifest coverage mismatch: missing={missing}, " + f"unexpected={unexpected}" + ) + return { + "schema": str(manifest["schema"]), + "expected_trials": len(expected), + "matched_trials": len(seen), + } + + +def _delta( + source: Mapping[str, Any], + target: Mapping[str, Any], + features: Iterable[str], +) -> dict[str, float]: + return { + feature: float(target[feature]) - float(source[feature]) + for feature in features + } + + +def _action_pair(source: Mapping[str, Any], target: Mapping[str, Any]) -> dict[str, Any]: + if source["tp"] != target["tp"]: + raise ValueError("action endpoints changed TP") + if source["level"] != target["level"] or source["replicate"] != target["replicate"]: + raise ValueError("action endpoints changed load role or repeat") + if source["request_hash"] != target["request_hash"]: + raise ValueError("action endpoints changed request sequence") + if not math.isclose( + source["offered_rate_per_gpu"], + target["offered_rate_per_gpu"], + rel_tol=0.0, + abs_tol=1e-12, + ): + raise ValueError("action endpoints changed offered load") + if source["mns"] >= target["mns"]: + raise ValueError("action must increase MNS") + beneficial = target["full_feasible"] and not source["full_feasible"] + return { + "kind": "matched_mns_increase", + "group": { + "tp": source["tp"], + "level": source["level"], + "replicate": source["replicate"], + "request_hash": source["request_hash"], + "offered_rate_per_gpu": source["offered_rate_per_gpu"], + }, + "source": { + key: source[key] + for key in ( + "trial_id", + "result_sha256", + "requests_sha256", + "cell", + "mns", + "full_pass_rate", + "full_feasible", + "early_stopped", + ) + }, + "target": { + key: target[key] + for key in ( + "trial_id", + "result_sha256", + "requests_sha256", + "cell", + "mns", + "full_pass_rate", + "full_feasible", + "early_stopped", + ) + }, + "delta_state": _delta(source["state"], target["state"], V0.ALL_FEATURES), + "delta_outcome": _delta(source["outcome"], target["outcome"], OUTCOME_FEATURES), + "full_action_efficacy": int(beneficial), + "full_feasibility_transition": ( + f"{str(source['full_feasible']).lower()}->" + f"{str(target['full_feasible']).lower()}" + ), + } + + +def _repeat_pair(source: Mapping[str, Any], target: Mapping[str, Any]) -> dict[str, Any]: + if source["cell"] != target["cell"] or source["level"] != target["level"]: + raise ValueError("repeat endpoints changed config or load role") + if target["replicate"] != source["replicate"] + 1: + raise ValueError("repeat endpoints are not consecutive pre-arranged bands") + if not math.isclose( + source["offered_rate_per_gpu"], + target["offered_rate_per_gpu"], + rel_tol=0.0, + abs_tol=1e-12, + ): + raise ValueError("repeat endpoints changed offered load") + return { + "kind": "same_config_workload_repeat", + "group": { + "cell": source["cell"], + "tp": source["tp"], + "mns": source["mns"], + "level": source["level"], + "source_replicate": source["replicate"], + "target_replicate": target["replicate"], + }, + "source": { + key: source[key] + for key in ("trial_id", "result_sha256", "requests_sha256") + }, + "target": { + key: target[key] + for key in ("trial_id", "result_sha256", "requests_sha256") + }, + "delta_state": _delta(source["state"], target["state"], V0.ALL_FEATURES), + "delta_outcome": _delta(source["outcome"], target["outcome"], OUTCOME_FEATURES), + } + + +def build_pairs( + trials: list[dict[str, Any]], +) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: + action_groups: dict[tuple[Any, ...], list[dict[str, Any]]] = defaultdict(list) + repeat_groups: dict[tuple[str, str], list[dict[str, Any]]] = defaultdict(list) + for trial in trials: + action_groups[ + ( + trial["tp"], + trial["level"], + trial["replicate"], + trial["request_hash"], + trial["offered_rate_per_gpu"], + ) + ].append(trial) + repeat_groups[(trial["cell"], trial["level"])].append(trial) + + actions = [] + for group in action_groups.values(): + if len(group) != 2: + continue + source, target = sorted(group, key=lambda trial: trial["mns"]) + actions.append(_action_pair(source, target)) + + repeats = [] + for group in repeat_groups.values(): + ordered = sorted(group, key=lambda trial: trial["replicate"]) + if len(ordered) != 3: + raise ValueError("each prospective repeat group must contain three runs") + repeats.extend( + _repeat_pair(source, target) + for source, target in zip(ordered, ordered[1:], strict=False) + ) + return actions, repeats + + +def _balanced_accuracy(labels: list[int], predictions: list[int]) -> float: + positive = [prediction for label, prediction in zip(labels, predictions) if label == 1] + negative = [prediction for label, prediction in zip(labels, predictions) if label == 0] + if not positive or not negative: + raise ValueError("balanced accuracy requires both classes") + sensitivity = sum(prediction == 1 for prediction in positive) / len(positive) + specificity = sum(prediction == 0 for prediction in negative) / len(negative) + return (sensitivity + specificity) / 2.0 + + +def _threshold_candidates(values: list[float]) -> list[float]: + unique = sorted(set(values)) + if len(unique) == 1: + return [unique[0] - 1.0, unique[0], unique[0] + 1.0] + scale = max(1.0, max(abs(value) for value in unique)) + candidates = [unique[0] - scale * 1e-6] + candidates.extend( + (left + right) / 2.0 + for left, right in zip(unique, unique[1:], strict=False) + ) + candidates.append(unique[-1] + scale * 1e-6) + return candidates + + +def _fit_threshold(values: list[float], labels: list[int]) -> tuple[float, int, float]: + best: tuple[float, int, float, float] | None = None + for threshold in _threshold_candidates(values): + for direction in (-1, 1): + predictions = [int(direction * (value - threshold) >= 0.0) for value in values] + balanced = _balanced_accuracy(labels, predictions) + accuracy = sum( + prediction == label + for prediction, label in zip(predictions, labels, strict=True) + ) / len(labels) + candidate = (balanced, accuracy, -abs(threshold), float(direction)) + if best is None or candidate > best: + best = candidate + selected_threshold = threshold + selected_direction = direction + assert best is not None + return selected_threshold, selected_direction, best[0] + + +def one_feature_leave_repeat_out( + actions: list[dict[str, Any]], + *, + delta_key: str, + features: tuple[str, ...], +) -> dict[str, Any]: + labels = [int(pair["full_action_efficacy"]) for pair in actions] + results = {} + for feature in features: + predictions = [] + held_out_labels = [] + folds = [] + for held_out in (1, 2, 3): + train = [pair for pair in actions if pair["group"]["replicate"] != held_out] + test = [pair for pair in actions if pair["group"]["replicate"] == held_out] + train_values = [float(pair[delta_key][feature]) for pair in train] + train_labels = [int(pair["full_action_efficacy"]) for pair in train] + threshold, direction, train_balanced = _fit_threshold( + train_values, train_labels + ) + test_values = [float(pair[delta_key][feature]) for pair in test] + test_predictions = [ + int(direction * (value - threshold) >= 0.0) for value in test_values + ] + test_labels = [int(pair["full_action_efficacy"]) for pair in test] + predictions.extend(test_predictions) + held_out_labels.extend(test_labels) + folds.append( + { + "held_out_replicate": held_out, + "threshold": threshold, + "direction": direction, + "train_balanced_accuracy": train_balanced, + "test_labels": test_labels, + "test_predictions": test_predictions, + } + ) + balanced = _balanced_accuracy(held_out_labels, predictions) + accuracy = sum( + prediction == label + for prediction, label in zip(predictions, held_out_labels, strict=True) + ) / len(held_out_labels) + results[feature] = { + "balanced_accuracy": balanced, + "accuracy": accuracy, + "folds": folds, + } + best_feature = max( + results, + key=lambda feature: ( + results[feature]["balanced_accuracy"], + results[feature]["accuracy"], + feature, + ), + ) + return { + "labels": V0.numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + "features": results, + "best_feature": best_feature, + "best_balanced_accuracy": results[best_feature]["balanced_accuracy"], + "best_accuracy": results[best_feature]["accuracy"], + } + + +def analyze_horizon(trials: list[dict[str, Any]], horizon_s: float) -> dict[str, Any]: + actions, repeats = build_pairs(trials) + response = V0.response_statistics(actions, repeats) + qualifying_response = sorted( + feature for feature, item in response.items() if item["qualifies"] + ) + outcome_cv = one_feature_leave_repeat_out( + actions, + delta_key="delta_outcome", + features=OUTCOME_FEATURES, + ) + telemetry_cv = one_feature_leave_repeat_out( + actions, + delta_key="delta_state", + features=V0.GATE_FEATURES, + ) + outcome_best = float(outcome_cv["best_balanced_accuracy"]) + efficacy_qualifying = sorted( + feature + for feature, item in telemetry_cv["features"].items() + if item["balanced_accuracy"] >= MIN_EFFICACY_BALANCED_ACCURACY + and item["balanced_accuracy"] + >= outcome_best + MIN_EFFICACY_DELTA_OVER_OUTCOME + ) + action_hashes_match = all( + pair["group"]["request_hash"] for pair in actions + ) + labels = [int(pair["full_action_efficacy"]) for pair in actions] + invariants = { + "expected_action_pair_count": len(actions) == EXPECTED_ACTION_PAIRS, + "expected_repeat_pair_count": len(repeats) == EXPECTED_REPEAT_PAIRS, + "matched_action_request_hashes": action_hashes_match, + "efficacy_label_balance": ( + sum(labels) >= MIN_EFFICACY_CLASS + and len(labels) - sum(labels) >= MIN_EFFICACY_CLASS + ), + "finite_deltas": all( + math.isfinite(value) + for pair in [*actions, *repeats] + for values in (pair["delta_state"], pair["delta_outcome"]) + for value in values.values() + ), + "probabilities_bounded": all( + 0.0 <= trial["outcome"][feature] <= 1.0 + for trial in trials + for feature in ( + "admitted_fraction", + "completed_over_admitted", + "completed_pass_rate", + "completed_fail_fraction_of_total", + "outstanding_over_admitted", + "admitted_input_tokens_mean_over_limit", + ) + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + transitions = defaultdict(int) + for pair in actions: + transitions[pair["full_feasibility_transition"]] += 1 + return { + "horizon_s": horizon_s, + "actions": actions, + "repeats": repeats, + "response_statistics": response, + "qualifying_response_features": qualifying_response, + "efficacy": { + "outcome_delta": outcome_cv, + "telemetry_delta": telemetry_cv, + "telemetry_qualifying_features": efficacy_qualifying, + "minimum_balanced_accuracy": MIN_EFFICACY_BALANCED_ACCURACY, + "minimum_delta_over_best_outcome": MIN_EFFICACY_DELTA_OVER_OUTCOME, + "feasibility_transitions": dict(sorted(transitions.items())), + }, + "sanity": { + "trials": len(trials), + "action_pairs": len(actions), + "repeat_pairs": len(repeats), + "invariants": invariants, + "red_flags": red_flags, + }, + } + + +def audit(*, run_root: Path, manifest_path: Path, output_path: Path) -> dict[str, Any]: + trials_by_horizon, streams = load_trials(run_root) + manifest_validation = validate_manifest( + trials_by_horizon[min(trials_by_horizon)], manifest_path + ) + horizons = { + str(int(horizon)): analyze_horizon(trials, horizon) + for horizon, trials in sorted(trials_by_horizon.items()) + } + red_flags = sorted( + { + flag + for horizon in horizons.values() + for flag in horizon["sanity"]["red_flags"] + } + ) + stable_response = sorted( + set.intersection( + *( + set(horizon["qualifying_response_features"]) + for horizon in horizons.values() + ) + ) + ) + stable_efficacy = sorted( + set.intersection( + *( + set(horizon["efficacy"]["telemetry_qualifying_features"]) + for horizon in horizons.values() + ) + ) + ) + if red_flags: + decision = "STOP_DATA_INVALID" + elif len(stable_response) < V0.MIN_STABLE_FEATURES: + decision = "STOP_NO_PROSPECTIVE_RESPONSE" + elif not stable_efficacy: + decision = "STOP_NO_INCREMENTAL_TUNING_SIGNAL" + else: + decision = "OPEN_MATCHED_GPU_PILOT" + payload = { + "schema": SCHEMA, + "status": "COMPLETE", + "decision": decision, + "claim_boundary": ( + "Development-only confirmation on an already-consumed P1 task. " + "Passing can open a newly registered matched pilot but cannot be " + "reported as held-out tuning evidence." + ), + "frozen_gate": { + "response_thresholds_identical_to_phase6_v0": True, + "expected_action_pairs": EXPECTED_ACTION_PAIRS, + "expected_repeat_pairs": EXPECTED_REPEAT_PAIRS, + "minimum_stable_response_features": V0.MIN_STABLE_FEATURES, + "minimum_efficacy_class": MIN_EFFICACY_CLASS, + "minimum_efficacy_balanced_accuracy": MIN_EFFICACY_BALANCED_ACCURACY, + "minimum_efficacy_delta_over_best_outcome": ( + MIN_EFFICACY_DELTA_OVER_OUTCOME + ), + }, + "stable_response_features": stable_response, + "stable_incremental_efficacy_features": stable_efficacy, + "horizons": horizons, + "provenance": { + "analysis_script": str(Path(__file__).resolve()), + "analysis_script_sha256": sha256_file(Path(__file__).resolve()), + "phase6_v0_script_sha256": sha256_file(HERE / "analyze_phase6.py"), + "run_root": str(run_root.resolve()), + "manifest": str(manifest_path.resolve()), + "manifest_sha256": sha256_file(manifest_path), + "manifest_validation": manifest_validation, + "streams": streams, + }, + "sanity": { + "stream_count": len(streams), + "stream_bytes": V0.numeric(item["bytes"] for item in streams), + "red_flags": red_flags, + }, + } + output_path.parent.mkdir(parents=True, exist_ok=True) + output_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = audit( + run_root=args.run_root, + manifest_path=args.manifest, + output_path=args.output, + ) + print( + json.dumps( + { + "decision": payload["decision"], + "stable_response_features": payload["stable_response_features"], + "stable_incremental_efficacy_features": payload[ + "stable_incremental_efficacy_features" + ], + "sanity": payload["sanity"], + }, + indent=2, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v0/analyze_phase6.py b/runs/intervention-response-v0/analyze_phase6.py new file mode 100644 index 0000000..900bdb5 --- /dev/null +++ b/runs/intervention-response-v0/analyze_phase6.py @@ -0,0 +1,520 @@ +#!/usr/bin/env python3 +"""Audit whether a controlled knob change produces identifiable telemetry deltas. + +This is a development-only feasibility audit. It compares adjacent MNS +interventions at an identical TP, offered-load anchor, and request sequence +against same-config primary/confirmation repeat noise. It does not claim that +the observed response is causal or that it improves an end-to-end tuner. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import sys +from collections import defaultdict +from pathlib import Path +from statistics import median +from typing import Any, Iterable, Mapping + + +HERE = Path(__file__).resolve().parent +COMMON_STATE_DIR = HERE.parent / "telemetry-residual" +sys.path.insert(0, str(COMMON_STATE_DIR)) + +from common_state import load_jsonl, summarize_engine # noqa: E402 + + +SCHEMA = "intervention-response-audit-v0" +HORIZONS_S = (5.0, 10.0) +GATE_FEATURES = ( + "scheduler_steps_per_s", + "decode_batch_size.mean", + "prefill_token_fraction", + "queue_waiting_mean", + "queue_running_mean", + "kv_usage_mean", + "graph_padding_fraction", +) +ALL_FEATURES = ( + "scheduler_steps_per_s", + "batch_size.mean", + "batch_tokens.mean", + "decode_batch_size.mean", + "prefill_token_fraction", + "queue_waiting_mean", + "queue_running_mean", + "preemptions", + "kv_usage_mean", + "kv_usage_max", + "kv_usage_end_minus_start", + "graph_none_share", + "graph_full_share", + "graph_padding_fraction", +) +EXPECTED_ACTION_PAIRS = 17 +MIN_REPEAT_PAIRS = 20 +MIN_STABLE_FEATURES = 2 +MIN_SIGN_CONSISTENCY = 0.75 +MIN_EFFECT_TO_NOISE = 2.0 +MIN_ABOVE_NOISE_P95_FRACTION = 0.5 + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def numeric(values: Iterable[float]) -> dict[str, Any]: + finite = [float(value) for value in values] + if not finite: + raise ValueError("numeric summary requires at least one value") + if any(not math.isfinite(value) for value in finite): + raise ValueError("numeric summary received a non-finite value") + return { + "n": len(finite), + "min": min(finite), + "max": max(finite), + "distinct_n": len(set(finite)), + } + + +def quantile(values: Iterable[float], probability: float) -> float: + ordered = sorted(float(value) for value in values) + if not ordered: + raise ValueError("quantile requires at least one value") + if not 0.0 <= probability <= 1.0: + raise ValueError("quantile probability must be in [0, 1]") + position = probability * (len(ordered) - 1) + lower = math.floor(position) + upper = math.ceil(position) + if lower == upper: + return ordered[lower] + weight = position - lower + return ordered[lower] * (1.0 - weight) + ordered[upper] * weight + + +def flatten_state(summary: Mapping[str, Any]) -> dict[str, float]: + common = summary["common"] + engine = summary["engine_only"] + state = { + "scheduler_steps_per_s": float(common["scheduler_steps_per_s"]), + "batch_size.mean": float(common["batch_size"]["mean"]), + "batch_tokens.mean": float(common["batch_tokens"]["mean"]), + "decode_batch_size.mean": float(common["decode_batch_size"]["mean"]), + "prefill_token_fraction": float(common["prefill_token_fraction"]), + "queue_waiting_mean": float(common["queue_waiting_mean"]), + "queue_running_mean": float(common["queue_running_mean"]), + "preemptions": float(common["preemptions"]), + "kv_usage_mean": float(engine["kv_usage_mean"]), + "kv_usage_max": float(engine["kv_usage_max"]), + "kv_usage_end_minus_start": float(engine["kv_usage_end_minus_start"]), + "graph_none_share": float(engine["graph_none_share"]), + "graph_full_share": float(engine["graph_full_share"]), + "graph_padding_fraction": float(engine["graph_padding_fraction"]), + } + if set(state) != set(ALL_FEATURES): + raise ValueError("flattened state does not match the frozen feature set") + if any(not math.isfinite(value) for value in state.values()): + raise ValueError("flattened state contains a non-finite value") + return state + + +def _trial_role(path: Path) -> str: + return "confirmation" if path.parent.name.startswith("confirm-") else "primary" + + +def load_trials( + raw_root: Path, + *, + horizons_s: tuple[float, ...] = HORIZONS_S, +) -> tuple[dict[float, list[dict[str, Any]]], list[dict[str, Any]]]: + by_horizon = {horizon: [] for horizon in horizons_s} + stream_provenance = [] + for cell_dir in sorted(path for path in raw_root.iterdir() if path.is_dir()): + streams = sorted((cell_dir / "opprof").glob("*.jsonl")) + if len(streams) != 1: + raise ValueError(f"{cell_dir}: expected exactly one Layer-1 stream") + stream = streams[0] + records = load_jsonl(stream) + stream_provenance.append( + { + "path": str(stream), + "sha256": sha256_file(stream), + "bytes": stream.stat().st_size, + } + ) + result_paths = sorted(cell_dir.glob("anchor-*/result.json")) + result_paths.extend(sorted(cell_dir.glob("confirm-*-anchor-*/result.json"))) + for result_path in result_paths: + result = json.loads(result_path.read_text(encoding="utf-8")) + start_ns = int(result["interval"]["start_mono_ns"]) + elapsed_s = float(result["interval"]["elapsed_s"]) + for horizon_s in horizons_s: + if elapsed_s < horizon_s: + raise ValueError( + f"{result_path}: elapsed {elapsed_s} is shorter than {horizon_s}s" + ) + state = flatten_state( + summarize_engine( + records, + start_ns=start_ns, + end_ns=start_ns + int(horizon_s * 1e9), + request_count=int(result["selection"]["count"]), + ) + ) + by_horizon[horizon_s].append( + { + "trial_id": str(result_path.relative_to(raw_root)), + "result_sha256": sha256_file(result_path), + "role": _trial_role(result_path), + "cell": str(result["cell"]), + "study_sha256": str(result["study_sha256"]), + "tp": int(result["tp"]), + "mns": int(result["mns"]), + "anchor": float(result["anchor"]), + "request_hash": str( + result["selection"]["request_id_order_sha256"] + ), + "request_count": int(result["selection"]["count"]), + "early_stopped": bool(result["early_stopped"]), + "full_pass_rate": float(result["pass_rate"]), + "full_feasible": bool(result["feasible"]), + "state": state, + } + ) + return by_horizon, stream_provenance + + +def _group_key(trial: Mapping[str, Any]) -> tuple[Any, ...]: + return ( + trial["study_sha256"], + trial["tp"], + trial["anchor"], + trial["request_hash"], + ) + + +def _delta(source: Mapping[str, Any], target: Mapping[str, Any]) -> dict[str, float]: + return { + feature: float(target["state"][feature]) - float(source["state"][feature]) + for feature in ALL_FEATURES + } + + +def _pair(source: Mapping[str, Any], target: Mapping[str, Any], kind: str) -> dict[str, Any]: + if _group_key(source) != _group_key(target): + raise ValueError("pair endpoints do not share workload identity") + return { + "kind": kind, + "group": { + "study_sha256": source["study_sha256"], + "tp": source["tp"], + "anchor": source["anchor"], + "request_hash": source["request_hash"], + }, + "source": { + "trial_id": source["trial_id"], + "cell": source["cell"], + "mns": source["mns"], + "early_stopped": source["early_stopped"], + "full_pass_rate": source["full_pass_rate"], + "full_feasible": source["full_feasible"], + }, + "target": { + "trial_id": target["trial_id"], + "cell": target["cell"], + "mns": target["mns"], + "early_stopped": target["early_stopped"], + "full_pass_rate": target["full_pass_rate"], + "full_feasible": target["full_feasible"], + }, + "delta_state": _delta(source, target), + "descriptive_full_outcome": { + "delta_pass_rate": target["full_pass_rate"] - source["full_pass_rate"], + "feasibility_transition": ( + f"{str(source['full_feasible']).lower()}->" + f"{str(target['full_feasible']).lower()}" + ), + }, + } + + +def build_pairs( + trials: list[dict[str, Any]], +) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: + primary_groups: dict[tuple[Any, ...], list[dict[str, Any]]] = defaultdict(list) + primary_by_cell_anchor: dict[tuple[Any, ...], dict[str, Any]] = {} + confirmations = [] + for trial in trials: + if trial["role"] == "primary": + primary_groups[_group_key(trial)].append(trial) + primary_by_cell_anchor[ + (trial["cell"], trial["anchor"], trial["request_hash"]) + ] = trial + else: + confirmations.append(trial) + + actions = [] + for group in primary_groups.values(): + ordered = sorted(group, key=lambda item: item["mns"]) + for source, target in zip(ordered, ordered[1:], strict=False): + if target["mns"] == source["mns"] * 2: + actions.append(_pair(source, target, "mns_increase")) + + repeats = [] + for confirmation in confirmations: + key = ( + confirmation["cell"], + confirmation["anchor"], + confirmation["request_hash"], + ) + primary = primary_by_cell_anchor.get(key) + if primary is None: + raise ValueError(f"{confirmation['trial_id']}: missing matched primary") + if primary["mns"] != confirmation["mns"]: + raise ValueError("repeat endpoints changed MNS") + repeats.append(_pair(primary, confirmation, "same_config_repeat")) + return actions, repeats + + +def response_statistics( + actions: list[dict[str, Any]], + repeats: list[dict[str, Any]], +) -> dict[str, Any]: + statistics = {} + for feature in ALL_FEATURES: + action = [float(pair["delta_state"][feature]) for pair in actions] + noise = [float(pair["delta_state"][feature]) for pair in repeats] + action_abs = [abs(value) for value in action] + noise_abs = [abs(value) for value in noise] + positive = sum(value > 1e-12 for value in action) + negative = sum(value < -1e-12 for value in action) + zero = len(action) - positive - negative + nonzero = positive + negative + sign_consistency = max(positive, negative) / nonzero if nonzero else 0.0 + action_median = median(action_abs) + noise_median = median(noise_abs) + noise_p95 = quantile(noise_abs, 0.95) + effect_to_noise = ( + action_median / noise_median + if noise_median > 0 + else (math.inf if action_median > 0 else 0.0) + ) + above_noise = sum(value > noise_p95 for value in action_abs) / len(action_abs) + qualifies = ( + feature in GATE_FEATURES + and sign_consistency >= MIN_SIGN_CONSISTENCY + and effect_to_noise >= MIN_EFFECT_TO_NOISE + and above_noise >= MIN_ABOVE_NOISE_P95_FRACTION + ) + statistics[feature] = { + "action_delta": numeric(action), + "repeat_delta": numeric(noise), + "action_abs_median": action_median, + "repeat_abs_median": noise_median, + "repeat_abs_p95": noise_p95, + "effect_to_repeat_median": ( + effect_to_noise if math.isfinite(effect_to_noise) else None + ), + "effect_to_repeat_median_is_infinite": math.isinf(effect_to_noise), + "action_signs": { + "positive": positive, + "negative": negative, + "zero": zero, + "consistency": sign_consistency, + }, + "action_above_repeat_p95_fraction": above_noise, + "gate_feature": feature in GATE_FEATURES, + "qualifies": qualifies, + } + return statistics + + +def analyze_horizon(trials: list[dict[str, Any]], horizon_s: float) -> dict[str, Any]: + actions, repeats = build_pairs(trials) + feature_statistics = response_statistics(actions, repeats) + qualifying = sorted( + feature for feature, item in feature_statistics.items() if item["qualifies"] + ) + all_values = [ + value + for trial in trials + for value in trial["state"].values() + ] + action_vectors = { + tuple(round(float(pair["delta_state"][feature]), 12) for feature in ALL_FEATURES) + for pair in actions + } + pair_invariants = { + "expected_action_pair_count": len(actions) == EXPECTED_ACTION_PAIRS, + "sufficient_repeat_pair_count": len(repeats) >= MIN_REPEAT_PAIRS, + "all_pair_hashes_match": all( + pair["group"]["request_hash"] for pair in [*actions, *repeats] + ), + "all_values_finite": all(math.isfinite(value) for value in all_values), + "state_vectors_not_all_identical": len(action_vectors) > 1, + "ratios_bounded": all( + 0.0 <= trial["state"][feature] <= 1.0 + for trial in trials + for feature in ( + "prefill_token_fraction", + "kv_usage_mean", + "kv_usage_max", + "graph_none_share", + "graph_full_share", + "graph_padding_fraction", + ) + ), + "nonnegative_counters": all( + trial["state"][feature] >= 0.0 + for trial in trials + for feature in ( + "scheduler_steps_per_s", + "batch_size.mean", + "batch_tokens.mean", + "decode_batch_size.mean", + "queue_waiting_mean", + "queue_running_mean", + "preemptions", + ) + ), + } + red_flags = [name for name, passed in pair_invariants.items() if not passed] + pass_deltas = [ + pair["descriptive_full_outcome"]["delta_pass_rate"] for pair in actions + ] + transitions = defaultdict(int) + for pair in actions: + transitions[pair["descriptive_full_outcome"]["feasibility_transition"]] += 1 + return { + "horizon_s": horizon_s, + "actions": actions, + "repeats": repeats, + "feature_statistics": feature_statistics, + "qualifying_features": qualifying, + "descriptive_full_outcome": { + "delta_pass_rate": numeric(pass_deltas), + "positive": sum(value > 1e-12 for value in pass_deltas), + "negative": sum(value < -1e-12 for value in pass_deltas), + "zero": sum(abs(value) <= 1e-12 for value in pass_deltas), + "feasibility_transitions": dict(sorted(transitions.items())), + "limitation": ( + "Full outcomes may use different elapsed durations when a trial " + "early-stopped; they are descriptive and are not a gate input." + ), + }, + "sanity": { + "trials": len(trials), + "action_pairs": len(actions), + "repeat_pairs": len(repeats), + "distinct_action_vectors": len(action_vectors), + "invariants": pair_invariants, + "red_flags": red_flags, + }, + } + + +def audit( + *, + metrics_path: Path, + raw_root: Path, + output_path: Path, +) -> dict[str, Any]: + trials_by_horizon, streams = load_trials(raw_root) + horizons = { + str(int(horizon)): analyze_horizon(trials, horizon) + for horizon, trials in sorted(trials_by_horizon.items()) + } + red_flags = sorted( + { + red_flag + for horizon in horizons.values() + for red_flag in horizon["sanity"]["red_flags"] + } + ) + stable_features = sorted( + set.intersection( + *(set(horizon["qualifying_features"]) for horizon in horizons.values()) + ) + ) + if red_flags: + decision = "STOP_DATA_INVALID" + elif len(stable_features) < MIN_STABLE_FEATURES: + decision = "STOP_NO_IDENTIFIABLE_RESPONSE" + else: + decision = "OPEN_MATCHED_PILOT" + payload = { + "schema": SCHEMA, + "status": "COMPLETE", + "decision": decision, + "claim_boundary": ( + "Development-only identifiability gate. Passing opens a controlled " + "real-GPU pilot; it does not establish tuning benefit or causality." + ), + "frozen_gate": { + "horizons_s": list(HORIZONS_S), + "expected_action_pairs": EXPECTED_ACTION_PAIRS, + "minimum_repeat_pairs": MIN_REPEAT_PAIRS, + "minimum_stable_features": MIN_STABLE_FEATURES, + "minimum_sign_consistency": MIN_SIGN_CONSISTENCY, + "minimum_effect_to_repeat_median": MIN_EFFECT_TO_NOISE, + "minimum_action_above_repeat_p95_fraction": ( + MIN_ABOVE_NOISE_P95_FRACTION + ), + "gate_features": list(GATE_FEATURES), + }, + "stable_qualifying_features": stable_features, + "horizons": horizons, + "provenance": { + "analysis_script": str(Path(__file__).resolve()), + "analysis_script_sha256": sha256_file(Path(__file__).resolve()), + "phase6_metrics": str(metrics_path.resolve()), + "phase6_metrics_sha256": sha256_file(metrics_path), + "raw_root": str(raw_root.resolve()), + "streams": streams, + }, + "sanity": { + "stream_count": len(streams), + "stream_bytes": numeric(item["bytes"] for item in streams), + "red_flags": red_flags, + }, + } + output_path.parent.mkdir(parents=True, exist_ok=True) + output_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--metrics", type=Path, required=True) + parser.add_argument("--raw-root", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = audit( + metrics_path=args.metrics, + raw_root=args.raw_root, + output_path=args.output, + ) + print( + json.dumps( + { + "decision": payload["decision"], + "stable_qualifying_features": payload[ + "stable_qualifying_features" + ], + "sanity": payload["sanity"], + }, + indent=2, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v0/p1-audit.json b/runs/intervention-response-v0/p1-audit.json new file mode 100644 index 0000000..7164d08 --- /dev/null +++ b/runs/intervention-response-v0/p1-audit.json @@ -0,0 +1,6818 @@ +{ + "claim_boundary": "Development-only confirmation on an already-consumed P1 task. Passing can open a newly registered matched pilot but cannot be reported as held-out tuning evidence.", + "decision": "STOP_NO_PROSPECTIVE_RESPONSE", + "frozen_gate": { + "expected_action_pairs": 12, + "expected_repeat_pairs": 24, + "minimum_efficacy_balanced_accuracy": 0.75, + "minimum_efficacy_class": 4, + "minimum_efficacy_delta_over_best_outcome": 0.15, + "minimum_stable_response_features": 2, + "response_thresholds_identical_to_phase6_v0": true + }, + "horizons": { + "10": { + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.04424092126093565, + "tpot_mean_over_slo": 0.008821253352839065, + "ttft_max_over_slo_max": 0.001164793997304514, + "ttft_mean_over_slo_max": -0.0007446444380963049 + }, + "delta_state": { + "batch_size.mean": 0.028437757317164802, + "batch_tokens.mean": -0.16003443371509718, + "decode_batch_size.mean": 0.02853132719514928, + "graph_full_share": -0.0012762931357137575, + "graph_none_share": -0.0069204281757616545, + "graph_padding_fraction": 0.006029111786335234, + "kv_usage_end_minus_start": 0.00036424504378218536, + "kv_usage_max": 0.009254174352690048, + "kv_usage_mean": 0.00277340001356393, + "preemptions": 0.0, + "prefill_token_fraction": -0.0003954091503666213, + "queue_running_mean": 0.21264213909999974, + "queue_waiting_mean": -0.073977916, + "scheduler_steps_per_s": 0.20000000000000284 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 1, + "request_hash": "3f0ab9be34dbee884f5dbe2135b983d31a47578a43f606a5f390b0528dbedbe3", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.20670391061452514, + "mns": 8, + "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", + "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", + "trial_id": "cells/tp1_mns8/high-rep1/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.24581005586592178, + "mns": 64, + "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", + "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", + "trial_id": "cells/tp1_mns64/high-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.032258064516129004, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.03225806451612906, + "tpot_max_over_slo": 0.33050636819556467, + "tpot_mean_over_slo": 0.2980276034718437, + "ttft_max_over_slo_max": -0.14499169250484553, + "ttft_mean_over_slo_max": -0.08870032973580096 + }, + "delta_state": { + "batch_size.mean": 4.5469521321594915, + "batch_tokens.mean": 177.93791776288268, + "decode_batch_size.mean": 4.490206104102559, + "graph_full_share": -0.049141465319289424, + "graph_none_share": 0.04533918395046808, + "graph_padding_fraction": 0.004209922294522723, + "kv_usage_end_minus_start": 0.05537535994708087, + "kv_usage_max": 0.08922043547542868, + "kv_usage_mean": 0.04639671324585863, + "preemptions": 0.0, + "prefill_token_fraction": 0.003967110353963799, + "queue_running_mean": 4.6810500217, + "queue_waiting_mean": -2.3262808059, + "scheduler_steps_per_s": -14.599999999999998 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 2, + "request_hash": "b19c91876a8417fe06cb2775f65a1111b202f88ce10ec33576af3d84c059f02a", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7988826815642458, + "mns": 8, + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.88268156424581, + "mns": 64, + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.018804627718169264, + "tpot_mean_over_slo": 0.011575705913082518, + "ttft_max_over_slo_max": 0.0012841941691779934, + "ttft_mean_over_slo_max": -0.0028507176665395206 + }, + "delta_state": { + "batch_size.mean": 0.028484697519924218, + "batch_tokens.mean": 0.386581442994256, + "decode_batch_size.mean": 0.02831307423779572, + "graph_full_share": -0.00017162328212849687, + "graph_none_share": -0.003114247473621861, + "graph_padding_fraction": 0.003200143463807169, + "kv_usage_end_minus_start": 8.07298706054782e-05, + "kv_usage_max": 0.0034809583857736692, + "kv_usage_mean": 0.0010653256527363222, + "preemptions": 0.0, + "prefill_token_fraction": -0.00013286942159473902, + "queue_running_mean": 0.1416205103000001, + "queue_waiting_mean": -0.028138744, + "scheduler_steps_per_s": -0.6000000000000085 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 3, + "request_hash": "a85849f1f057a67c82523cff57a52af22d2eb9514a953e076990aa6c25c1ef7e", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6424581005586593, + "mns": 8, + "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", + "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", + "trial_id": "cells/tp1_mns8/high-rep3/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.9273743016759777, + "mns": 64, + "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", + "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", + "trial_id": "cells/tp1_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.0040397713397167245, + "tpot_mean_over_slo": -0.007906587129236364, + "ttft_max_over_slo_max": -0.00039426567188155515, + "ttft_mean_over_slo_max": -0.0011763758957386017 + }, + "delta_state": { + "batch_size.mean": -0.08160779537149798, + "batch_tokens.mean": -1.7609001853017006, + "decode_batch_size.mean": -0.08094394190416265, + "graph_full_share": 0.0006306607939687847, + "graph_none_share": -0.005369997286859742, + "graph_padding_fraction": 0.0002696603812660705, + "kv_usage_end_minus_start": 9.361443848254769e-05, + "kv_usage_max": 0.0003780815386457137, + "kv_usage_mean": 2.4214752986723154e-05, + "preemptions": 0.0, + "prefill_token_fraction": -6.638371684375421e-05, + "queue_running_mean": -0.0994468485000004, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.3000000000000114 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 1, + "request_hash": "c8ea9278f180ebb23cee595e167627cbed711032e70c9ca5ddc7205320cd75c0", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", + "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", + "trial_id": "cells/tp1_mns8/low-rep1/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", + "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", + "trial_id": "cells/tp1_mns64/low-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.08836641039618653, + "tpot_mean_over_slo": 0.05249869848340616, + "ttft_max_over_slo_max": -0.03703611367382109, + "ttft_mean_over_slo_max": -0.011082230367607901 + }, + "delta_state": { + "batch_size.mean": 0.2702547536010833, + "batch_tokens.mean": 7.3674632893748395, + "decode_batch_size.mean": 0.2679754243751127, + "graph_full_share": -0.000425589040939367, + "graph_none_share": -0.003064812355221179, + "graph_padding_fraction": 0.008784696835623326, + "kv_usage_end_minus_start": 2.2346672411965862e-05, + "kv_usage_max": 0.018441793350708946, + "kv_usage_mean": 0.009225433346662204, + "preemptions": 0.0, + "prefill_token_fraction": 2.292768411660262e-05, + "queue_running_mean": 0.7380613438000001, + "queue_waiting_mean": -0.2325565183, + "scheduler_steps_per_s": -3.8000000000000043 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 2, + "request_hash": "1f27fe31ca32fd67f2f70a54eee83146329a7838ff2a33be48ce4372640a858c", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.0020900686586818318, + "tpot_mean_over_slo": -0.0010018654846283281, + "ttft_max_over_slo_max": 0.0005588701654535982, + "ttft_mean_over_slo_max": 6.347352852269228e-05 + }, + "delta_state": { + "batch_size.mean": -0.012547944623296114, + "batch_tokens.mean": -0.15241343405688212, + "decode_batch_size.mean": -0.012475695661875008, + "graph_full_share": 7.224896142110637e-05, + "graph_none_share": -0.0058946652641868255, + "graph_padding_fraction": 0.00029565151617282764, + "kv_usage_end_minus_start": 0.00011133071931368477, + "kv_usage_max": 0.0002683613903166071, + "kv_usage_mean": -1.5294343296619328e-05, + "preemptions": 0.0, + "prefill_token_fraction": 9.183515573107925e-05, + "queue_running_mean": -0.008641511400000024, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.20000000000000284 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 3, + "request_hash": "a46c62ebd8b7df5534831feb70aed45d23b1a5e7f8bf3d4cd82a79f182fbc69e", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", + "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", + "trial_id": "cells/tp1_mns8/low-rep3/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", + "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", + "trial_id": "cells/tp1_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.008, + "completed_over_admitted": -0.1610169491525424, + "completed_pass_rate": 0.08823529411764708, + "outstanding_over_admitted": 0.1610169491525424, + "tpot_max_over_slo": 0.29320454629189446, + "tpot_mean_over_slo": 0.2535434351882483, + "ttft_max_over_slo_max": -0.2690165124949999, + "ttft_mean_over_slo_max": -0.13813761862324808 + }, + "delta_state": { + "batch_size.mean": 10.093788314560394, + "batch_tokens.mean": 434.02936828958076, + "decode_batch_size.mean": 9.924284699384675, + "graph_full_share": -0.09005855324219958, + "graph_none_share": 0.07778357071650377, + "graph_padding_fraction": 0.0017247330951379231, + "kv_usage_end_minus_start": 0.05233364294717768, + "kv_usage_max": 0.049202174029810086, + "kv_usage_mean": 0.019156054903148284, + "preemptions": 0.0, + "prefill_token_fraction": 0.013365962658000052, + "queue_running_mean": 20.948292951200003, + "queue_waiting_mean": -13.1590807183, + "scheduler_steps_per_s": -34.5 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 1, + "request_hash": "5240182174ec5e25247be01c017cbefe2f1e59426bb24a57746dc40be4a82208", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.10666666666666667, + "mns": 16, + "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", + "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", + "trial_id": "cells/tp4_mns16/high-rep1/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", + "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", + "trial_id": "cells/tp4_mns64/high-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.012, + "completed_over_admitted": -0.1578947368421053, + "completed_pass_rate": 0.13636363636363635, + "outstanding_over_admitted": 0.1578947368421053, + "tpot_max_over_slo": 0.4737703299230537, + "tpot_mean_over_slo": 0.3241796650751651, + "ttft_max_over_slo_max": -0.3133456981643879, + "ttft_mean_over_slo_max": -0.1578698176905678 + }, + "delta_state": { + "batch_size.mean": 19.42960244648318, + "batch_tokens.mean": 656.7778499176663, + "decode_batch_size.mean": 19.184206069160197, + "graph_full_share": -0.13147494707127738, + "graph_none_share": 0.11760056457304163, + "graph_padding_fraction": 0.0010721073978083185, + "kv_usage_end_minus_start": 0.04855346265723404, + "kv_usage_max": 0.042070962023699754, + "kv_usage_mean": 0.022662705875080867, + "preemptions": 0.0, + "prefill_token_fraction": 0.007938598543200115, + "queue_running_mean": 24.0049637595, + "queue_waiting_mean": -15.5760687653, + "scheduler_steps_per_s": -32.900000000000006 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 2, + "request_hash": "51be592ba9a59c8d66eb1bcc70c3d2b011c87c66128aff36574973b40389ad96", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.092, + "mns": 16, + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.12844036697247707, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.12844036697247707, + "tpot_max_over_slo": 0.2467615478745419, + "tpot_mean_over_slo": 0.024430165887946775, + "ttft_max_over_slo_max": -0.08022667251255673, + "ttft_mean_over_slo_max": -0.04385035177332515 + }, + "delta_state": { + "batch_size.mean": 2.1439298581773514, + "batch_tokens.mean": 116.96904413675719, + "decode_batch_size.mean": 2.0885971125866956, + "graph_full_share": -0.028466450255056452, + "graph_none_share": 0.016098373855303175, + "graph_padding_fraction": 0.004743554546834421, + "kv_usage_end_minus_start": 0.0262231865121626, + "kv_usage_max": 0.020094834170883313, + "kv_usage_mean": 0.005378376045436206, + "preemptions": 0.0, + "prefill_token_fraction": 0.005752175072525123, + "queue_running_mean": 4.9523955869, + "queue_waiting_mean": -4.365988725199999, + "scheduler_steps_per_s": -10.100000000000009 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 3, + "request_hash": "604e338cf576d187157293922dada2bceb311567268ddae74eb7a089debf0242", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.14666666666666667, + "mns": 16, + "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", + "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", + "trial_id": "cells/tp4_mns16/high-rep3/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", + "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", + "trial_id": "cells/tp4_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.03209297007177109, + "tpot_mean_over_slo": -0.01184639150074901, + "ttft_max_over_slo_max": -0.04185720401195188, + "ttft_mean_over_slo_max": -0.005481651257700743 + }, + "delta_state": { + "batch_size.mean": -0.9669213868166988, + "batch_tokens.mean": -22.564692860834526, + "decode_batch_size.mean": -0.9550554240524107, + "graph_full_share": 0.009900102186990445, + "graph_none_share": -0.015869663207727508, + "graph_padding_fraction": -0.0007445684074620391, + "kv_usage_end_minus_start": 5.549832249451825e-05, + "kv_usage_max": 0.004800695185764425, + "kv_usage_mean": -0.0003621004636635179, + "preemptions": 0.0, + "prefill_token_fraction": -0.00013559912520566542, + "queue_running_mean": -0.46276202019999957, + "queue_waiting_mean": -0.0902122614, + "scheduler_steps_per_s": 11.900000000000006 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "6fb170f366b68a4e4e0fc976d370cb1f5b7b915d4dc22ebe5db3c6b9615afe68", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6196078431372549, + "mns": 16, + "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", + "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", + "trial_id": "cells/tp4_mns16/low-rep1/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", + "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", + "trial_id": "cells/tp4_mns64/low-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.05128205128205132, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.051282051282051294, + "tpot_max_over_slo": 0.03807669731890995, + "tpot_mean_over_slo": -0.014506958738293885, + "ttft_max_over_slo_max": -0.12963614600206105, + "ttft_mean_over_slo_max": -0.048670619592321884 + }, + "delta_state": { + "batch_size.mean": -1.4718645603306726, + "batch_tokens.mean": -31.86018426395526, + "decode_batch_size.mean": -1.4608669627184998, + "graph_full_share": 0.02356281394840931, + "graph_none_share": -0.027804605789997328, + "graph_padding_fraction": 0.0035831460549140947, + "kv_usage_end_minus_start": -0.002621229743826836, + "kv_usage_max": 0.0030741209471074615, + "kv_usage_mean": -8.774771567193577e-05, + "preemptions": 0.0, + "prefill_token_fraction": 0.0001064077287293852, + "queue_running_mean": -0.40493394709999997, + "queue_waiting_mean": -2.0223754996, + "scheduler_steps_per_s": 10.599999999999994 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "eb4c0445556d23d73d27d728b45e63fec44e17e84a4f2f341787108b7f7a4901", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6803921568627451, + "mns": 16, + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.04088999417529798, + "tpot_mean_over_slo": -0.0032529423795618906, + "ttft_max_over_slo_max": 0.0002604483355147097, + "ttft_mean_over_slo_max": -0.001620827191598239 + }, + "delta_state": { + "batch_size.mean": 0.14852247907791494, + "batch_tokens.mean": 11.553519812007579, + "decode_batch_size.mean": 0.14728805801238387, + "graph_full_share": 0.0005649503708670123, + "graph_none_share": -0.0023940560373101466, + "graph_padding_fraction": -0.00035714669425012356, + "kv_usage_end_minus_start": 0.0014602804352583654, + "kv_usage_max": 0.003991863602817669, + "kv_usage_mean": 4.656051020407101e-06, + "preemptions": 0.0, + "prefill_token_fraction": 0.0018350834191880327, + "queue_running_mean": 0.06592412640000056, + "queue_waiting_mean": -0.2173940436, + "scheduler_steps_per_s": 1.1000000000000085 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "a660c44d1f8b9c0ada5a19285fa579871bcfcae7c6fa71649546efda9aa3baef", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6294117647058823, + "mns": 16, + "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", + "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", + "trial_id": "cells/tp4_mns16/low-rep3/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", + "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", + "trial_id": "cells/tp4_mns64/low-rep3/result.json" + } + } + ], + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "outstanding_over_admitted", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.006, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.004, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.004, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.08034921574430304, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.06422018348623854, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.09507640067911716, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.06818181818181818, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.04411764705882354, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.04411764705882354, + "train_balanced_accuracy": 0.75 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.08034921574430307, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.06422018348623854, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.09507640067911718, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.028440662518539606, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.025448798894970176, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.014026599366027181, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": -0.0021274039320951094, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.018002935900514647, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.009876489314992687, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 0.00040965925048415397, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.00040965925048415397, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.039446658842886485, + "train_balanced_accuracy": 0.875 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.027466291070466525, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.0013986015436684204, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.0033290135767196725, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0878005661250898, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.08790969260376658, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.5179996829782867, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -3.074758903864796e-05, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -4.3743156492026536e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.0038965341747184087, + "train_balanced_accuracy": 0.875 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -5.152102948427755e-05, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.004075888029500068, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -3.176648134260631e-05, + "train_balanced_accuracy": 0.75 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 9.912144223023223e-05, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.000963459287459556, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 6.466770642299391e-05, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 4.816722804299999, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 2.5825188629999998, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.2521903978000002, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.1227663938, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.0820950887, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.0820950887, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": 0.6500000000000057, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -5.3500000000000085, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 6.450000000000003, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "horizon_s": 10.0, + "qualifying_response_features": [], + "repeats": [ + { + "delta_outcome": { + "admitted_fraction": 0.027932960893854747, + "admitted_input_tokens_mean_over_limit": 0.06722636258335918, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.21464019851116622, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.21464019851116628, + "tpot_max_over_slo": 0.3907463973292743, + "tpot_mean_over_slo": 0.45304077139972754, + "ttft_max_over_slo_max": -0.026048107499567152, + "ttft_mean_over_slo_max": 0.02040584175783646 + }, + "delta_state": { + "batch_size.mean": 8.048982941677576, + "batch_tokens.mean": 298.1121465228864, + "decode_batch_size.mean": 7.965265224708595, + "graph_full_share": -0.07611315423133669, + "graph_none_share": 0.08050759417399073, + "graph_padding_fraction": -0.006831997792023563, + "kv_usage_end_minus_start": 0.08232590668398432, + "kv_usage_max": 0.11409731241882304, + "kv_usage_mean": 0.08571139047521727, + "preemptions": 0.0, + "prefill_token_fraction": 0.010675483227336025, + "queue_running_mean": 7.2005321570000005, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -46.900000000000006 + }, + "group": { + "cell": "tp1_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", + "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", + "trial_id": "cells/tp1_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.03910614525139666, + "admitted_input_tokens_mean_over_limit": -0.15364353631132394, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.31720430107526887, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.31720430107526887, + "tpot_max_over_slo": -0.4318866070898938, + "tpot_mean_over_slo": -0.4648576303649653, + "ttft_max_over_slo_max": -0.0451326228309578, + "ttft_mean_over_slo_max": -0.03512127695333522 + }, + "delta_state": { + "batch_size.mean": -8.639628684110797, + "batch_tokens.mean": -331.31332381027744, + "decode_batch_size.mean": -8.548044927723337, + "graph_full_share": 0.08397919364981532, + "graph_none_share": -0.08346278303674427, + "graph_padding_fraction": 0.0049766447911436746, + "kv_usage_end_minus_start": -0.12798481366769898, + "kv_usage_max": -0.12688580277750028, + "kv_usage_mean": -0.09580544440945148, + "preemptions": 0.0, + "prefill_token_fraction": -0.024511024899494305, + "queue_running_mean": -7.6525189056, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 65.0 + }, + "group": { + "cell": "tp1_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", + "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", + "trial_id": "cells/tp1_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.008196721311475419, + "admitted_input_tokens_mean_over_limit": 0.07372581845238091, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.09285714285714286, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.09285714285714285, + "tpot_max_over_slo": 0.1702932155907418, + "tpot_mean_over_slo": 0.1739500633542023, + "ttft_max_over_slo_max": 0.02082575849878293, + "ttft_mean_over_slo_max": 0.029532855802237655 + }, + "delta_state": { + "batch_size.mean": 1.3717277486910993, + "batch_tokens.mean": 56.51418492510525, + "decode_batch_size.mean": 1.3587752164958689, + "graph_full_share": -0.010646964922293112, + "graph_none_share": 0.011895900019023516, + "graph_padding_fraction": 0.0086692890425959, + "kv_usage_end_minus_start": -0.017684084324108285, + "kv_usage_max": 0.05015486062543706, + "kv_usage_mean": 0.03571575463439401, + "preemptions": 0.0, + "prefill_token_fraction": 0.011636868678620127, + "queue_running_mean": 2.2687814585000003, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -27.10000000000001 + }, + "group": { + "cell": "tp1_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", + "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", + "trial_id": "cells/tp1_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.02459016393442623, + "admitted_input_tokens_mean_over_limit": -0.09311379825367644, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.24411764705882344, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.24411764705882355, + "tpot_max_over_slo": -0.3084416280309926, + "tpot_mean_over_slo": -0.21178917402407615, + "ttft_max_over_slo_max": -0.07642558032724386, + "ttft_mean_over_slo_max": -0.031511784150528625 + }, + "delta_state": { + "batch_size.mean": -1.5740567152122056, + "batch_tokens.mean": -51.36471646204379, + "decode_batch_size.mean": -1.5621527698392743, + "graph_full_share": 0.008413543976771298, + "graph_none_share": -0.010745558883376392, + "graph_padding_fraction": -0.007424342975302652, + "kv_usage_end_minus_start": 0.022080127884903655, + "kv_usage_max": -0.0773803576780897, + "kv_usage_mean": -0.04564071465536516, + "preemptions": 0.0, + "prefill_token_fraction": -0.004854783915929306, + "queue_running_mean": -3.3275810848000003, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 11.400000000000006 + }, + "group": { + "cell": "tp1_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", + "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", + "trial_id": "cells/tp1_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.027932960893854747, + "admitted_input_tokens_mean_over_limit": 0.06722636258335918, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.18238213399503722, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.18238213399503722, + "tpot_max_over_slo": 0.10448095039464528, + "tpot_mean_over_slo": 0.1638344212807229, + "ttft_max_over_slo_max": 0.1201083790025829, + "ttft_mean_over_slo_max": 0.10836152705554111 + }, + "delta_state": { + "batch_size.mean": 3.5304685668352485, + "batch_tokens.mean": 120.01419432628865, + "decode_batch_size.mean": 3.503590447801186, + "graph_full_share": -0.028247982047761022, + "graph_none_share": 0.028247982047760994, + "graph_padding_fraction": -0.005012808300211052, + "kv_usage_end_minus_start": 0.027314791780685632, + "kv_usage_max": 0.0341310512960844, + "kv_usage_mean": 0.042088077242922566, + "preemptions": 0.0, + "prefill_token_fraction": 0.006312963723005605, + "queue_running_mean": 2.7321242744000003, + "queue_waiting_mean": 2.2523028899, + "scheduler_steps_per_s": -32.1 + }, + "group": { + "cell": "tp1_mns8", + "level": "high", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", + "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", + "trial_id": "cells/tp1_mns8/high-rep1/result.json" + }, + "target": { + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.03910614525139666, + "admitted_input_tokens_mean_over_limit": -0.15364353631132394, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.28494623655913986, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.28494623655913975, + "tpot_max_over_slo": -0.12018486661249839, + "tpot_mean_over_slo": -0.17840573280620411, + "ttft_max_over_slo_max": -0.19140850950498134, + "ttft_mean_over_slo_max": -0.12097088902259666 + }, + "delta_state": { + "batch_size.mean": -4.121161249471228, + "batch_tokens.mean": -153.761987490389, + "decode_batch_size.mean": -4.086151897858574, + "graph_full_share": 0.035009351612654394, + "graph_none_share": -0.03500935161265434, + "graph_padding_fraction": 0.0059864236218592284, + "kv_usage_end_minus_start": -0.07269018359122359, + "kv_usage_max": -0.041146325687845264, + "kv_usage_mean": -0.050474056816329174, + "preemptions": 0.0, + "prefill_token_fraction": -0.020411045123935767, + "queue_running_mean": -3.1130893942, + "queue_waiting_mean": -2.2981420619, + "scheduler_steps_per_s": 51.00000000000001 + }, + "group": { + "cell": "tp1_mns8", + "level": "high", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + }, + "target": { + "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", + "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", + "trial_id": "cells/tp1_mns8/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.008196721311475419, + "admitted_input_tokens_mean_over_limit": 0.07372581845238091, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.09285714285714286, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.09285714285714285, + "tpot_max_over_slo": 0.07788703385483853, + "tpot_mean_over_slo": 0.11354477774155977, + "ttft_max_over_slo_max": 0.05746760650072247, + "ttft_mean_over_slo_max": 0.039438710274106954 + }, + "delta_state": { + "batch_size.mean": 1.019865199718518, + "batch_tokens.mean": 47.38582145042871, + "decode_batch_size.mean": 1.0098558502165935, + "graph_full_share": -0.00959071508738496, + "graph_none_share": 0.009590715087384953, + "graph_padding_fraction": 0.00015425258823864325, + "kv_usage_end_minus_start": -0.017612816558037703, + "kv_usage_max": 0.032091148813373827, + "kv_usage_mean": 0.02651453604071853, + "preemptions": 0.0, + "prefill_token_fraction": 0.01154755727765977, + "queue_running_mean": 1.4312732661999998, + "queue_waiting_mean": 0.2325565183, + "scheduler_steps_per_s": -20.999999999999993 + }, + "group": { + "cell": "tp1_mns8", + "level": "low", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", + "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", + "trial_id": "cells/tp1_mns8/low-rep1/result.json" + }, + "target": { + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.02459016393442623, + "admitted_input_tokens_mean_over_limit": -0.09311379825367644, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.24411764705882344, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.24411764705882355, + "tpot_max_over_slo": -0.21798514897612425, + "tpot_mean_over_slo": -0.15828861005604167, + "ttft_max_over_slo_max": -0.11402056416651855, + "ttft_mean_over_slo_max": -0.04265748804665922 + }, + "delta_state": { + "batch_size.mean": -1.2912540169878262, + "batch_tokens.mean": -43.844839738612066, + "decode_batch_size.mean": -1.2817016498022866, + "graph_full_share": 0.007915705974410825, + "graph_none_share": -0.007915705974410745, + "graph_padding_fraction": 0.0010647023441478472, + "kv_usage_end_minus_start": 0.021991143838001936, + "kv_usage_max": -0.05920692571769737, + "kv_usage_mean": -0.036399986965406333, + "preemptions": 0.0, + "prefill_token_fraction": -0.004923691387543783, + "queue_running_mean": -2.5808782296, + "queue_waiting_mean": -0.2325565183, + "scheduler_steps_per_s": 7.399999999999999 + }, + "group": { + "cell": "tp1_mns8", + "level": "low", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + }, + "target": { + "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", + "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", + "trial_id": "cells/tp1_mns8/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.014492753623188387, + "admitted_input_tokens_mean_over_limit": 0.04833232120973391, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.008403361344537785, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.00840336134453784, + "tpot_max_over_slo": 0.09841551118728337, + "tpot_mean_over_slo": 0.06345367037672023, + "ttft_max_over_slo_max": 0.010166719497647136, + "ttft_mean_over_slo_max": 0.006259302157370798 + }, + "delta_state": { + "batch_size.mean": 2.0365641486137616, + "batch_tokens.mean": 91.46332360534291, + "decode_batch_size.mean": 2.012753956673029, + "graph_full_share": -0.01772084409024577, + "graph_none_share": 0.021565271074194634, + "graph_padding_fraction": -0.002786564210742205, + "kv_usage_end_minus_start": 0.007622555602450509, + "kv_usage_max": 0.020987606952384663, + "kv_usage_mean": 0.007630049119681775, + "preemptions": 0.0, + "prefill_token_fraction": 0.005242973565105835, + "queue_running_mean": 0.8145702931999992, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -13.199999999999996 + }, + "group": { + "cell": "tp2_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "19537ec05f1e40543d850964db12234561954c209d55cd8cd0a1da716ab105fc", + "result_sha256": "49523b3269db649aa70072058666da3309ecabb93242cd504b6413f5de0f5d54", + "trial_id": "cells/tp2_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", + "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", + "trial_id": "cells/tp2_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.005797101449275366, + "admitted_input_tokens_mean_over_limit": -0.08122977688274602, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.10543840177580466, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.10543840177580466, + "tpot_max_over_slo": -0.08790813528521968, + "tpot_mean_over_slo": -0.0976607763237542, + "ttft_max_over_slo_max": -0.031135886325500906, + "ttft_mean_over_slo_max": -0.007596982260572371 + }, + "delta_state": { + "batch_size.mean": -2.9554801420404466, + "batch_tokens.mean": -131.8397196967789, + "decode_batch_size.mean": -2.9177065756667595, + "graph_full_share": 0.026547586067947715, + "graph_none_share": -0.024236910368448883, + "graph_padding_fraction": 0.005457793019402147, + "kv_usage_end_minus_start": -0.01758559388149228, + "kv_usage_max": -0.016281062553556214, + "kv_usage_mean": -0.00956401468061617, + "preemptions": 0.0, + "prefill_token_fraction": -0.00900387036264405, + "queue_running_mean": -2.3104728836999993, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 20.099999999999994 + }, + "group": { + "cell": "tp2_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", + "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", + "trial_id": "cells/tp2_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "ccba3d0a18a7db8501b040855bda473ee55c519619d4b9fa2a105bda06300829", + "result_sha256": "55447419925f30397d19863e149890241c58863d0adaafd511a67757905cb58e", + "trial_id": "cells/tp2_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.004255319148936149, + "admitted_input_tokens_mean_over_limit": 0.027061719804318485, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.002024291497975783, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0020242914979757, + "tpot_max_over_slo": 0.09163424472588544, + "tpot_mean_over_slo": 0.0001830762350235604, + "ttft_max_over_slo_max": 0.0006185861614843163, + "ttft_mean_over_slo_max": 0.0017585336856889636 + }, + "delta_state": { + "batch_size.mean": -0.4199832128448815, + "batch_tokens.mean": 6.494131535483177, + "decode_batch_size.mean": -0.4188308558741509, + "graph_full_share": 0.0010917066038501622, + "graph_none_share": 0.00011588552243251699, + "graph_padding_fraction": -0.0006875875614091621, + "kv_usage_end_minus_start": 0.009093350727084282, + "kv_usage_max": 0.011843098134008634, + "kv_usage_mean": 0.0004732259215823806, + "preemptions": 0.0, + "prefill_token_fraction": 0.005759524366025781, + "queue_running_mean": -0.24529797560000066, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.799999999999997 + }, + "group": { + "cell": "tp2_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "8685f475baf247f7ce52a6a2df1183cd77a6f580cd57a3a87c407b4e96c0895f", + "result_sha256": "7a8f4e462d496bd56b73b0a686653cbe499a238df62a8c9a50db687d051f1c68", + "trial_id": "cells/tp2_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", + "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", + "trial_id": "cells/tp2_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.008510638297872353, + "admitted_input_tokens_mean_over_limit": -0.08059763768960165, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.2543859649122807, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.2543859649122807, + "tpot_max_over_slo": -0.1787865552757903, + "tpot_mean_over_slo": -0.07545848129810456, + "ttft_max_over_slo_max": -0.024053132161498063, + "ttft_mean_over_slo_max": -0.008335181439306037 + }, + "delta_state": { + "batch_size.mean": -1.151692859112214, + "batch_tokens.mean": -39.27803694513372, + "decode_batch_size.mean": -1.1449765646539838, + "graph_full_share": 0.006457127102288407, + "graph_none_share": -0.008497380755445271, + "graph_padding_fraction": 6.114154911974513e-05, + "kv_usage_end_minus_start": 0.012533732366445394, + "kv_usage_max": -0.022906035375820144, + "kv_usage_mean": -0.00690787067830519, + "preemptions": 0.0, + "prefill_token_fraction": -0.0041771367400077475, + "queue_running_mean": -1.9580360157999994, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 14.099999999999994 + }, + "group": { + "cell": "tp2_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", + "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", + "trial_id": "cells/tp2_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "5455d1cdeabd6e2937130ed8fc8d86c1aaa6f04cbf26c3cb69ad064be5f67d73", + "result_sha256": "2de5c4524f57b487b94b4fdba35defc2a566236fc3e31f7260b7a9f85bb7007c", + "trial_id": "cells/tp2_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.01785714285714285, + "admitted_input_tokens_mean_over_limit": 0.0482287348533163, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.007421150278293154, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.007421150278293098, + "tpot_max_over_slo": 0.045055894329630586, + "tpot_mean_over_slo": 0.03517161669315627, + "ttft_max_over_slo_max": 0.04909342716564424, + "ttft_mean_over_slo_max": 0.07044418369208602 + }, + "delta_state": { + "batch_size.mean": 0.9503756445719365, + "batch_tokens.mean": 50.369704265273356, + "decode_batch_size.mean": 0.9415787316873265, + "graph_full_share": -0.009959703582283153, + "graph_none_share": 0.009959703582283236, + "graph_padding_fraction": -0.001694114329886807, + "kv_usage_end_minus_start": 0.0035348314893507427, + "kv_usage_max": 0.009660171254290972, + "kv_usage_mean": 0.004883298444633271, + "preemptions": 0.0, + "prefill_token_fraction": 0.007370470138318375, + "queue_running_mean": 0.27710616229999996, + "queue_waiting_mean": 1.1825361130999998, + "scheduler_steps_per_s": -17.900000000000006 + }, + "group": { + "cell": "tp2_mns8", + "level": "high", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5d41a316f329e2efcc0f728cdd078377eb72fb9258738636db704bcfcca290aa", + "result_sha256": "466b3991190a232c984ba1b54fb6793cac1893dce196ee96f7238d70d49fcb74", + "trial_id": "cells/tp2_mns8/high-rep1/result.json" + }, + "target": { + "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", + "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", + "trial_id": "cells/tp2_mns8/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.005952380952380931, + "admitted_input_tokens_mean_over_limit": -0.08086076227365946, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.04961984793917562, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.049619847939175676, + "tpot_max_over_slo": -0.03665214141383477, + "tpot_mean_over_slo": -0.03040908016559296, + "ttft_max_over_slo_max": -0.06188676766275117, + "ttft_mean_over_slo_max": -0.061940220978646765 + }, + "delta_state": { + "batch_size.mean": -0.9308857183911066, + "batch_tokens.mean": -38.259866334430285, + "decode_batch_size.mean": -0.9247474057174099, + "graph_full_share": 0.009718264940999921, + "graph_none_share": -0.009718264941000004, + "graph_padding_fraction": 0.0021664497846041342, + "kv_usage_end_minus_start": 0.0029988642599186877, + "kv_usage_max": -0.006508173500248904, + "kv_usage_mean": -0.0037595414789287837, + "preemptions": 0.0, + "prefill_token_fraction": -0.0037648024150247483, + "queue_running_mean": -0.21647643550000062, + "queue_waiting_mean": -1.8036835434999998, + "scheduler_steps_per_s": 15.700000000000003 + }, + "group": { + "cell": "tp2_mns8", + "level": "high", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", + "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", + "trial_id": "cells/tp2_mns8/high-rep2/result.json" + }, + "target": { + "requests_sha256": "5a68fda7c29477e5b3e66d70ad2b05b77be6c200cc1beea89d663a5729f31a2b", + "result_sha256": "f2116797b35057949acbf1f57898befac9ff18b94c808462f0969bdcf2446dd6", + "trial_id": "cells/tp2_mns8/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.008733624454148492, + "admitted_input_tokens_mean_over_limit": 0.03586033261326227, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.07276507276507282, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.07276507276507277, + "tpot_max_over_slo": -0.00577808629105403, + "tpot_mean_over_slo": -0.014899153875042898, + "ttft_max_over_slo_max": 0.014347533161829545, + "ttft_mean_over_slo_max": 0.00013980660489499447 + }, + "delta_state": { + "batch_size.mean": -0.6370388590554663, + "batch_tokens.mean": -0.24447251544799542, + "decode_batch_size.mean": -0.6327331911137266, + "graph_full_share": 0.004305667941739033, + "graph_none_share": -0.004305667941739005, + "graph_padding_fraction": 0.0015483374551458753, + "kv_usage_end_minus_start": -0.0016334239373173576, + "kv_usage_max": 0.0027181195206921194, + "kv_usage_mean": -0.0003114508043975048, + "preemptions": 0.0, + "prefill_token_fraction": 0.005116159306570833, + "queue_running_mean": -0.5131508089000008, + "queue_waiting_mean": -0.13343640950000002, + "scheduler_steps_per_s": 7.5 + }, + "group": { + "cell": "tp2_mns8", + "level": "low", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "455a1c21388700f0e3234c49d15ba3b0426b0d3b551c096b297793afa0d4602f", + "result_sha256": "13de25a73cc8c752245f169cfca1d64220f697c15b835a646ba969a1005047fb", + "trial_id": "cells/tp2_mns8/low-rep1/result.json" + }, + "target": { + "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", + "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", + "trial_id": "cells/tp2_mns8/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.008733624454148464, + "admitted_input_tokens_mean_over_limit": -0.08428101999879345, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.23320463320463325, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.2332046332046332, + "tpot_max_over_slo": -0.02744940897875181, + "tpot_mean_over_slo": -0.059641868091682704, + "ttft_max_over_slo_max": -0.10803754650017558, + "ttft_mean_over_slo_max": -0.032639687448619975 + }, + "delta_state": { + "batch_size.mean": -1.203120302435217, + "batch_tokens.mean": -43.54127643086966, + "decode_batch_size.mean": -1.1934092223734076, + "graph_full_share": 0.010598391508127913, + "graph_none_share": -0.0105983915081279, + "graph_padding_fraction": 0.004406513335831962, + "kv_usage_end_minus_start": 0.00967293237880118, + "kv_usage_max": -0.008881742659163083, + "kv_usage_mean": -0.005494017047481575, + "preemptions": 0.0, + "prefill_token_fraction": -0.0054281680100684015, + "queue_running_mean": -1.7641761531999998, + "queue_waiting_mean": -0.32204582449999997, + "scheduler_steps_per_s": 15.600000000000009 + }, + "group": { + "cell": "tp2_mns8", + "level": "low", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", + "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", + "trial_id": "cells/tp2_mns8/low-rep2/result.json" + }, + "target": { + "requests_sha256": "cc8c01e4560411cd5b84cdc9bebdda9a86b88615ee1c39a996affe669d26c167", + "result_sha256": "c9a7d026bd89cc027c45b8f1899d56e21d53d4234e95db365a9492f670a78d36", + "trial_id": "cells/tp2_mns8/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.005333333333333329, + "admitted_input_tokens_mean_over_limit": 0.00313201278155667, + "completed_fail_fraction_of_total": 0.004, + "completed_over_admitted": 0.00267618198037467, + "completed_pass_rate": -0.048128342245989275, + "outstanding_over_admitted": -0.00267618198037467, + "tpot_max_over_slo": 0.01678296282344799, + "tpot_mean_over_slo": 0.03333551369285431, + "ttft_max_over_slo_max": 0.04254181500679499, + "ttft_mean_over_slo_max": 0.027600657575446386 + }, + "delta_state": { + "batch_size.mean": 2.617504152501592, + "batch_tokens.mean": 83.36000248373932, + "decode_batch_size.mean": 2.6009989288874404, + "graph_full_share": -0.030922553904903838, + "graph_none_share": 0.027604433474596013, + "graph_padding_fraction": -8.22833741022888e-05, + "kv_usage_end_minus_start": -0.001912904196656684, + "kv_usage_max": 0.0016425470701958433, + "kv_usage_mean": 0.002711903470305097, + "preemptions": 0.0, + "prefill_token_fraction": 0.0031462566114718316, + "queue_running_mean": 1.2096441719000008, + "queue_waiting_mean": 2.3781169752999993, + "scheduler_steps_per_s": -13.399999999999991 + }, + "group": { + "cell": "tp4_mns16", + "level": "high", + "mns": 16, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", + "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", + "trial_id": "cells/tp4_mns16/high-rep1/result.json" + }, + "target": { + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.006666666666666654, + "admitted_input_tokens_mean_over_limit": -0.045669445146517385, + "completed_fail_fraction_of_total": -0.012, + "completed_over_admitted": 0.09077740222114916, + "completed_pass_rate": 0.13636363636363635, + "outstanding_over_admitted": -0.09077740222114916, + "tpot_max_over_slo": 0.006141297796037171, + "tpot_mean_over_slo": -0.021391237509928385, + "ttft_max_over_slo_max": -0.2410895283295152, + "ttft_mean_over_slo_max": -0.12306385203762282 + }, + "delta_state": { + "batch_size.mean": 0.023181238824859562, + "batch_tokens.mean": 2.5905518549044473, + "decode_batch_size.mean": 0.011653673102646422, + "graph_full_share": 0.0012768372268986417, + "graph_none_share": 0.001837564686330409, + "graph_padding_fraction": 0.0006975685594821312, + "kv_usage_end_minus_start": 0.001387493177308352, + "kv_usage_max": 0.0022138677902641124, + "kv_usage_mean": -0.0011780510256150529, + "preemptions": 0.0, + "prefill_token_fraction": 0.0002680116189619497, + "queue_running_mean": 0.1377117206999987, + "queue_waiting_mean": -11.247484751, + "scheduler_steps_per_s": 2.5 + }, + "group": { + "cell": "tp4_mns16", + "level": "high", + "mns": 16, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + }, + "target": { + "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", + "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", + "trial_id": "cells/tp4_mns16/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0215686274509804, + "admitted_input_tokens_mean_over_limit": -0.007677676072761208, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.12342135476463834, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.12342135476463834, + "tpot_max_over_slo": 0.042659409131532156, + "tpot_mean_over_slo": 0.06205778281240598, + "ttft_max_over_slo_max": 0.09863472549477593, + "ttft_mean_over_slo_max": 0.04373689836276399 + }, + "delta_state": { + "batch_size.mean": 3.8253490040280305, + "batch_tokens.mean": 77.16682826085241, + "decode_batch_size.mean": 3.80033895365794, + "graph_full_share": -0.03715089743892919, + "graph_none_share": 0.034228013337432306, + "graph_padding_fraction": -0.0048637698153220395, + "kv_usage_end_minus_start": -0.007677122175915652, + "kv_usage_max": 0.005958058937853616, + "kv_usage_mean": 0.004083928262059737, + "preemptions": 0.0, + "prefill_token_fraction": -0.0015885934328132567, + "queue_running_mean": 4.248843562199999, + "queue_waiting_mean": 1.9316373561999998, + "scheduler_steps_per_s": -12.799999999999997 + }, + "group": { + "cell": "tp4_mns16", + "level": "low", + "mns": 16, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", + "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", + "trial_id": "cells/tp4_mns16/low-rep1/result.json" + }, + "target": { + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.023529411764705882, + "admitted_input_tokens_mean_over_limit": 0.013229832504734862, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.1573426573426573, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.15734265734265732, + "tpot_max_over_slo": -0.05231123732945586, + "tpot_mean_over_slo": -0.08222875707353122, + "ttft_max_over_slo_max": -0.11858330133448666, + "ttft_mean_over_slo_max": -0.04432077170326809 + }, + "delta_state": { + "batch_size.mean": -6.136759360232643, + "batch_tokens.mean": -119.22330840733238, + "decode_batch_size.mean": -6.093614010489692, + "graph_full_share": 0.04765150334943136, + "graph_none_share": -0.049295061536064805, + "graph_padding_fraction": 0.001850754011979509, + "kv_usage_end_minus_start": 0.00735065319301953, + "kv_usage_max": -0.003932931028326259, + "kv_usage_mean": -0.00419858552745247, + "preemptions": 0.0, + "prefill_token_fraction": 0.0044424417100876745, + "queue_running_mean": -5.4834562212, + "queue_waiting_mean": -1.804526599, + "scheduler_steps_per_s": 31.299999999999997 + }, + "group": { + "cell": "tp4_mns16", + "level": "low", + "mns": 16, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + }, + "target": { + "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", + "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", + "trial_id": "cells/tp4_mns16/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.005333333333333329, + "admitted_input_tokens_mean_over_limit": 0.00313201278155667, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.005798394290811748, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.005798394290811748, + "tpot_max_over_slo": 0.1973487464546072, + "tpot_mean_over_slo": 0.10397174357977113, + "ttft_max_over_slo_max": -0.001787370662592977, + "ttft_mean_over_slo_max": 0.007868458508126674 + }, + "delta_state": { + "batch_size.mean": 11.95331828442438, + "batch_tokens.mean": 306.10848411182496, + "decode_batch_size.mean": 11.860920298662965, + "graph_full_share": -0.07233894773398164, + "graph_none_share": 0.06742142733113388, + "graph_padding_fraction": -0.0007349090714318936, + "kv_usage_end_minus_start": -0.0056930844866003305, + "kv_usage_max": -0.005488664935914489, + "kv_usage_mean": 0.006218554442237678, + "preemptions": 0.0, + "prefill_token_fraction": -0.002281107503328106, + "queue_running_mean": 4.266314980200001, + "queue_waiting_mean": -0.038871071699999996, + "scheduler_steps_per_s": -11.799999999999997 + }, + "group": { + "cell": "tp4_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", + "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", + "trial_id": "cells/tp4_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.006666666666666654, + "admitted_input_tokens_mean_over_limit": -0.045669445146517385, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.1202317720907774, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.1202317720907774, + "tpot_max_over_slo": -0.2208674842524746, + "tpot_mean_over_slo": -0.3211407366971467, + "ttft_max_over_slo_max": -0.007970502677684024, + "ttft_mean_over_slo_max": -0.009044386120380173 + }, + "delta_state": { + "batch_size.mean": -17.26249134948097, + "batch_tokens.mean": -537.2182539260048, + "decode_batch_size.mean": -17.083955283470857, + "graph_full_share": 0.10428533404311957, + "graph_none_share": -0.09966462603140805, + "graph_padding_fraction": 0.004369015708508233, + "kv_usage_end_minus_start": -0.020942782967763085, + "kv_usage_max": -0.019762260062552328, + "kv_usage_mean": -0.018462380855259714, + "preemptions": 0.0, + "prefill_token_fraction": -0.0019184118517130422, + "queue_running_mean": -18.914856451900004, + "queue_waiting_mean": -0.0374047109, + "scheduler_steps_per_s": 25.299999999999997 + }, + "group": { + "cell": "tp4_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", + "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", + "trial_id": "cells/tp4_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0215686274509804, + "admitted_input_tokens_mean_over_limit": -0.007677676072761208, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.17470340604668966, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.17470340604668963, + "tpot_max_over_slo": 0.048643136378671015, + "tpot_mean_over_slo": 0.059397215574861106, + "ttft_max_over_slo_max": 0.01085578350466676, + "ttft_mean_over_slo_max": 0.0005479300281428442 + }, + "delta_state": { + "batch_size.mean": 3.3204058305140567, + "batch_tokens.mean": 67.87133685773168, + "decode_batch_size.mean": 3.294527414991851, + "graph_full_share": -0.02348818567751032, + "graph_none_share": 0.022293070755162486, + "graph_padding_fraction": -0.0005360553529459057, + "kv_usage_end_minus_start": -0.010353850242237006, + "kv_usage_max": 0.004231484699196653, + "kv_usage_mean": 0.0043582810100513195, + "preemptions": 0.0, + "prefill_token_fraction": -0.001346586578878206, + "queue_running_mean": 4.306671635299999, + "queue_waiting_mean": -0.000525882, + "scheduler_steps_per_s": -14.100000000000009 + }, + "group": { + "cell": "tp4_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", + "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", + "trial_id": "cells/tp4_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.023529411764705882, + "admitted_input_tokens_mean_over_limit": 0.013229832504734862, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.2086247086247086, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.2086247086247086, + "tpot_max_over_slo": -0.049497940473067825, + "tpot_mean_over_slo": -0.07097474071479923, + "ttft_max_over_slo_max": 0.0113132930030891, + "ttft_mean_over_slo_max": 0.0027290206974555573 + }, + "delta_state": { + "batch_size.mean": -4.516372320824056, + "batch_tokens.mean": -75.80960433136954, + "decode_batch_size.mean": -4.485458989758809, + "graph_full_share": 0.024653639771889058, + "graph_none_share": -0.023884511783377624, + "graph_padding_fraction": -0.0020895387371847093, + "kv_usage_end_minus_start": 0.011432163372104731, + "kv_usage_max": -0.003015188372616051, + "kv_usage_mean": -0.004106181760760127, + "preemptions": 0.0, + "prefill_token_fraction": 0.006171117400546322, + "queue_running_mean": -5.0125981476999995, + "queue_waiting_mean": 0.000454857, + "scheduler_steps_per_s": 21.80000000000001 + }, + "group": { + "cell": "tp4_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", + "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", + "trial_id": "cells/tp4_mns64/low-rep3/result.json" + } + } + ], + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.6185880702088911, + "action_delta": { + "distinct_n": 12, + "max": 19.42960244648318, + "min": -1.4718645603306726, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.2658268122367889, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 2.327034150557677, + "repeat_abs_p95": 11.456264844377335, + "repeat_delta": { + "distinct_n": 24, + "max": 11.95331828442438, + "min": -17.26249134948097, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 17.059106336421053, + "action_delta": { + "distinct_n": 12, + "max": 656.7778499176663, + "min": -31.86018426395526, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.2374581652269279, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 71.84047059455061, + "repeat_abs_p95": 327.5325978555095, + "repeat_delta": { + "distinct_n": 24, + "max": 306.10848411182496, + "min": -537.2182539260048, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.6115154242137617, + "action_delta": { + "distinct_n": 12, + "max": 19.184206069160197, + "min": -1.4608669627184998, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.2650837352505827, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.3068764427802346, + "repeat_abs_p95": 11.363988993022012, + "repeat_delta": { + "distinct_n": 24, + "max": 11.860920298662965, + "min": -17.083955283470857, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.005588197661352101, + "action_delta": { + "distinct_n": 12, + "max": 0.02356281394840931, + "min": -0.13147494707127738, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.27121228977463446, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.020604514883878045, + "repeat_abs_p95": 0.08279928773704351, + "repeat_delta": { + "distinct_n": 24, + "max": 0.10428533404311957, + "min": -0.07611315423133669, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.011395045691744581, + "action_delta": { + "distinct_n": 12, + "max": 0.11760056457304163, + "min": -0.027804605789997328, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 8, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.5196295717736035, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.02192917091467856, + "repeat_abs_p95": 0.08301950470733124, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08050759417399073, + "min": -0.09966462603140805, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.0024624382794725458, + "action_delta": { + "distinct_n": 12, + "max": 0.008784696835623326, + "min": -0.0007445684074620391, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 1.15716396642797, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0021279942608944218, + "repeat_abs_p95": 0.007335491197810787, + "repeat_delta": { + "distinct_n": 24, + "max": 0.0086692890425959, + "min": -0.007424342975302652, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0009122627395202754, + "action_delta": { + "distinct_n": 12, + "max": 0.05537535994708087, + "min": -0.002621229743826836, + "n": 12 + }, + "action_signs": { + "consistency": 0.9166666666666666, + "negative": 1, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 0.08374755984910728, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010893006807170869, + "repeat_abs_p95": 0.08088054822007018, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08232590668398432, + "min": -0.12798481366769898, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0070274347692272365, + "action_delta": { + "distinct_n": 12, + "max": 0.08922043547542868, + "min": 0.0002683613903166071, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 0.49974360815926006, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.014062080343782424, + "repeat_abs_p95": 0.10858976920771296, + "repeat_delta": { + "distinct_n": 24, + "max": 0.11409731241882304, + "min": -0.12688580277750028, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0019193628331501261, + "action_delta": { + "distinct_n": 12, + "max": 0.04639671324585863, + "min": -0.0003621004636635179, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.29244258288516367, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.006563212560271434, + "repeat_abs_p95": 0.08042579042638397, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08571139047521727, + "min": -0.09580544440945148, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00026550413778614335, + "action_delta": { + "distinct_n": 12, + "max": 0.013365962658000052, + "min": -0.0003954091503666213, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.05125991549197503, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.005179566435838334, + "repeat_abs_p95": 0.0190949186571384, + "repeat_delta": { + "distinct_n": 24, + "max": 0.011636868678620127, + "min": -0.024511024899494305, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.43384798364999977, + "action_delta": { + "distinct_n": 12, + "max": 24.0049637595, + "min": -0.46276202019999957, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.1773939239284337, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.4456755566499995, + "repeat_abs_p95": 7.584720893309999, + "repeat_delta": { + "distinct_n": 24, + "max": 7.2005321570000005, + "min": -18.914856451900004, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.22497528095, + "action_delta": { + "distinct_n": 11, + "max": 0.0, + "min": -15.5760687653, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 10, + "positive": 0, + "zero": 2 + }, + "effect_to_repeat_median": 2.611323424650003, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0861537406, + "repeat_abs_p95": 2.3661207382899994, + "repeat_delta": { + "distinct_n": 17, + "max": 2.3781169752999993, + "min": -11.247484751, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.950000000000006, + "action_delta": { + "distinct_n": 11, + "max": 11.900000000000006, + "min": -34.5, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.44408945686900986, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 15.650000000000006, + "repeat_abs_p95": 50.385, + "repeat_delta": { + "distinct_n": 24, + "max": 65.0, + "min": -46.900000000000006, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + } + }, + "5": { + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.0008452736944928041, + "tpot_mean_over_slo": -0.0021708008398531042, + "ttft_max_over_slo_max": -0.00014104516594670713, + "ttft_mean_over_slo_max": -0.00020825385150534392 + }, + "delta_state": { + "batch_size.mean": -0.06880455035794819, + "batch_tokens.mean": -1.0156712758654507, + "decode_batch_size.mean": -0.067941551436697, + "graph_full_share": 0.0008629989212514122, + "graph_none_share": -0.006688241639697952, + "graph_padding_fraction": 0.00030316259031496623, + "kv_usage_end_minus_start": 2.700440393244019e-05, + "kv_usage_max": 0.00012663114366773254, + "kv_usage_mean": -6.934750080297807e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.000948468871704522, + "queue_running_mean": -0.028601565200000234, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 4.0 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 1, + "request_hash": "3f0ab9be34dbee884f5dbe2135b983d31a47578a43f606a5f390b0528dbedbe3", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.20670391061452514, + "mns": 8, + "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", + "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", + "trial_id": "cells/tp1_mns8/high-rep1/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.24581005586592178, + "mns": 64, + "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", + "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", + "trial_id": "cells/tp1_mns64/high-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.375, + "completed_pass_rate": -1.0, + "outstanding_over_admitted": 0.375, + "tpot_max_over_slo": -0.562845061415821, + "tpot_mean_over_slo": -0.49685783692809143, + "ttft_max_over_slo_max": -0.1160576971694051, + "ttft_mean_over_slo_max": -0.06709235994559196 + }, + "delta_state": { + "batch_size.mean": 1.9650793650793652, + "batch_tokens.mean": 127.72380952380956, + "decode_batch_size.mean": 1.9158730158730162, + "graph_full_share": -0.041269841269841345, + "graph_none_share": 0.033333333333333326, + "graph_padding_fraction": 0.0032253509426556993, + "kv_usage_end_minus_start": 0.08606331121732713, + "kv_usage_max": 0.04171319199456147, + "kv_usage_mean": 0.01377304273482051, + "preemptions": 0.0, + "prefill_token_fraction": 0.0018893720284961502, + "queue_running_mean": 1.9109891312, + "queue_waiting_mean": -1.558876379, + "scheduler_steps_per_s": -10.8 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 2, + "request_hash": "b19c91876a8417fe06cb2775f65a1111b202f88ce10ec33576af3d84c059f02a", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7988826815642458, + "mns": 8, + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.88268156424581, + "mns": 64, + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.003354239532372044, + "tpot_mean_over_slo": -0.002150681978709712, + "ttft_max_over_slo_max": -0.0004135148386315776, + "ttft_mean_over_slo_max": -0.00028004554942959864 + }, + "delta_state": { + "batch_size.mean": -0.03143779072203845, + "batch_tokens.mean": -11.489561245442736, + "decode_batch_size.mean": -0.02922096970204935, + "graph_full_share": 0.0022168210199891014, + "graph_none_share": -0.00629845367305033, + "graph_padding_fraction": 0.001725218661472923, + "kv_usage_end_minus_start": -0.01687018923627459, + "kv_usage_max": -0.01682167078536223, + "kv_usage_mean": -0.00010651327558790225, + "preemptions": 0.0, + "prefill_token_fraction": -0.009907841139181417, + "queue_running_mean": -0.027900191600000124, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "false->false", + "group": { + "level": "high", + "offered_rate_per_gpu": 2.9833333333333334, + "replicate": 3, + "request_hash": "a85849f1f057a67c82523cff57a52af22d2eb9514a953e076990aa6c25c1ef7e", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6424581005586593, + "mns": 8, + "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", + "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", + "trial_id": "cells/tp1_mns8/high-rep3/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.9273743016759777, + "mns": 64, + "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", + "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", + "trial_id": "cells/tp1_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.019319413383821316, + "tpot_mean_over_slo": -0.01717012501436968, + "ttft_max_over_slo_max": -0.0006420235052549544, + "ttft_mean_over_slo_max": -0.0016948439976355693 + }, + "delta_state": { + "batch_size.mean": -0.06047484495981781, + "batch_tokens.mean": -1.7159186818832382, + "decode_batch_size.mean": -0.059649187185791375, + "graph_full_share": 0.0008256577740266557, + "graph_none_share": -0.007128178782430002, + "graph_padding_fraction": 0.0006142370124512038, + "kv_usage_end_minus_start": 9.20038674978585e-05, + "kv_usage_max": 0.00016629145416469093, + "kv_usage_mean": -5.445550568086016e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.00036424705933824075, + "queue_running_mean": -0.16474551560000017, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 3.6000000000000085 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 1, + "request_hash": "c8ea9278f180ebb23cee595e167627cbed711032e70c9ca5ddc7205320cd75c0", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", + "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", + "trial_id": "cells/tp1_mns8/low-rep1/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", + "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", + "trial_id": "cells/tp1_mns64/low-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.15384615384615385, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.15384615384615385, + "tpot_max_over_slo": 0.040732697633234394, + "tpot_mean_over_slo": 0.014527851366132327, + "ttft_max_over_slo_max": -0.03623925517119157, + "ttft_mean_over_slo_max": -0.015724865515949205 + }, + "delta_state": { + "batch_size.mean": 0.05866351271436976, + "batch_tokens.mean": 16.67403903015969, + "decode_batch_size.mean": 0.053696037847427824, + "graph_full_share": 0.0010053222945002327, + "graph_none_share": -0.012241277350680077, + "graph_padding_fraction": 0.0038721626639551332, + "kv_usage_end_minus_start": 0.02762750849475526, + "kv_usage_max": 0.01654350181309283, + "kv_usage_mean": 0.004906340576442361, + "preemptions": 0.0, + "prefill_token_fraction": 0.0010152672534560203, + "queue_running_mean": 0.6418094667999998, + "queue_waiting_mean": -0.3949698864, + "scheduler_steps_per_s": -2.3999999999999986 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 2, + "request_hash": "1f27fe31ca32fd67f2f70a54eee83146329a7838ff2a33be48ce4372640a858c", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": -0.0007078415662871984, + "tpot_mean_over_slo": -0.000578866876769335, + "ttft_max_over_slo_max": -5.364099827906421e-05, + "ttft_mean_over_slo_max": 4.2679616146623633e-05 + }, + "delta_state": { + "batch_size.mean": -0.001802258455126493, + "batch_tokens.mean": -0.2360817774774091, + "decode_batch_size.mean": -0.00173185773422313, + "graph_full_share": 7.040072090336302e-05, + "graph_none_share": -7.040072090338037e-05, + "graph_padding_fraction": 0.0, + "kv_usage_end_minus_start": 0.00011716903913294718, + "kv_usage_max": 0.00013871042605262396, + "kv_usage_mean": 8.130641993444807e-06, + "preemptions": 0.0, + "prefill_token_fraction": -3.2355679021600636e-05, + "queue_running_mean": -0.0025477285999999655, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.19999999999999574 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.033333333333333, + "replicate": 3, + "request_hash": "a46c62ebd8b7df5534831feb70aed45d23b1a5e7f8bf3d4cd82a79f182fbc69e", + "tp": 1 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", + "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", + "trial_id": "cells/tp1_mns8/low-rep3/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", + "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", + "trial_id": "cells/tp1_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.27586206896551724, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.27586206896551724, + "tpot_max_over_slo": 0.207529695590914, + "tpot_mean_over_slo": 0.05943365186246716, + "ttft_max_over_slo_max": -0.10603234449323887, + "ttft_mean_over_slo_max": -0.017237129418515705 + }, + "delta_state": { + "batch_size.mean": 4.237397287896355, + "batch_tokens.mean": 201.35936633563546, + "decode_batch_size.mean": 4.139436406714726, + "graph_full_share": -0.05746836719261983, + "graph_none_share": 0.044500176437969455, + "graph_padding_fraction": 0.0022441507829442276, + "kv_usage_end_minus_start": 0.02343349054119659, + "kv_usage_max": 0.0207004044749316, + "kv_usage_mean": 0.007304490614720681, + "preemptions": 0.0, + "prefill_token_fraction": 0.016050535408914768, + "queue_running_mean": 9.929849956000002, + "queue_waiting_mean": -5.506802823999999, + "scheduler_steps_per_s": -29.19999999999999 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 1, + "request_hash": "5240182174ec5e25247be01c017cbefe2f1e59426bb24a57746dc40be4a82208", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.10666666666666667, + "mns": 16, + "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", + "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", + "trial_id": "cells/tp4_mns16/high-rep1/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", + "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", + "trial_id": "cells/tp4_mns64/high-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.2631578947368421, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.26315789473684204, + "tpot_max_over_slo": 0.2331512886664177, + "tpot_mean_over_slo": 0.22120977902069205, + "ttft_max_over_slo_max": -0.17202288067589203, + "ttft_mean_over_slo_max": -0.05936201432352671 + }, + "delta_state": { + "batch_size.mean": 10.325762195121952, + "batch_tokens.mean": 461.67439024390245, + "decode_batch_size.mean": 10.164329268292683, + "graph_full_share": -0.08612804878048774, + "graph_none_share": 0.07111280487804877, + "graph_padding_fraction": 0.0007778735353099922, + "kv_usage_end_minus_start": 0.03093665701006798, + "kv_usage_max": 0.026463529328836888, + "kv_usage_mean": 0.010597923626744828, + "preemptions": 0.0, + "prefill_token_fraction": 0.009329314654311416, + "queue_running_mean": 12.835752482200002, + "queue_waiting_mean": -6.7273728942, + "scheduler_steps_per_s": -23.0 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 2, + "request_hash": "51be592ba9a59c8d66eb1bcc70c3d2b011c87c66128aff36574973b40389ad96", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.092, + "mns": 16, + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.07843137254901955, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0784313725490196, + "tpot_max_over_slo": 0.05021820802747551, + "tpot_mean_over_slo": -0.012818677862446504, + "ttft_max_over_slo_max": -0.08022667251255673, + "ttft_mean_over_slo_max": -0.02098792773025101 + }, + "delta_state": { + "batch_size.mean": 0.32844164919636576, + "batch_tokens.mean": 20.103174603174637, + "decode_batch_size.mean": 0.3230008984725963, + "graph_full_share": 0.004542278127183774, + "graph_none_share": -0.017769791354697015, + "graph_padding_fraction": 0.006616409980351494, + "kv_usage_end_minus_start": 0.007401906100479305, + "kv_usage_max": 0.004352176770658023, + "kv_usage_mean": 0.0009581054950093242, + "preemptions": 0.0, + "prefill_token_fraction": 0.0019433062474734086, + "queue_running_mean": 0.822417270599999, + "queue_waiting_mean": -1.9255238218, + "scheduler_steps_per_s": 1.3999999999999915 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "high", + "offered_rate_per_gpu": 3.125, + "replicate": 3, + "request_hash": "604e338cf576d187157293922dada2bceb311567268ddae74eb7a089debf0242", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.14666666666666667, + "mns": 16, + "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", + "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", + "trial_id": "cells/tp4_mns16/high-rep3/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", + "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", + "trial_id": "cells/tp4_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.03209297007177109, + "tpot_mean_over_slo": 0.00165350785497842, + "ttft_max_over_slo_max": -0.04185720401195188, + "ttft_mean_over_slo_max": -0.008807509172654063 + }, + "delta_state": { + "batch_size.mean": -1.1226839723743751, + "batch_tokens.mean": -34.76613479399856, + "decode_batch_size.mean": -1.1063943796141942, + "graph_full_share": 0.012598237675637058, + "graph_none_share": -0.01938556799237913, + "graph_padding_fraction": -9.478587342680138e-05, + "kv_usage_end_minus_start": 3.947117882996398e-05, + "kv_usage_max": 0.004831301652910902, + "kv_usage_mean": 1.4883144002681564e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.00096100288388834, + "queue_running_mean": -0.07484160059999923, + "queue_waiting_mean": -0.1708327272, + "scheduler_steps_per_s": 12.400000000000006 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "6fb170f366b68a4e4e0fc976d370cb1f5b7b915d4dc22ebe5db3c6b9615afe68", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6196078431372549, + "mns": 16, + "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", + "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", + "trial_id": "cells/tp4_mns16/low-rep1/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", + "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", + "trial_id": "cells/tp4_mns64/low-rep1/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.10869565217391303, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.10869565217391308, + "tpot_max_over_slo": 0.031286743469536304, + "tpot_mean_over_slo": -0.024426390991776098, + "ttft_max_over_slo_max": -0.014295891169846676, + "ttft_mean_over_slo_max": -0.003572963410477855 + }, + "delta_state": { + "batch_size.mean": 0.7171517639433311, + "batch_tokens.mean": 19.4169881786475, + "decode_batch_size.mean": 0.701919812339888, + "graph_full_share": 0.004382851322571657, + "graph_none_share": -0.007268742862434022, + "graph_padding_fraction": 0.002283307241934776, + "kv_usage_end_minus_start": 0.008891646331802172, + "kv_usage_max": 0.0030741209471074615, + "kv_usage_mean": 0.0003701503920676315, + "preemptions": 0.0, + "prefill_token_fraction": 0.00017846155000900144, + "queue_running_mean": 0.8948316157999994, + "queue_waiting_mean": -1.1446055668, + "scheduler_steps_per_s": -0.8000000000000114 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "eb4c0445556d23d73d27d728b45e63fec44e17e84a4f2f341787108b7f7a4901", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6803921568627451, + "mns": 16, + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.06666666666666676, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.06666666666666665, + "tpot_max_over_slo": -0.0004955744884145297, + "tpot_mean_over_slo": -0.0060310517702504285, + "ttft_max_over_slo_max": 0.0002604483355147097, + "ttft_mean_over_slo_max": -0.003228990856401895 + }, + "delta_state": { + "batch_size.mean": -0.09294265910522981, + "batch_tokens.mean": -6.586326402016368, + "decode_batch_size.mean": -0.0904221802142402, + "graph_full_share": 0.0001575299306868505, + "graph_none_share": -0.0036231884057971037, + "graph_padding_fraction": 0.0004465429838699502, + "kv_usage_end_minus_start": -0.002635620484154888, + "kv_usage_max": 0.0003123116904728551, + "kv_usage_mean": -0.00022787683290566817, + "preemptions": 0.0, + "prefill_token_fraction": -0.0009035229748303175, + "queue_running_mean": -0.21642387019999987, + "queue_waiting_mean": -0.0184107002, + "scheduler_steps_per_s": 4.6000000000000085 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "a660c44d1f8b9c0ada5a19285fa579871bcfcae7c6fa71649546efda9aa3baef", + "tp": 4 + }, + "kind": "matched_mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.6294117647058823, + "mns": 16, + "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", + "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", + "trial_id": "cells/tp4_mns16/low-rep3/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", + "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", + "trial_id": "cells/tp4_mns64/low-rep3/result.json" + } + } + ], + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333334, + "best_feature": "tpot_max_over_slo", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.03333333333333338, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.039215686274509776, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.05434782608695651, + "train_balanced_accuracy": 0.625 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.5, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.5, + "train_balanced_accuracy": 0.625 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.033333333333333326, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0392156862745098, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.05434782608695654, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.01522073488752175, + "train_balanced_accuracy": 0.875 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 0 + ], + "threshold": -0.00409086687448007, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 0 + ], + "threshold": 0.0005373204891045424, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00025864649243734217, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.00010340366861782274, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": -0.021249613758603417, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.0074689573375508155, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.0017545182029157468, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.002461917427018732, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.002633903704056712, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.18834846816001205, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0791818658254686, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.3778079250936579, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.25, + "balanced_accuracy": 0.25, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.0002232714919349751, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.0019846847222085755, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.0027543290922952377, + "train_balanced_accuracy": 0.75 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0001671950542467852, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 1.1506892998063185e-05, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 0 + ], + "threshold": -1.97861808390893e-05, + "train_balanced_accuracy": 0.75 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0019163391379847794, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.000955475284225904, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.005609343341403783, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.7321133686999994, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.40993477099999953, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.7683205412999996, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.0092053501, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.0092053501, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.0854163636, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.9999999999999929, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 4.300000000000004, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -16.9, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "horizon_s": 5.0, + "qualifying_response_features": [], + "repeats": [ + { + "delta_outcome": { + "admitted_fraction": 0.027932960893854754, + "admitted_input_tokens_mean_over_limit": 0.25888824462890625, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.8181818181818182, + "completed_pass_rate": -1.0, + "outstanding_over_admitted": 0.8181818181818181, + "tpot_max_over_slo": -0.22724595748614082, + "tpot_mean_over_slo": -0.16274999088627282, + "ttft_max_over_slo_max": -0.034610742829196774, + "ttft_mean_over_slo_max": -0.0228428806456375 + }, + "delta_state": { + "batch_size.mean": 5.528833410386808, + "batch_tokens.mean": 384.9526121128063, + "decode_batch_size.mean": 5.423208506703652, + "graph_full_share": -0.09768839574664823, + "graph_none_share": 0.09557713052858684, + "graph_padding_fraction": -0.01691520035196146, + "kv_usage_end_minus_start": 0.16350284743730636, + "kv_usage_max": 0.1339294634828655, + "kv_usage_mean": 0.07938311082464784, + "preemptions": 0.0, + "prefill_token_fraction": 0.07360971724683552, + "queue_running_mean": 5.2693423148, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -77.8 + }, + "group": { + "cell": "tp1_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", + "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", + "trial_id": "cells/tp1_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.011173184357541902, + "admitted_input_tokens_mean_over_limit": -0.18295179094587052, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.5, + "completed_pass_rate": 1.0, + "outstanding_over_admitted": -0.5, + "tpot_max_over_slo": 0.16581994141305348, + "tpot_mean_over_slo": 0.14577003201512728, + "ttft_max_over_slo_max": 0.04351074216538109, + "ttft_mean_over_slo_max": 0.02181711295270361 + }, + "delta_state": { + "batch_size.mean": -5.595011337868481, + "batch_tokens.mean": -365.5517006802721, + "decode_batch_size.mean": -5.494557823129252, + "graph_full_share": 0.09251700680272112, + "graph_none_share": -0.08866213151927438, + "graph_padding_fraction": 0.004841272439424705, + "kv_usage_end_minus_start": -0.1340793286042561, + "kv_usage_max": -0.12953341992207013, + "kv_usage_mean": -0.0768057629061545, + "preemptions": 0.0, + "prefill_token_fraction": -0.033037405519596885, + "queue_running_mean": -4.789778502000001, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 72.8 + }, + "group": { + "cell": "tp1_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", + "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", + "trial_id": "cells/tp1_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.02459016393442623, + "admitted_input_tokens_mean_over_limit": 0.20505746694711535, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.4692307692307692, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.4692307692307693, + "tpot_max_over_slo": 0.38761381102051196, + "tpot_mean_over_slo": 0.33823327695869493, + "ttft_max_over_slo_max": -0.043946471996605396, + "ttft_mean_over_slo_max": -0.0057849611233298966 + }, + "delta_state": { + "batch_size.mean": 3.1123123406666036, + "batch_tokens.mean": 225.4005759607214, + "decode_batch_size.mean": 3.0546690586346896, + "graph_full_share": -0.04640732697573413, + "graph_none_share": 0.041473892927957695, + "graph_padding_fraction": 0.005316401000291102, + "kv_usage_end_minus_start": 0.11649515436107505, + "kv_usage_max": 0.10080927165551001, + "kv_usage_mean": 0.05787529741227895, + "preemptions": 0.0, + "prefill_token_fraction": 0.025873876159429043, + "queue_running_mean": 4.3325417076, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -59.6 + }, + "group": { + "cell": "tp1_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", + "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", + "trial_id": "cells/tp1_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.06557377049180327, + "admitted_input_tokens_mean_over_limit": 0.058956204927884615, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.3692307692307692, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.36923076923076925, + "tpot_max_over_slo": -0.41323108928586083, + "tpot_mean_over_slo": -0.3718732634627717, + "ttft_max_over_slo_max": 0.05646172333702755, + "ttft_mean_over_slo_max": 0.02716241966764857 + }, + "delta_state": { + "batch_size.mean": -3.6161048689138573, + "batch_tokens.mean": -189.70786516853934, + "decode_batch_size.mean": -3.556179775280899, + "graph_full_share": 0.048689138576779034, + "graph_none_share": -0.037453183520599245, + "graph_padding_fraction": -0.008781536045622663, + "kv_usage_end_minus_start": -0.11025077430312713, + "kv_usage_max": -0.10765311219902085, + "kv_usage_mean": -0.06249167924800681, + "preemptions": 0.0, + "prefill_token_fraction": 0.00031032253091034523, + "queue_running_mean": -5.3659050614, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 17.799999999999997 + }, + "group": { + "cell": "tp1_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", + "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", + "trial_id": "cells/tp1_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.027932960893854754, + "admitted_input_tokens_mean_over_limit": 0.25888824462890625, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.44318181818181823, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.4431818181818182, + "tpot_max_over_slo": 0.3347538302351874, + "tpot_mean_over_slo": 0.33193704520196554, + "ttft_max_over_slo_max": 0.08130590917426161, + "ttft_mean_over_slo_max": 0.044041225448449114 + }, + "delta_state": { + "batch_size.mean": 3.494949494949495, + "batch_tokens.mean": 256.21313131313127, + "decode_batch_size.mean": 3.4393939393939394, + "graph_full_share": -0.05555555555555547, + "graph_none_share": 0.05555555555555555, + "graph_padding_fraction": -0.01983738870430219, + "kv_usage_end_minus_start": 0.07746654062391167, + "kv_usage_max": 0.09234290263197176, + "kv_usage_mean": 0.06554072058902435, + "preemptions": 0.0, + "prefill_token_fraction": 0.07077187634663484, + "queue_running_mean": 3.3297516184, + "queue_waiting_mean": 1.558876379, + "scheduler_steps_per_s": -63.0 + }, + "group": { + "cell": "tp1_mns8", + "level": "high", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", + "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", + "trial_id": "cells/tp1_mns8/high-rep1/result.json" + }, + "target": { + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.011173184357541902, + "admitted_input_tokens_mean_over_limit": -0.18295179094587052, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.125, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.125, + "tpot_max_over_slo": -0.3936708804703955, + "tpot_mean_over_slo": -0.34893712293425444, + "ttft_max_over_slo_max": -0.07213344016539244, + "ttft_mean_over_slo_max": -0.04499520144345875 + }, + "delta_state": { + "batch_size.mean": -3.598494182067077, + "batch_tokens.mean": -226.33832991101983, + "decode_batch_size.mean": -3.5494638375541867, + "graph_full_share": 0.04903034451289068, + "graph_none_share": -0.04903034451289072, + "graph_padding_fraction": 0.006341404720607481, + "kv_usage_end_minus_start": -0.031145828150654364, + "kv_usage_max": -0.07099855714214642, + "kv_usage_mean": -0.06292620689574609, + "preemptions": 0.0, + "prefill_token_fraction": -0.021240192351919318, + "queue_running_mean": -2.8508891792, + "queue_waiting_mean": -1.558876379, + "scheduler_steps_per_s": 61.400000000000006 + }, + "group": { + "cell": "tp1_mns8", + "level": "high", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + }, + "target": { + "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", + "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", + "trial_id": "cells/tp1_mns8/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.02459016393442623, + "admitted_input_tokens_mean_over_limit": 0.20505746694711535, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.3153846153846153, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.31538461538461543, + "tpot_max_over_slo": 0.3275617000034563, + "tpot_mean_over_slo": 0.3065353005781929, + "ttft_max_over_slo_max": -0.008349240330668778, + "ttft_mean_over_slo_max": 0.008245060394983739 + }, + "delta_state": { + "batch_size.mean": 2.993173982992416, + "batch_tokens.mean": 207.01061824867847, + "decode_batch_size.mean": 2.9413238336014706, + "graph_full_share": -0.04658699149620771, + "graph_none_share": 0.04658699149620778, + "graph_padding_fraction": 0.0020584753487871725, + "kv_usage_end_minus_start": 0.08895964973381765, + "kv_usage_max": 0.08443206129658187, + "kv_usage_mean": 0.05291450133015573, + "preemptions": 0.0, + "prefill_token_fraction": 0.024494361846634782, + "queue_running_mean": 3.5259867252, + "queue_waiting_mean": 0.3949698864, + "scheduler_steps_per_s": -53.599999999999994 + }, + "group": { + "cell": "tp1_mns8", + "level": "low", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", + "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", + "trial_id": "cells/tp1_mns8/low-rep1/result.json" + }, + "target": { + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.06557377049180327, + "admitted_input_tokens_mean_over_limit": 0.058956204927884615, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.21538461538461534, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.2153846153846154, + "tpot_max_over_slo": -0.3717905500863392, + "tpot_mean_over_slo": -0.35676654521987006, + "ttft_max_over_slo_max": 0.02027610916411504, + "ttft_mean_over_slo_max": 0.01139487453555274 + }, + "delta_state": { + "batch_size.mean": -3.5556390977443613, + "batch_tokens.mean": -172.79774436090224, + "decode_batch_size.mean": -3.500751879699248, + "graph_full_share": 0.049624060150375904, + "graph_none_share": -0.049624060150375945, + "graph_padding_fraction": -0.00490937338166753, + "kv_usage_end_minus_start": -0.08274043484750482, + "kv_usage_max": -0.09124832081198064, + "kv_usage_mean": -0.057593469313557896, + "preemptions": 0.0, + "prefill_token_fraction": 0.0013579454633879662, + "queue_running_mean": -4.721547866, + "queue_waiting_mean": -0.3949698864, + "scheduler_steps_per_s": 15.200000000000003 + }, + "group": { + "cell": "tp1_mns8", + "level": "low", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 1 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + }, + "target": { + "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", + "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", + "trial_id": "cells/tp1_mns8/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.005797101449275366, + "admitted_input_tokens_mean_over_limit": 0.08894061053240737, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.4148148148148148, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.41481481481481475, + "tpot_max_over_slo": 0.18239212881473055, + "tpot_mean_over_slo": 0.22534509859074725, + "ttft_max_over_slo_max": 0.009043970329609387, + "ttft_mean_over_slo_max": 0.0034825192906686835 + }, + "delta_state": { + "batch_size.mean": 4.136376094514087, + "batch_tokens.mean": 294.2141424923907, + "decode_batch_size.mean": 4.035904389596526, + "graph_full_share": -0.08650808533964194, + "graph_none_share": 0.08037736393404786, + "graph_padding_fraction": -0.007242852466043042, + "kv_usage_end_minus_start": 0.03774188185038829, + "kv_usage_max": 0.02683241888245158, + "kv_usage_mean": 0.01589560565810534, + "preemptions": 0.0, + "prefill_token_fraction": 0.01670564225806337, + "queue_running_mean": 3.658735814, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -40.39999999999999 + }, + "group": { + "cell": "tp2_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "19537ec05f1e40543d850964db12234561954c209d55cd8cd0a1da716ab105fc", + "result_sha256": "49523b3269db649aa70072058666da3309ecabb93242cd504b6413f5de0f5d54", + "trial_id": "cells/tp2_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", + "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", + "trial_id": "cells/tp2_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": -0.07515914351851849, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.3333333333333333, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.3333333333333333, + "tpot_max_over_slo": -0.2056584360561972, + "tpot_mean_over_slo": -0.22610848720556842, + "ttft_max_over_slo_max": 0.0024637060026483013, + "ttft_mean_over_slo_max": -0.0009699875146560837 + }, + "delta_state": { + "batch_size.mean": -3.8698393491171377, + "batch_tokens.mean": -263.8179979938542, + "decode_batch_size.mean": -3.7790214466540357, + "graph_full_share": 0.0771251612081455, + "graph_none_share": -0.06919610871399684, + "graph_padding_fraction": 0.008047836044516822, + "kv_usage_end_minus_start": -0.028942690148230588, + "kv_usage_max": -0.029901904359948328, + "kv_usage_mean": -0.01598686261764442, + "preemptions": 0.0, + "prefill_token_fraction": -0.011907711608659222, + "queue_running_mean": -3.4762310214000003, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 33.2 + }, + "group": { + "cell": "tp2_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", + "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", + "trial_id": "cells/tp2_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "ccba3d0a18a7db8501b040855bda473ee55c519619d4b9fa2a105bda06300829", + "result_sha256": "55447419925f30397d19863e149890241c58863d0adaafd511a67757905cb58e", + "trial_id": "cells/tp2_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.017021276595744678, + "admitted_input_tokens_mean_over_limit": 0.045296223958333326, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.33333333333333337, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.33333333333333337, + "tpot_max_over_slo": 0.22320900677050662, + "tpot_mean_over_slo": 0.15232562887487847, + "ttft_max_over_slo_max": 0.0061178085064360205, + "ttft_mean_over_slo_max": 0.018187332690771045 + }, + "delta_state": { + "batch_size.mean": 4.4320356115196375, + "batch_tokens.mean": 99.22318191879197, + "decode_batch_size.mean": 4.402188829591008, + "graph_full_share": -0.026162662010280058, + "graph_none_share": 0.02238508644647205, + "graph_padding_fraction": 0.010726240511491474, + "kv_usage_end_minus_start": -0.01923544232564689, + "kv_usage_max": 0.02004118226348972, + "kv_usage_mean": 0.014893345374172863, + "preemptions": 0.0, + "prefill_token_fraction": -0.00708953487763786, + "queue_running_mean": 4.799239215399999, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -38.400000000000006 + }, + "group": { + "cell": "tp2_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "8685f475baf247f7ce52a6a2df1183cd77a6f580cd57a3a87c407b4e96c0895f", + "result_sha256": "7a8f4e462d496bd56b73b0a686653cbe499a238df62a8c9a50db687d051f1c68", + "trial_id": "cells/tp2_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", + "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", + "trial_id": "cells/tp2_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.042553191489361694, + "admitted_input_tokens_mean_over_limit": -0.026056198846726164, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.11904761904761907, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.11904761904761904, + "tpot_max_over_slo": -0.2355341601559496, + "tpot_mean_over_slo": -0.1643432670810964, + "ttft_max_over_slo_max": -0.026516160830700145, + "ttft_mean_over_slo_max": -0.01345709933763525 + }, + "delta_state": { + "batch_size.mean": -5.32536868824603, + "batch_tokens.mean": -133.89198850405924, + "decode_batch_size.mean": -5.279523379974407, + "graph_full_share": 0.04036585621682853, + "graph_none_share": -0.034001132811680546, + "graph_padding_fraction": -0.008571843774165616, + "kv_usage_end_minus_start": 0.0019567969919042083, + "kv_usage_max": -0.028865953011293133, + "kv_usage_mean": -0.018137793999508895, + "preemptions": 0.0, + "prefill_token_fraction": 0.0041833878419771375, + "queue_running_mean": -6.512575867199999, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 57.599999999999994 + }, + "group": { + "cell": "tp2_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", + "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", + "trial_id": "cells/tp2_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "5455d1cdeabd6e2937130ed8fc8d86c1aaa6f04cbf26c3cb69ad064be5f67d73", + "result_sha256": "2de5c4524f57b487b94b4fdba35defc2a566236fc3e31f7260b7a9f85bb7007c", + "trial_id": "cells/tp2_mns64/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.005952380952380959, + "admitted_input_tokens_mean_over_limit": 0.11506300706129807, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.16666666666666663, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.16666666666666669, + "tpot_max_over_slo": 0.0634281492909402, + "tpot_mean_over_slo": 0.06688242088001703, + "ttft_max_over_slo_max": 0.13068378849614723, + "ttft_mean_over_slo_max": 0.05409936663613152 + }, + "delta_state": { + "batch_size.mean": 1.552585603284773, + "batch_tokens.mean": 89.12552429250253, + "decode_batch_size.mean": 1.5358014452458475, + "graph_full_share": -0.018898322943787682, + "graph_none_share": 0.01889832294378769, + "graph_padding_fraction": -0.003394579943621764, + "kv_usage_end_minus_start": 0.000535967229432166, + "kv_usage_max": 0.01052792772099076, + "kv_usage_mean": 0.006505251073294799, + "preemptions": 0.0, + "prefill_token_fraction": 0.010329730802724923, + "queue_running_mean": 0.5542123245999999, + "queue_waiting_mean": 2.0910913376, + "scheduler_steps_per_s": -32.39999999999999 + }, + "group": { + "cell": "tp2_mns8", + "level": "high", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5d41a316f329e2efcc0f728cdd078377eb72fb9258738636db704bcfcca290aa", + "result_sha256": "466b3991190a232c984ba1b54fb6793cac1893dce196ee96f7238d70d49fcb74", + "trial_id": "cells/tp2_mns8/high-rep1/result.json" + }, + "target": { + "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", + "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", + "trial_id": "cells/tp2_mns8/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.0029761904761904795, + "admitted_input_tokens_mean_over_limit": -0.09617018479567307, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.14, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.14, + "tpot_max_over_slo": -0.07476021338613548, + "tpot_mean_over_slo": -0.06040260059613095, + "ttft_max_over_slo_max": -0.0990663813330078, + "ttft_mean_over_slo_max": -0.03931008928012512 + }, + "delta_state": { + "batch_size.mean": -1.5731600373589236, + "batch_tokens.mean": -75.68624207727024, + "decode_batch_size.mean": -1.5588509992086177, + "graph_full_share": 0.01874362795074891, + "graph_none_share": -0.018743627950748964, + "graph_padding_fraction": 0.004413545938114254, + "kv_usage_end_minus_start": 0.003471025866799371, + "kv_usage_max": -0.012199635031839073, + "kv_usage_mean": -0.007236177184589677, + "preemptions": 0.0, + "prefill_token_fraction": -0.006096255385559024, + "queue_running_mean": -0.43295287099999946, + "queue_waiting_mean": -1.9712983662, + "scheduler_steps_per_s": 28.0 + }, + "group": { + "cell": "tp2_mns8", + "level": "high", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", + "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", + "trial_id": "cells/tp2_mns8/high-rep2/result.json" + }, + "target": { + "requests_sha256": "5a68fda7c29477e5b3e66d70ad2b05b77be6c200cc1beea89d663a5729f31a2b", + "result_sha256": "f2116797b35057949acbf1f57898befac9ff18b94c808462f0969bdcf2446dd6", + "trial_id": "cells/tp2_mns8/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.017467248908296942, + "admitted_input_tokens_mean_over_limit": 0.045296223958333326, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.16666666666666663, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.16666666666666669, + "tpot_max_over_slo": 0.06375630676753702, + "tpot_mean_over_slo": 0.0900300343637355, + "ttft_max_over_slo_max": 0.051512272330001, + "ttft_mean_over_slo_max": 0.0365448741568495 + }, + "delta_state": { + "batch_size.mean": 3.2164733311475753, + "batch_tokens.mean": 83.75307177093255, + "decode_batch_size.mean": 3.188651751468882, + "graph_full_share": -0.027821579678693587, + "graph_none_share": 0.027821579678693545, + "graph_padding_fraction": -0.0015661122787500655, + "kv_usage_end_minus_start": -0.011063894950423037, + "kv_usage_max": 0.003075431006980267, + "kv_usage_mean": 0.009516064845812437, + "preemptions": 0.0, + "prefill_token_fraction": -0.003183579698054051, + "queue_running_mean": 2.5618635706, + "queue_waiting_mean": 1.0030269042, + "scheduler_steps_per_s": -25.60000000000001 + }, + "group": { + "cell": "tp2_mns8", + "level": "low", + "mns": 8, + "source_replicate": 1, + "target_replicate": 2, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "455a1c21388700f0e3234c49d15ba3b0426b0d3b551c096b297793afa0d4602f", + "result_sha256": "13de25a73cc8c752245f169cfca1d64220f697c15b835a646ba969a1005047fb", + "trial_id": "cells/tp2_mns8/low-rep1/result.json" + }, + "target": { + "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", + "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", + "trial_id": "cells/tp2_mns8/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.04803493449781659, + "admitted_input_tokens_mean_over_limit": -0.009297688802083315, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.10256410256410264, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.10256410256410253, + "tpot_max_over_slo": -0.07959600126436789, + "tpot_mean_over_slo": -0.121404858913335, + "ttft_max_over_slo_max": -0.07178450832725503, + "ttft_mean_over_slo_max": -0.03313614673934353 + }, + "delta_state": { + "batch_size.mean": -4.304456872572413, + "batch_tokens.mean": -123.40500688035755, + "decode_batch_size.mean": -4.261865986626437, + "graph_full_share": 0.04259088594597549, + "graph_none_share": -0.04259088594597552, + "graph_padding_fraction": 0.005238254196308781, + "kv_usage_end_minus_start": -0.0053979556678535046, + "kv_usage_max": -0.011102178323953882, + "kv_usage_mean": -0.01280541269135944, + "preemptions": 0.0, + "prefill_token_fraction": 0.000758348233463435, + "queue_running_mean": -4.5151126308, + "queue_waiting_mean": -1.010089547, + "scheduler_steps_per_s": 49.60000000000001 + }, + "group": { + "cell": "tp2_mns8", + "level": "low", + "mns": 8, + "source_replicate": 2, + "target_replicate": 3, + "tp": 2 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", + "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", + "trial_id": "cells/tp2_mns8/low-rep2/result.json" + }, + "target": { + "requests_sha256": "cc8c01e4560411cd5b84cdc9bebdda9a86b88615ee1c39a996affe669d26c167", + "result_sha256": "c9a7d026bd89cc027c45b8f1899d56e21d53d4234e95db365a9492f670a78d36", + "trial_id": "cells/tp2_mns8/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.0013333333333333391, + "admitted_input_tokens_mean_over_limit": 0.0868109341325431, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.025105868118572294, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.025105868118572294, + "tpot_max_over_slo": 0.029982028815044692, + "tpot_mean_over_slo": 0.05498080777948092, + "ttft_max_over_slo_max": 0.05707128983340226, + "ttft_mean_over_slo_max": 0.04147020173474 + }, + "delta_state": { + "batch_size.mean": 3.8799424686192463, + "batch_tokens.mean": 164.1232217573222, + "decode_batch_size.mean": 3.8344665271966534, + "graph_full_share": -0.04971234309623429, + "graph_none_share": 0.042429393305439336, + "graph_padding_fraction": -0.000239797608965738, + "kv_usage_end_minus_start": 0.003295296296107475, + "kv_usage_max": 0.004723598096277648, + "kv_usage_mean": 0.003950784826693938, + "preemptions": 0.0, + "prefill_token_fraction": 0.013157938495005683, + "queue_running_mean": 2.4192883438, + "queue_waiting_mean": 1.2621521842000005, + "scheduler_steps_per_s": -31.599999999999994 + }, + "group": { + "cell": "tp4_mns16", + "level": "high", + "mns": 16, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", + "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", + "trial_id": "cells/tp4_mns16/high-rep1/result.json" + }, + "target": { + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.007999999999999993, + "admitted_input_tokens_mean_over_limit": -0.08157169117647056, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.1713106295149639, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.1713106295149639, + "tpot_max_over_slo": -0.04532225479511176, + "tpot_mean_over_slo": -0.04295654146532207, + "ttft_max_over_slo_max": -0.031098171496220545, + "ttft_mean_over_slo_max": -0.018361275960999665 + }, + "delta_state": { + "batch_size.mean": 0.28468665768194157, + "batch_tokens.mean": -55.978571428571456, + "decode_batch_size.mean": 0.2811657681940698, + "graph_full_share": 0.016324123989218342, + "graph_none_share": -0.006949123989218334, + "graph_padding_fraction": 0.0015473261462516127, + "kv_usage_end_minus_start": -0.0032850941403919087, + "kv_usage_max": -0.00460117222769163, + "kv_usage_mean": -0.0019998572533929705, + "preemptions": 0.0, + "prefill_token_fraction": -0.008209306549717632, + "queue_running_mean": 0.27542344140000097, + "queue_waiting_mean": -4.8449581018000005, + "scheduler_steps_per_s": 10.200000000000003 + }, + "group": { + "cell": "tp4_mns16", + "level": "high", + "mns": 16, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + }, + "target": { + "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", + "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", + "trial_id": "cells/tp4_mns16/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0196078431372549, + "admitted_input_tokens_mean_over_limit": -0.009788771182442646, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.20048309178743962, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.20048309178743962, + "tpot_max_over_slo": 0.042659409131532156, + "tpot_mean_over_slo": 0.04424057361688544, + "ttft_max_over_slo_max": -0.048664552169308685, + "ttft_mean_over_slo_max": -0.015179490139230319 + }, + "delta_state": { + "batch_size.mean": 2.2298197150334396, + "batch_tokens.mean": 48.29412619947661, + "decode_batch_size.mean": 2.216560046525153, + "graph_full_share": -0.03405059610351846, + "graph_none_share": 0.02576330328583891, + "graph_padding_fraction": -0.0005586075805253821, + "kv_usage_end_minus_start": 0.005978463249284638, + "kv_usage_max": 0.005988665405000093, + "kv_usage_mean": 0.0033715964225947276, + "preemptions": 0.0, + "prefill_token_fraction": -0.0013328880320482295, + "queue_running_mean": 2.578025910000001, + "queue_waiting_mean": 0.9737728395999999, + "scheduler_steps_per_s": -3.5999999999999943 + }, + "group": { + "cell": "tp4_mns16", + "level": "low", + "mns": 16, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", + "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", + "trial_id": "cells/tp4_mns16/low-rep1/result.json" + }, + "target": { + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.03137254901960784, + "admitted_input_tokens_mean_over_limit": 0.04353239639945655, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.17826086956521736, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.17826086956521742, + "tpot_max_over_slo": -0.06113145905583517, + "tpot_mean_over_slo": -0.07880983479690457, + "ttft_max_over_slo_max": 0.028715976329597957, + "ttft_mean_over_slo_max": 0.012690293574927458 + }, + "delta_state": { + "batch_size.mean": -5.802285141359179, + "batch_tokens.mean": -131.81382573186144, + "decode_batch_size.mean": -5.749516966234634, + "graph_full_share": 0.0604758274237851, + "graph_none_share": -0.055969252942589476, + "graph_padding_fraction": 0.00012825203851504416, + "kv_usage_end_minus_start": -0.0044583420476747015, + "kv_usage_max": -0.003932931028326259, + "kv_usage_mean": -0.00443657185283492, + "preemptions": 0.0, + "prefill_token_fraction": 0.003659185530759501, + "queue_running_mean": -4.786444050200001, + "queue_waiting_mean": -1.1252851525999998, + "scheduler_steps_per_s": 33.39999999999999 + }, + "group": { + "cell": "tp4_mns16", + "level": "low", + "mns": 16, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + }, + "target": { + "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", + "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", + "trial_id": "cells/tp4_mns16/low-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.0013333333333333391, + "admitted_input_tokens_mean_over_limit": 0.0868109341325431, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.03781004234724741, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0378100423472475, + "tpot_max_over_slo": 0.055603621890548405, + "tpot_mean_over_slo": 0.21675693493770581, + "ttft_max_over_slo_max": -0.008919246349250898, + "ttft_mean_over_slo_max": -0.0006546831702710018 + }, + "delta_state": { + "batch_size.mean": 9.968307375844843, + "batch_tokens.mean": 424.4382456655892, + "decode_batch_size.mean": 9.85935938877461, + "graph_full_share": -0.0783720246841022, + "graph_none_share": 0.06904202174551866, + "graph_padding_fraction": -0.0017060748565999734, + "kv_usage_end_minus_start": 0.010798462764978867, + "kv_usage_max": 0.010486722950182936, + "kv_usage_mean": 0.007244217838718085, + "preemptions": 0.0, + "prefill_token_fraction": 0.006436717740402331, + "queue_running_mean": 5.32519087, + "queue_waiting_mean": 0.041582113999999996, + "scheduler_steps_per_s": -25.400000000000006 + }, + "group": { + "cell": "tp4_mns64", + "level": "high", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", + "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", + "trial_id": "cells/tp4_mns64/high-rep1/result.json" + }, + "target": { + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.007999999999999993, + "admitted_input_tokens_mean_over_limit": -0.08157169117647056, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.35603715170278644, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.3560371517027863, + "tpot_max_over_slo": -0.22825533543405396, + "tpot_mean_over_slo": -0.27698499834846063, + "ttft_max_over_slo_max": 0.06069803666711475, + "ttft_mean_over_slo_max": 0.020012810632276036 + }, + "delta_state": { + "batch_size.mean": -9.712633888243644, + "batch_tokens.mean": -497.54978706929927, + "decode_batch_size.mean": -9.560162601626017, + "graph_full_share": 0.10699445089688986, + "graph_none_share": -0.09583172022196412, + "graph_padding_fraction": 0.007385862591293114, + "kv_usage_end_minus_start": -0.026819845049980584, + "kv_usage_max": -0.026712524785870495, + "kv_usage_mean": -0.011639675385128475, + "preemptions": 0.0, + "prefill_token_fraction": -0.01559531495655564, + "queue_running_mean": -11.737911770200002, + "queue_waiting_mean": -0.0431090294, + "scheduler_steps_per_s": 34.599999999999994 + }, + "group": { + "cell": "tp4_mns64", + "level": "high", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + }, + "target": { + "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", + "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", + "trial_id": "cells/tp4_mns64/high-rep3/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0196078431372549, + "admitted_input_tokens_mean_over_limit": -0.009788771182442646, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": -0.30917874396135264, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.3091787439613527, + "tpot_max_over_slo": 0.04185318252929737, + "tpot_mean_over_slo": 0.01816067477013092, + "ttft_max_over_slo_max": -0.021103239327203482, + "ttft_mean_over_slo_max": -0.00994494437705411 + }, + "delta_state": { + "batch_size.mean": 4.069655451351146, + "batch_tokens.mean": 102.47724917212267, + "decode_batch_size.mean": 4.024874238479235, + "graph_full_share": -0.04226598245658386, + "graph_none_share": 0.03788012841578402, + "graph_padding_fraction": 0.0018194855348361953, + "kv_usage_end_minus_start": 0.014830638402256846, + "kv_usage_max": 0.004231484699196653, + "kv_usage_mean": 0.0037268636706596775, + "preemptions": 0.0, + "prefill_token_fraction": -0.00019342359815088805, + "queue_running_mean": 3.5476991263999995, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -16.80000000000001 + }, + "group": { + "cell": "tp4_mns64", + "level": "low", + "mns": 64, + "source_replicate": 1, + "target_replicate": 2, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", + "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", + "trial_id": "cells/tp4_mns64/low-rep1/result.json" + }, + "target": { + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": -0.03137254901960784, + "admitted_input_tokens_mean_over_limit": 0.04353239639945655, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.35362318840579715, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.35362318840579715, + "tpot_max_over_slo": -0.092913777013786, + "tpot_mean_over_slo": -0.060414495575378896, + "ttft_max_over_slo_max": 0.04327231583495934, + "ttft_mean_over_slo_max": 0.013034266129003418 + }, + "delta_state": { + "batch_size.mean": -6.61237956440774, + "batch_tokens.mean": -157.8171403125253, + "decode_batch_size.mean": -6.541858958788763, + "graph_full_share": 0.056250506031900294, + "graph_none_share": -0.05232369848595256, + "graph_padding_fraction": -0.0017085122195497817, + "kv_usage_end_minus_start": -0.01598560886363176, + "kv_usage_max": -0.006694740284960865, + "kv_usage_mean": -0.00503459907780822, + "preemptions": 0.0, + "prefill_token_fraction": 0.002577201005920182, + "queue_running_mean": -5.8976995362, + "queue_waiting_mean": 0.000909714, + "scheduler_steps_per_s": 38.80000000000001 + }, + "group": { + "cell": "tp4_mns64", + "level": "low", + "mns": 64, + "source_replicate": 2, + "target_replicate": 3, + "tp": 4 + }, + "kind": "same_config_workload_repeat", + "source": { + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + }, + "target": { + "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", + "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", + "trial_id": "cells/tp4_mns64/low-rep3/result.json" + } + } + ], + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.21069215415079778, + "action_delta": { + "distinct_n": 12, + "max": 10.325762195121952, + "min": -1.1226839723743751, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.054373699571412804, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 3.874890908868192, + "repeat_abs_p95": 9.247595739668252, + "repeat_delta": { + "distinct_n": 24, + "max": 9.968307375844843, + "min": -9.712633888243644, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 18.045513604403595, + "action_delta": { + "distinct_n": 12, + "max": 461.67439024390245, + "min": -34.76613479399856, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.10712015825142496, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 168.46048305911222, + "repeat_abs_p95": 418.5154006326717, + "repeat_delta": { + "distinct_n": 24, + "max": 424.4382456655892, + "min": -497.54978706929927, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.20671153934341824, + "action_delta": { + "distinct_n": 12, + "max": 10.164329268292683, + "min": -1.1063943796141942, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.05430140299778245, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.8067439869253445, + "repeat_abs_p95": 9.107417055200422, + "repeat_delta": { + "distinct_n": 24, + "max": 9.85935938877461, + "min": -9.560162601626017, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.003299836171280379, + "action_delta": { + "distinct_n": 12, + "max": 0.012598237675637058, + "min": -0.08612804878048774, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.06753691417406232, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.048859741544834856, + "repeat_abs_p95": 0.09691268740505915, + "repeat_delta": { + "distinct_n": 24, + "max": 0.10699445089688986, + "min": -0.09768839574664823, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00975501010655705, + "action_delta": { + "distinct_n": 12, + "max": 0.07111280487804877, + "min": -0.01938556799237913, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 9, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.21877645860951367, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.04458893872109165, + "repeat_abs_p95": 0.09453988067718995, + "repeat_delta": { + "distinct_n": 24, + "max": 0.09557713052858684, + "min": -0.09583172022196412, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0012515460983914576, + "action_delta": { + "distinct_n": 12, + "max": 0.006616409980351494, + "min": -9.478587342680138e-05, + "n": 12 + }, + "action_signs": { + "consistency": 0.9090909090909091, + "negative": 1, + "positive": 10, + "zero": 1 + }, + "effect_to_repeat_median": 0.2567104007991265, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.004875322910546117, + "repeat_abs_p95": 0.015986856375890948, + "repeat_delta": { + "distinct_n": 24, + "max": 0.010726240511491474, + "min": -0.01983738870430219, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.008146776216140739, + "action_delta": { + "distinct_n": 12, + "max": 0.08606331121732713, + "min": -0.01687018923627459, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 0.4626083515997178, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.017610525594639326, + "repeat_abs_p95": 0.1314417024677789, + "repeat_delta": { + "distinct_n": 24, + "max": 0.16350284743730636, + "min": -0.1340793286042561, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.004591739211784462, + "action_delta": { + "distinct_n": 12, + "max": 0.04171319199456147, + "min": -0.01682167078536223, + "n": 12 + }, + "action_signs": { + "consistency": 0.9166666666666666, + "negative": 1, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 0.19642246579235886, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.023376853524680108, + "repeat_abs_p95": 0.1262513737636127, + "repeat_delta": { + "distinct_n": 24, + "max": 0.1339294634828655, + "min": -0.12953341992207013, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0002990136124866498, + "action_delta": { + "distinct_n": 12, + "max": 0.01377304273482051, + "min": -0.00022787683290566817, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.021590398513840623, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.013849379032766153, + "repeat_abs_p95": 0.07511600655858494, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07938311082464784, + "min": -0.0768057629061545, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0009881350686721802, + "action_delta": { + "distinct_n": 12, + "max": 0.016050535408914768, + "min": -0.009907841139181417, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.12917776465155323, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.007649420713677746, + "repeat_abs_p95": 0.06511170572257907, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07360971724683552, + "min": -0.033037405519596885, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.42911666849999985, + "action_delta": { + "distinct_n": 12, + "max": 12.835752482200002, + "min": -0.21642387019999987, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.10739626232229334, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.9956387608, + "repeat_abs_p95": 6.420344417549997, + "repeat_delta": { + "distinct_n": 24, + "max": 5.32519087, + "min": -11.737911770200002, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.28290130680000003, + "action_delta": { + "distinct_n": 9, + "max": 0.0, + "min": -6.7273728942, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 8, + "positive": 0, + "zero": 4 + }, + "effect_to_repeat_median": 1.2915540857901293, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.2190394579, + "repeat_abs_p95": 2.0731223918899997, + "repeat_delta": { + "distinct_n": 16, + "max": 2.0910913376, + "min": -4.8449581018000005, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 3.8000000000000043, + "action_delta": { + "distinct_n": 12, + "max": 12.400000000000006, + "min": -29.19999999999999, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.11176470588235309, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 33.99999999999999, + "repeat_abs_p95": 71.32999999999997, + "repeat_delta": { + "distinct_n": 24, + "max": 72.8, + "min": -77.8, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + } + } + }, + "provenance": { + "analysis_script": "/home/gahow/phd/aituner/runs/intervention-response-v0/analyze_p1.py", + "analysis_script_sha256": "902adcba58bcb73f97c455fdea3710f6cb28fa8d8f4ea5bc4915143ed106b17a", + "manifest": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/pilot-manifest.json", + "manifest_sha256": "86b249a087ab1bcd51cd8de3d1da6bf34afd41af977eb58c93229b941dbf5c8a", + "manifest_validation": { + "expected_trials": 36, + "matched_trials": 36, + "schema": "fidelity-prefix-pilot-manifest-v1" + }, + "phase6_v0_script_sha256": "f1b40c350f22e797b08b129df7b55386b407bd5ae7f466d36987e94d768ded2f", + "run_root": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b", + "streams": [ + { + "bytes": 24501686, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns64/opprof/opprof-v1-dp0-pid234511-1784005631788937266.jsonl", + "sha256": "a813076afe677e5c13010b2ec44f46998794ca44cdf30ee81be836e3623f0068" + }, + { + "bytes": 25524027, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns8/opprof/opprof-v1-dp0-pid227734-1784005177936445360.jsonl", + "sha256": "1708759d2e77c42bff1940d43214fb2d927c3eebeda62d53a3fb7d04817e9b22" + }, + { + "bytes": 29431988, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns64/opprof/opprof-v1-dp0-pid247791-1784006495467408663.jsonl", + "sha256": "65355de883f2eb5d31be161d82b26dddc53c41a0a35a0fdb807777bc5b21c5ec" + }, + { + "bytes": 26036993, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns8/opprof/opprof-v1-dp0-pid241594-1784006096655191299.jsonl", + "sha256": "485e203d37d45fdd3c4de768b95da2050478f111f4d739b129d4f4ebd4737b70" + }, + { + "bytes": 17449143, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns16/opprof/opprof-v1-dp0-pid256090-1784007028237607411.jsonl", + "sha256": "59e98014d01bfc68d9f97112ecd782cb03571ad437382ed67416c7b85ae48c56" + }, + { + "bytes": 22674678, + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns64/opprof/opprof-v1-dp0-pid262293-1784007376383569515.jsonl", + "sha256": "03e4f3cca3023a78e859b663703b9468dedf89e58320f0292e2973070e7616c8" + } + ] + }, + "sanity": { + "red_flags": [], + "stream_bytes": { + "distinct_n": 6, + "max": 29431988.0, + "min": 17449143.0, + "n": 6 + }, + "stream_count": 6 + }, + "schema": "intervention-response-p1-confirmation-v1", + "stable_incremental_efficacy_features": [], + "stable_response_features": [], + "status": "COMPLETE" +} diff --git a/runs/intervention-response-v0/phase6-audit.json b/runs/intervention-response-v0/phase6-audit.json new file mode 100644 index 0000000..6231482 --- /dev/null +++ b/runs/intervention-response-v0/phase6-audit.json @@ -0,0 +1,5121 @@ +{ + "claim_boundary": "Development-only identifiability gate. Passing opens a controlled real-GPU pilot; it does not establish tuning benefit or causality.", + "decision": "STOP_NO_IDENTIFIABLE_RESPONSE", + "frozen_gate": { + "expected_action_pairs": 17, + "gate_features": [ + "scheduler_steps_per_s", + "decode_batch_size.mean", + "prefill_token_fraction", + "queue_waiting_mean", + "queue_running_mean", + "kv_usage_mean", + "graph_padding_fraction" + ], + "horizons_s": [ + 5.0, + 10.0 + ], + "minimum_action_above_repeat_p95_fraction": 0.5, + "minimum_effect_to_repeat_median": 2.0, + "minimum_repeat_pairs": 20, + "minimum_sign_consistency": 0.75, + "minimum_stable_features": 2 + }, + "horizons": { + "10": { + "actions": [ + { + "delta_state": { + "batch_size.mean": 1.5764389277324504, + "batch_tokens.mean": 81.00995765564915, + "decode_batch_size.mean": 1.5631264281332364, + "graph_full_share": -0.011964790704335537, + "graph_none_share": 0.020610006604949223, + "graph_padding_fraction": -0.01937281246147585, + "kv_usage_end_minus_start": -0.02063938485088157, + "kv_usage_max": 0.05576515049492037, + "kv_usage_mean": 0.02372409715700479, + "preemptions": 0.0, + "prefill_token_fraction": 0.09894954087875252, + "queue_running_mean": 1.8681007873, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -51.89999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.003449406279594669, + "batch_tokens.mean": 0.2550229474757799, + "decode_batch_size.mean": 0.003361987324251192, + "graph_full_share": -8.37764988708134e-05, + "graph_none_share": -0.0012785022218984483, + "graph_padding_fraction": 0.00024814146059230855, + "kv_usage_end_minus_start": 4.607711554860394e-05, + "kv_usage_max": 0.00030157213266412963, + "kv_usage_mean": 0.0007577042092319561, + "preemptions": 0.0, + "prefill_token_fraction": 6.688650356756476e-05, + "queue_running_mean": 0.05724569279999958, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.20000000000000284 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.012661422160780589, + "batch_tokens.mean": -0.39155502891061644, + "decode_batch_size.mean": -0.012529760925139577, + "graph_full_share": 0.0001261753508223773, + "graph_none_share": -0.0014757030161394735, + "graph_padding_fraction": 0.0004267789129811959, + "kv_usage_end_minus_start": 7.934839022727402e-06, + "kv_usage_max": 0.03040170172083234, + "kv_usage_mean": 0.0025619580859006993, + "preemptions": 0.0, + "prefill_token_fraction": -2.6753484588915022e-05, + "queue_running_mean": 0.22427197499999973, + "queue_waiting_mean": -0.0255129477, + "scheduler_steps_per_s": 0.29999999999999716 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.04195804195804198, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.958041958041958, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.012870461691032453, + "batch_tokens.mean": -0.2637663627959057, + "decode_batch_size.mean": -0.01278327820794356, + "graph_full_share": 8.355083796041107e-05, + "graph_none_share": -0.002771708233208551, + "graph_padding_fraction": -4.1151267345866487e-05, + "kv_usage_end_minus_start": 1.7658454245395028e-05, + "kv_usage_max": 0.00011557359098801623, + "kv_usage_mean": -0.00031444354435781335, + "preemptions": 0.0, + "prefill_token_fraction": 2.6753484588915022e-05, + "queue_running_mean": -0.05587743330000006, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.20000000000000284 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.021751047667267098, + "batch_tokens.mean": 0.5249008791116694, + "decode_batch_size.mean": 0.021576208801315833, + "graph_full_share": -0.00016755391320344426, + "graph_none_share": -0.0012074809179768962, + "graph_padding_fraction": 9.628532857289349e-05, + "kv_usage_end_minus_start": 4.607711554860394e-05, + "kv_usage_max": 0.00030157213266412963, + "kv_usage_mean": 0.00019975354503918435, + "preemptions": 0.0, + "prefill_token_fraction": -1.337692842384719e-05, + "queue_running_mean": 0.0040136895999998146, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.3999999999999915 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 6.1794369551023545, + "batch_tokens.mean": 175.96941851979875, + "decode_batch_size.mean": 6.118283596420478, + "graph_full_share": -0.05096593309521069, + "graph_none_share": 0.047163651726389366, + "graph_padding_fraction": 0.0032959117320311947, + "kv_usage_end_minus_start": 0.07533673225448412, + "kv_usage_max": 0.06162226941794913, + "kv_usage_mean": 0.056370734075687004, + "preemptions": 0.0, + "prefill_token_fraction": -0.0018186060572318308, + "queue_running_mean": 5.9180508831, + "queue_waiting_mean": -2.7288210595, + "scheduler_steps_per_s": -13.3 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.06521739130434782, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.15579710144927536, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 4.764302767707134, + "batch_tokens.mean": 277.81382085064996, + "decode_batch_size.mean": 4.6529755062339735, + "graph_full_share": -0.09344327526748608, + "graph_none_share": 0.0681757891944469, + "graph_padding_fraction": 0.000877004096309171, + "kv_usage_end_minus_start": 0.13999257948102506, + "kv_usage_max": 0.13096340508568438, + "kv_usage_mean": 0.05558263246404621, + "preemptions": 0.0, + "prefill_token_fraction": 0.006009982585939322, + "queue_running_mean": 6.513921786700001, + "queue_waiting_mean": -2.611976143, + "scheduler_steps_per_s": -9.100000000000001 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.06521739130434782, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.15579710144927536, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.8274828140997492, + "batch_tokens.mean": 96.29113646336111, + "decode_batch_size.mean": -0.8609770367120078, + "graph_full_share": -0.017588123445955728, + "graph_none_share": 0.015211350007313162, + "graph_padding_fraction": 7.496352550239935e-05, + "kv_usage_end_minus_start": 0.03212970309780483, + "kv_usage_max": 0.01812081632758722, + "kv_usage_mean": 0.003998827610290162, + "preemptions": 0.0, + "prefill_token_fraction": 0.004183520933583096, + "queue_running_mean": 0.37492188330000076, + "queue_waiting_mean": -0.1123975519, + "scheduler_steps_per_s": -1.299999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.02898550724637681, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06159420289855073, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.07209087471218689, + "batch_tokens.mean": 0.5870128720027967, + "decode_batch_size.mean": 0.071372838111468, + "graph_full_share": -0.0007180366007190075, + "graph_none_share": 0.00017950915017974842, + "graph_padding_fraction": -0.00012257265767716408, + "kv_usage_end_minus_start": 0.0005957119907479269, + "kv_usage_max": 0.003445359565031225, + "kv_usage_mean": 0.0015147564690633879, + "preemptions": 0.0, + "prefill_token_fraction": 0.0006457635751305757, + "queue_running_mean": 0.2144143154, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -5.000000000000014 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.2421875, + "request_hash": "adb62bea7f7a12c1e33fa1572ec1d0e274100013ad90e14c6ef5c549b0e0d017", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.2421875/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.2421875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.36245150019298356, + "batch_tokens.mean": -0.3978304588945094, + "decode_batch_size.mean": -0.35854104454872315, + "graph_full_share": 0.003837325045198847, + "graph_none_share": -0.010532837670384139, + "graph_padding_fraction": -0.04819081852048884, + "kv_usage_end_minus_start": -0.006582719323285691, + "kv_usage_max": -0.0005696857286345836, + "kv_usage_mean": -0.0009069497569939534, + "preemptions": 0.0, + "prefill_token_fraction": -0.0024232654720200003, + "queue_running_mean": -0.4996837289, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 21.599999999999994 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.4921875, + "request_hash": "cd227aee3be472a35e8c35af60a44a1411c4dc689cbd98746b3fb8a8d329b44c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.4921875/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.4921875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.11406418543540475, + "batch_tokens.mean": 9.343502830968255, + "decode_batch_size.mean": -0.11416962019156696, + "graph_full_share": -0.0001054347561618707, + "graph_none_share": -0.006673028798106863, + "graph_padding_fraction": 0.00039894330767695045, + "kv_usage_end_minus_start": 0.013684335607858866, + "kv_usage_max": 0.015105694785835122, + "kv_usage_mean": 0.003178130879397629, + "preemptions": 0.0, + "prefill_token_fraction": 0.010932399371689994, + "queue_running_mean": 1.5075751790999998, + "queue_waiting_mean": -1.0976031745, + "scheduler_steps_per_s": 7.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.728937728937729, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.27106227106227104, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 5.413566807825205, + "batch_tokens.mean": 200.71125302018655, + "decode_batch_size.mean": 5.3571681691824065, + "graph_full_share": -0.05071523213218676, + "graph_none_share": 0.05388272583201268, + "graph_padding_fraction": -0.227748799260124, + "kv_usage_end_minus_start": 0.0335338856501185, + "kv_usage_max": 0.060946820037647065, + "kv_usage_mean": 0.028804550592809862, + "preemptions": 0.0, + "prefill_token_fraction": 0.8741864436917179, + "queue_running_mean": 10.8362644165, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -89.4 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.7698209718670077, + "feasibility_transition": "true->false" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.23017902813299232, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 2.075957134599138, + "batch_tokens.mean": 46.93942520046838, + "decode_batch_size.mean": 2.0480560866794146, + "graph_full_share": -0.0074377102879925605, + "graph_none_share": -0.0018112558732417336, + "graph_padding_fraction": 0.0006496717251760695, + "kv_usage_end_minus_start": 0.018695600546584812, + "kv_usage_max": 0.017004326138624792, + "kv_usage_mean": 0.00908001267992682, + "preemptions": 0.0, + "prefill_token_fraction": 0.0002714485379902154, + "queue_running_mean": 8.942033995400001, + "queue_waiting_mean": -3.5751631934000008, + "scheduler_steps_per_s": -12.200000000000003 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.032423208191126277, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06313993174061433, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.35039583066624225, + "batch_tokens.mean": 122.17784504581891, + "decode_batch_size.mean": -0.430057526564223, + "graph_full_share": -0.016886297279774087, + "graph_none_share": 0.014833841973330465, + "graph_padding_fraction": -0.0008982073370737251, + "kv_usage_end_minus_start": 0.023420994136330364, + "kv_usage_max": 0.017803055174036908, + "kv_usage_mean": 0.0044696216831996635, + "preemptions": 0.0, + "prefill_token_fraction": 0.015941265487523948, + "queue_running_mean": 5.9603496991, + "queue_waiting_mean": -5.3947101023999995, + "scheduler_steps_per_s": -6.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.08361774744027303, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -1.3459024035954341, + "batch_tokens.mean": -76.90460019071148, + "decode_batch_size.mean": -1.335767993913132, + "graph_full_share": 0.014601482572194069, + "graph_none_share": -0.03533131042635292, + "graph_padding_fraction": 0.215661617603708, + "kv_usage_end_minus_start": -0.0019582245100192575, + "kv_usage_max": 0.004037693486432903, + "kv_usage_mean": -0.0013060500299530847, + "preemptions": 0.0, + "prefill_token_fraction": -0.7248107341009736, + "queue_running_mean": -1.1173600969999997, + "queue_waiting_mean": -5.5313452697, + "scheduler_steps_per_s": 55.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8216666666666667, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.17833333333333334, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.001077049918674966, + "batch_tokens.mean": -0.009036864794174093, + "decode_batch_size.mean": 0.0014499324412282633, + "graph_full_share": -0.0018530405936193484, + "graph_none_share": -0.0005575282018593251, + "graph_padding_fraction": -0.005954146779449193, + "kv_usage_end_minus_start": 2.5931683662605565e-05, + "kv_usage_max": 1.368725807104898e-05, + "kv_usage_mean": -2.4961532697604337e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.0012108546438647505, + "queue_running_mean": -0.027323256100000748, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.4000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.012380697130775609, + "batch_tokens.mean": 0.015975045041020763, + "decode_batch_size.mean": 0.012248179747174959, + "graph_full_share": 0.0004243534193951737, + "graph_none_share": 2.977918732597739e-06, + "graph_padding_fraction": 0.0021324803565644923, + "kv_usage_end_minus_start": -4.493314282427718e-06, + "kv_usage_max": 1.9874666237429395e-05, + "kv_usage_mean": 2.2389395203431008e-05, + "preemptions": 0.0, + "prefill_token_fraction": 9.913223620203193e-05, + "queue_running_mean": 0.015902196000000757, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.5 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.034252608017/result.json" + } + } + ], + "descriptive_full_outcome": { + "delta_pass_rate": { + "distinct_n": 10, + "max": 0.8216666666666667, + "min": -0.7698209718670077, + "n": 17 + }, + "feasibility_transitions": { + "false->false": 5, + "false->true": 2, + "true->false": 1, + "true->true": 9 + }, + "limitation": "Full outcomes may use different elapsed durations when a trial early-stopped; they are descriptive and are not a gate input.", + "negative": 4, + "positive": 5, + "zero": 8 + }, + "feature_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.058823529411764705, + "action_abs_median": 0.35039583066624225, + "action_delta": { + "distinct_n": 17, + "max": 6.1794369551023545, + "min": -1.3459024035954341, + "n": 17 + }, + "action_signs": { + "consistency": 0.5882352941176471, + "negative": 7, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 6.562183065127602, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.05339622914945741, + "repeat_abs_p95": 5.572276889098532, + "repeat_delta": { + "distinct_n": 29, + "max": 0.648283558063266, + "min": -5.684231334296602, + "n": 29 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.343502830968255, + "action_delta": { + "distinct_n": 17, + "max": 277.81382085064996, + "min": -76.90460019071148, + "n": 17 + }, + "action_signs": { + "consistency": 0.7058823529411765, + "negative": 5, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 8.491093823400865, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 1.1003885983709436, + "repeat_abs_p95": 315.3187671058407, + "repeat_delta": { + "distinct_n": 29, + "max": 49.961881401529, + "min": -364.45844383082266, + "n": 29 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.058823529411764705, + "action_abs_median": 0.35854104454872315, + "action_delta": { + "distinct_n": 17, + "max": 6.118283596420478, + "min": -1.335767993913132, + "n": 17 + }, + "action_signs": { + "consistency": 0.5882352941176471, + "negative": 7, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 6.76984750225626, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.05296146544353153, + "repeat_abs_p95": 5.508199821866692, + "repeat_delta": { + "distinct_n": 29, + "max": 0.6427639871873927, + "min": -5.614987163710968, + "n": 29 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.058823529411764705, + "action_abs_median": 0.003837325045198847, + "action_delta": { + "distinct_n": 17, + "max": 0.014601482572194069, + "min": -0.09344327526748608, + "n": 17 + }, + "action_signs": { + "consistency": 0.7058823529411765, + "negative": 12, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 4.142422601449927, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0009263480369809951, + "repeat_abs_p95": 0.05226163287124676, + "repeat_delta": { + "distinct_n": 28, + "max": 0.055525557955363625, + "min": -0.006350826071218729, + "n": 29 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.058823529411764705, + "action_abs_median": 0.006673028798106863, + "action_delta": { + "distinct_n": 17, + "max": 0.0681757891944469, + "min": -0.03533131042635292, + "n": 17 + }, + "action_signs": { + "consistency": 0.5294117647058824, + "negative": 9, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 6.753166603769785, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0009881332994778795, + "repeat_abs_p95": 0.0591340838533234, + "repeat_delta": { + "distinct_n": 22, + "max": 0.014663378024668453, + "min": -0.06263498920086392, + "n": 29 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.000877004096309171, + "action_delta": { + "distinct_n": 17, + "max": 0.215661617603708, + "min": -0.227748799260124, + "n": 17 + }, + "action_signs": { + "consistency": 0.5882352941176471, + "negative": 7, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 0.05354985977835987, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.016377336933075942, + "repeat_abs_p95": 0.24277823872412585, + "repeat_delta": { + "distinct_n": 29, + "max": 0.2448027632667215, + "min": -0.016377336933075942, + "n": 29 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.11764705882352941, + "action_abs_median": 0.006582719323285691, + "action_delta": { + "distinct_n": 16, + "max": 0.13999257948102506, + "min": -0.02063938485088157, + "n": 17 + }, + "action_signs": { + "consistency": 0.7647058823529411, + "negative": 4, + "positive": 13, + "zero": 0 + }, + "effect_to_repeat_median": 322.61413699448417, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 2.0404311431021505e-05, + "repeat_abs_p95": 0.050972874042718504, + "repeat_delta": { + "distinct_n": 19, + "max": 0.004504676952132969, + "min": -0.07211360414777734, + "n": 29 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.23529411764705882, + "action_abs_median": 0.015105694785835122, + "action_delta": { + "distinct_n": 16, + "max": 0.13096340508568438, + "min": -0.0005696857286345836, + "n": 17 + }, + "action_signs": { + "consistency": 0.9411764705882353, + "negative": 1, + "positive": 16, + "zero": 0 + }, + "effect_to_repeat_median": 2958.616486447546, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 5.1056616682254585e-06, + "repeat_abs_p95": 0.051096537054163664, + "repeat_delta": { + "distinct_n": 16, + "max": 0.004527588437235774, + "min": -0.0609037076826664, + "n": 29 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.17647058823529413, + "action_abs_median": 0.0025619580859006993, + "action_delta": { + "distinct_n": 17, + "max": 0.056370734075687004, + "min": -0.0013060500299530847, + "n": 17 + }, + "action_signs": { + "consistency": 0.7647058823529411, + "negative": 4, + "positive": 13, + "zero": 0 + }, + "effect_to_repeat_median": 7.0681027141206565, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0003624675799889579, + "repeat_abs_p95": 0.026328182817430976, + "repeat_delta": { + "distinct_n": 29, + "max": 0.00522340215966962, + "min": -0.034746555150565066, + "n": 29 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 17 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 17 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 29 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0018186060572318308, + "action_delta": { + "distinct_n": 17, + "max": 0.8741864436917179, + "min": -0.7248107341009736, + "n": 17 + }, + "action_signs": { + "consistency": 0.6470588235294118, + "negative": 6, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 0.01492870595090838, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.12181940371872435, + "repeat_abs_p95": 0.904349172685805, + "repeat_delta": { + "distinct_n": 29, + "max": 0.12181940371872435, + "min": -0.9109256423065466, + "n": 29 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4996837289, + "action_delta": { + "distinct_n": 17, + "max": 10.8362644165, + "min": -1.1173600969999997, + "n": 17 + }, + "action_signs": { + "consistency": 0.7647058823529411, + "negative": 4, + "positive": 13, + "zero": 0 + }, + "effect_to_repeat_median": 6.675982397855274, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.07484796979999953, + "repeat_abs_p95": 19.390367244779988, + "repeat_delta": { + "distinct_n": 29, + "max": 0.11568243179999982, + "min": -22.007979622, + "n": 29 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 9, + "max": 0.0, + "min": -5.5313452697, + "n": 17 + }, + "action_signs": { + "consistency": 1.0, + "negative": 8, + "positive": 0, + "zero": 9 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 5.569089138, + "repeat_delta": { + "distinct_n": 10, + "max": 0.0, + "min": -9.1442523314, + "n": 29 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.0, + "action_delta": { + "distinct_n": 17, + "max": 55.0, + "min": -89.4, + "n": 17 + }, + "action_signs": { + "consistency": 0.6470588235294118, + "negative": 11, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 2.8571428571428648, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.0999999999999943, + "repeat_abs_p95": 136.72000000000003, + "repeat_delta": { + "distinct_n": 24, + "max": 143.3, + "min": -28.89999999999999, + "n": 29 + } + } + }, + "horizon_s": 10.0, + "qualifying_features": [], + "repeats": [ + { + "delta_state": { + "batch_size.mean": -0.032046759505486744, + "batch_tokens.mean": -0.6567641003932749, + "decode_batch_size.mean": -0.031829677307312565, + "graph_full_share": 0.00020803710658368058, + "graph_none_share": -0.00019899201499314415, + "graph_padding_fraction": -0.0002750356645350475, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -4.977600796418624e-05, + "kv_usage_mean": -0.0014423226092831615, + "preemptions": 0.0, + "prefill_token_fraction": 6.688650356756476e-05, + "queue_running_mean": -0.1142690622, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.026614995471553993, + "batch_tokens.mean": -0.6496552246722729, + "decode_batch_size.mean": -0.02639849605068978, + "graph_full_share": 0.00020747861166137138, + "graph_none_share": -0.00018041618405337592, + "graph_padding_fraction": -9.631257646229227e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.00015576460470852899, + "preemptions": 0.0, + "prefill_token_fraction": 1.3376556165067832e-05, + "queue_running_mean": -0.04869389649999967, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/confirm-1-anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.010808600034953653, + "batch_tokens.mean": -0.5085699460020408, + "decode_batch_size.mean": -0.010635633606656203, + "graph_full_share": 0.0001657594937848339, + "graph_none_share": -0.00014413869024775677, + "graph_padding_fraction": -1.428081475827233e-05, + "kv_usage_end_minus_start": 9.965122072741472e-05, + "kv_usage_max": 0.0, + "kv_usage_mean": -4.734747194812261e-05, + "preemptions": 0.0, + "prefill_token_fraction": -9.363068207879355e-05, + "queue_running_mean": -0.009244884900000194, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.4000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.016288869514675675, + "batch_tokens.mean": -0.5180834059866299, + "decode_batch_size.mean": -0.016114501598372577, + "graph_full_share": 0.00016710258645746823, + "graph_none_share": -0.00013804126707352496, + "graph_padding_fraction": -0.00023422744992916602, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -4.995504046367838e-05, + "kv_usage_mean": -0.0007559762079877713, + "preemptions": 0.0, + "prefill_token_fraction": -4.013301897864974e-05, + "queue_running_mean": -0.07447232549999949, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.4000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/confirm-1-anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.04857934121216845, + "batch_tokens.mean": -1.5419494800797793, + "decode_batch_size.mean": -0.04806041200816624, + "graph_full_share": 0.0004973071538354468, + "graph_none_share": -0.0004108189531684403, + "graph_padding_fraction": -0.0002209229137654341, + "kv_usage_end_minus_start": 9.991008092724574e-05, + "kv_usage_max": -4.995504046367838e-05, + "kv_usage_mean": -0.0009120779275651192, + "preemptions": 0.0, + "prefill_token_fraction": -0.00012037560603939124, + "queue_running_mean": -0.07484796979999953, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.1999999999999886 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.648283558063266, + "batch_tokens.mean": 49.961881401529, + "decode_batch_size.mean": 0.6427639871873927, + "graph_full_share": -0.006350826071218729, + "graph_none_share": 0.014663378024668453, + "graph_padding_fraction": -0.016377336933075942, + "kv_usage_end_minus_start": 4.9753719090550597e-05, + "kv_usage_max": 0.004527588437235774, + "kv_usage_mean": 0.00522340215966962, + "preemptions": 0.0, + "prefill_token_fraction": 0.12181940371872435, + "queue_running_mean": 0.11568243179999982, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -28.89999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.2265625, + "request_hash": "a05848006335fc9f595432bab793a13d427244886e4f04e51ba74d495453ef90", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9920634920634921, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.2265625/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9920634920634921, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.2265625/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.003282275711160132, + "batch_tokens.mean": 0.03829321663019414, + "decode_batch_size.mean": 0.003282275711159688, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": 6.231700658618396e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -4.5836795611722825e-05, + "preemptions": 0.0, + "prefill_token_fraction": -2.6373628770337554e-05, + "queue_running_mean": -0.0024092132999999905, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.23046875, + "request_hash": "0f4c973153935c8abcd421eb631900670e31af0dc8a33bdf7e4d3d85e7ed7711", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9923076923076923, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.23046875/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9923076923076923, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.23046875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.016918595496772326, + "batch_tokens.mean": -0.47955440088175294, + "decode_batch_size.mean": -0.0167453944260747, + "graph_full_share": 0.00017320107069751511, + "graph_none_share": -0.000173201070697529, + "graph_padding_fraction": 3.159090949033175e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -1.8095624633056306e-05, + "preemptions": 0.0, + "prefill_token_fraction": -4.685685224570246e-05, + "queue_running_mean": -0.005980747899999805, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.234375, + "request_hash": "14d2a74fffe25e7d9b3e7490c392eae0dc31bb1f3fe6566c9c0a567543bdd6ac", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9924242424242424, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.234375/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9924242424242424, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.234375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.05339622914945741, + "batch_tokens.mean": -1.3045555990319286, + "decode_batch_size.mean": -0.05296146544353153, + "graph_full_share": 0.0004166485515122442, + "graph_none_share": -0.00041664855151225116, + "graph_padding_fraction": -5.497908988029605e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0003624675799889579, + "preemptions": 0.0, + "prefill_token_fraction": 2.675497368564983e-05, + "queue_running_mean": -0.03772782840000044, + "queue_waiting_mean": -0.0204415657, + "scheduler_steps_per_s": 1.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.006993006993006978, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.958041958041958, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.951048951048951, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -1.5223872504055418, + "batch_tokens.mean": -69.8055007051478, + "decode_batch_size.mean": -1.5056685448852059, + "graph_full_share": 0.017265452376541424, + "graph_none_share": -0.025711662075298437, + "graph_padding_fraction": 0.15174913486120387, + "kv_usage_end_minus_start": -0.009853977330746444, + "kv_usage_max": -0.01946543449402638, + "kv_usage_mean": -0.008432216411693306, + "preemptions": 0.0, + "prefill_token_fraction": -0.8614186464282653, + "queue_running_mean": -3.6705211939, + "queue_waiting_mean": -0.0093291794, + "scheduler_steps_per_s": 74.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/confirm-1-anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.007251432291120796, + "batch_tokens.mean": -0.00762511250828668, + "decode_batch_size.mean": -0.007211010152244679, + "graph_full_share": 3.9523869123314626e-05, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.0023343164243360137, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -1.4972595796737716e-05, + "preemptions": 0.0, + "prefill_token_fraction": 5.034578599764916e-05, + "queue_running_mean": -0.006232385800000184, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.30000000000001137 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.5/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/confirm-1-anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.001311475409836227, + "batch_tokens.mean": -0.001311475409836227, + "decode_batch_size.mean": -0.001311475409836227, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.00024782551484880955, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -2.119802975337863e-05, + "preemptions": 0.0, + "prefill_token_fraction": 1.920108243574914e-05, + "queue_running_mean": -0.0037054131000004986, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/confirm-1-anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -5.253278252178259, + "batch_tokens.mean": -203.60648072702782, + "decode_batch_size.mean": -5.198083023705538, + "graph_full_share": 0.03511966842665615, + "graph_none_share": -0.043478260869565216, + "graph_padding_fraction": 0.24157911714869673, + "kv_usage_end_minus_start": -0.07211360414777734, + "kv_usage_max": -0.05893470570957904, + "kv_usage_mean": -0.034746555150565066, + "preemptions": 0.0, + "prefill_token_fraction": -0.8735909388483541, + "queue_running_mean": -12.366947709400002, + "queue_waiting_mean": -1.6375549495, + "scheduler_steps_per_s": 86.29999999999998 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.7385786802030457, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.75390625, + "request_hash": "2677a6a12ff0e104d6246dd121d0667ff4fe3d25f8d68699547906420246ebdd", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.20558375634517767, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75390625/result.json" + }, + "target": { + "cell": "tp2_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.9441624365482234, + "mns": 32, + "trial_id": "tp2_mns32/confirm-1-anchor-0.75390625/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.0022141886965014734, + "batch_tokens.mean": -0.0023373966282491665, + "decode_batch_size.mean": -0.002200860915423153, + "graph_full_share": 1.3031608165570141e-05, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.0003699622628544341, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -6.601132195046752e-06, + "preemptions": 0.0, + "prefill_token_fraction": 1.2576053067120396e-05, + "queue_running_mean": 0.0006418031000001712, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.10000000000002274 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.5/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp2_mns64/confirm-1-anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -5.415534020939959, + "batch_tokens.mean": -200.71322023330129, + "decode_batch_size.mean": -5.35913538229716, + "graph_full_share": 0.05071523213218676, + "graph_none_share": -0.05388272583201268, + "graph_padding_fraction": 0.22642208124358779, + "kv_usage_end_minus_start": -0.03352133931883006, + "kv_usage_max": -0.0609037076826664, + "kv_usage_mean": -0.02880160581001033, + "preemptions": 0.0, + "prefill_token_fraction": -0.8741576384479508, + "queue_running_mean": -10.836573950600002, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 89.4 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.6624040920716113, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.23017902813299232, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.8925831202046036, + "mns": 64, + "trial_id": "tp2_mns64/confirm-1-anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.04236326637336152, + "batch_tokens.mean": -0.0506303786819462, + "decode_batch_size.mean": -0.041385582574491764, + "graph_full_share": 0.0009697259074838982, + "graph_none_share": -0.000746904948671601, + "graph_padding_fraction": -0.0024530500387316967, + "kv_usage_end_minus_start": -0.005895639523754825, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0001257250352028564, + "preemptions": 0.0, + "prefill_token_fraction": -0.0014409663380382937, + "queue_running_mean": -0.0696698146000001, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.0999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.4921875, + "request_hash": "cd227aee3be472a35e8c35af60a44a1411c4dc689cbd98746b3fb8a8d329b44c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.4921875/result.json" + }, + "target": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/confirm-1-anchor-0.4921875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.9687249841095076, + "batch_tokens.mean": -59.73987529620144, + "decode_batch_size.mean": -0.9579959096941315, + "graph_full_share": 0.011406581190443799, + "graph_none_share": -0.019512062147118037, + "graph_padding_fraction": 0.20562784534661108, + "kv_usage_end_minus_start": 0.004504676952132969, + "kv_usage_max": -0.0037772928550463547, + "kv_usage_mean": -0.0036990145914360204, + "preemptions": 0.0, + "prefill_token_fraction": -0.8491799860125482, + "queue_running_mean": -1.3942294618000002, + "queue_waiting_mean": -1.1069323539, + "scheduler_steps_per_s": 45.69999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.6190476190476191, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.27106227106227104, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.8901098901098901, + "mns": 8, + "trial_id": "tp2_mns8/confirm-1-anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.02348440960742426, + "batch_tokens.mean": -0.024462377280626946, + "decode_batch_size.mean": -0.02333965019280182, + "graph_full_share": -0.0009263480369809951, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.0002862441287893591, + "kv_usage_end_minus_start": -5.101077857672109e-06, + "kv_usage_max": -1.0202155715566263e-05, + "kv_usage_mean": -2.2000789048993996e-05, + "preemptions": 0.0, + "prefill_token_fraction": 5.985572526742006e-05, + "queue_running_mean": -0.02092175199999957, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.2469565217391304, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033182214016, + "request_hash": "e6b624a0212807df3733c54b4a2160fba24f19ca4776b14f34918d83428bba7f", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7530434782608696, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033182214016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.033182214016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -3.19743814378252, + "batch_tokens.mean": -194.8952745035574, + "decode_batch_size.mean": -3.1597031963470315, + "graph_full_share": 0.02834766910905817, + "graph_none_share": -0.04961240310077519, + "graph_padding_fraction": 0.24357765310774523, + "kv_usage_end_minus_start": -0.008034197625958361, + "kv_usage_max": -0.004550161449114243, + "kv_usage_mean": -0.008735063035535659, + "preemptions": 0.0, + "prefill_token_fraction": -0.8943849310253909, + "queue_running_mean": -6.701412029300001, + "queue_waiting_mean": -9.1442523314, + "scheduler_steps_per_s": 110.69999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.34982935153583616, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06313993174061433, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.4129692832764505, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.07980086487246574, + "batch_tokens.mean": -1.0880711819756304, + "decode_batch_size.mean": -0.07880182167344785, + "graph_full_share": -0.0018140521245323393, + "graph_none_share": -0.0016474464579901153, + "graph_padding_fraction": 0.02197042071969288, + "kv_usage_end_minus_start": 2.0404311431021505e-05, + "kv_usage_max": 0.0, + "kv_usage_mean": -8.46355932073951e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.13021608847689226, + "queue_running_mean": -0.09728112210000006, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 3.8000000000000114 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.006431552476177771, + "batch_tokens.mean": -0.006758232721721491, + "decode_batch_size.mean": -0.006383197052567624, + "graph_full_share": 0.000573798951080895, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.0006374012583605093, + "kv_usage_end_minus_start": -5.101077857672109e-06, + "kv_usage_max": -5.101077857783132e-06, + "kv_usage_mean": -1.0214925773703067e-05, + "preemptions": 0.0, + "prefill_token_fraction": 5.436516758668131e-06, + "queue_running_mean": -0.009119564499999733, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.19999999999998863 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.2469565217391304, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033182214016, + "request_hash": "e6b624a0212807df3733c54b4a2160fba24f19ca4776b14f34918d83428bba7f", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7530434782608696, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033182214016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-2-anchor-0.033182214016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -5.684231334296602, + "batch_tokens.mean": -242.26991192050903, + "decode_batch_size.mean": -5.614987163710968, + "graph_full_share": 0.03905791041999285, + "graph_none_share": -0.04780114722753346, + "graph_padding_fraction": 0.23825248667744758, + "kv_usage_end_minus_start": -0.028101561821904375, + "kv_usage_max": -0.02156120922490945, + "kv_usage_mean": -0.01815563294515909, + "preemptions": 0.0, + "prefill_token_fraction": -0.89493540395689, + "queue_running_mean": -16.051436879700002, + "queue_waiting_mean": -5.569089138, + "scheduler_steps_per_s": 136.8 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.9692832764505119, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.07573170158789821, + "batch_tokens.mean": -1.0733751057032785, + "decode_batch_size.mean": -0.07518292539697491, + "graph_full_share": 0.0010444071220520046, + "graph_none_share": -0.0010899182561307902, + "graph_padding_fraction": 0.027146526691836648, + "kv_usage_end_minus_start": -5.105661668114436e-06, + "kv_usage_max": -5.1056616682254585e-06, + "kv_usage_mean": -6.434458670689751e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.1289638873088742, + "queue_running_mean": -0.0711253370999998, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.0999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -5.676772134537581, + "batch_tokens.mean": -242.26212742181514, + "decode_batch_size.mean": -5.607576114913048, + "graph_full_share": 0.039014238618063524, + "graph_none_share": -0.04780114722753346, + "graph_padding_fraction": 0.23544552622699638, + "kv_usage_end_minus_start": -0.02810666748357249, + "kv_usage_max": -0.021556103563241225, + "kv_usage_mean": -0.01814814993745567, + "preemptions": 0.0, + "prefill_token_fraction": -0.8949517370921197, + "queue_running_mean": -16.0413130873, + "queue_waiting_mean": -5.569089138, + "scheduler_steps_per_s": 136.60000000000002 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.9692832764505119, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-2-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -4.9280804627990085, + "batch_tokens.mean": -364.01800389606234, + "decode_batch_size.mean": -4.782726800328038, + "graph_full_share": 0.05329256669728677, + "graph_none_share": -0.06263498920086392, + "graph_padding_fraction": 0.2448027632667215, + "kv_usage_end_minus_start": -0.05014161119802052, + "kv_usage_max": -0.039318834800568414, + "kv_usage_mean": -0.022291899962727596, + "preemptions": 0.0, + "prefill_token_fraction": -0.9106141297482617, + "queue_running_mean": -21.6163208215, + "queue_waiting_mean": -0.1743790356, + "scheduler_steps_per_s": 129.10000000000002 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8430034129692833, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9573378839590444, + "mns": 64, + "trial_id": "tp4_mns64/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.0986378480703829, + "batch_tokens.mean": -1.1003885983709436, + "decode_batch_size.mean": -0.09787917001031765, + "graph_full_share": 0.00015255866453844913, + "graph_none_share": -0.001092896174863388, + "graph_padding_fraction": 0.026001990897698585, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -8.10913166477481e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.1290526738911407, + "queue_running_mean": -0.09155397830000034, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.9000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -5.343712123283302, + "batch_tokens.mean": -364.45844383082266, + "decode_batch_size.mean": -5.194686323828272, + "graph_full_share": 0.055525557955363625, + "graph_none_share": -0.06263498920086392, + "graph_padding_fraction": 0.2382752171480348, + "kv_usage_end_minus_start": -0.05152704927251717, + "kv_usage_max": -0.0393392840710407, + "kv_usage_mean": -0.022618048328561978, + "preemptions": 0.0, + "prefill_token_fraction": -0.9109256423065466, + "queue_running_mean": -22.007979622, + "queue_waiting_mean": -0.1743790356, + "scheduler_steps_per_s": 143.3 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8856655290102389, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/confirm-2-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.5747954473735191, + "batch_tokens.mean": -30.476142999621462, + "decode_batch_size.mean": -0.5686361081847746, + "graph_full_share": 0.006538565006766528, + "graph_none_share": -0.017039834454924494, + "graph_padding_fraction": 0.1583386005746088, + "kv_usage_end_minus_start": -0.002356121069944095, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0006074327722416259, + "preemptions": 0.0, + "prefill_token_fraction": -0.8386345381065662, + "queue_running_mean": -0.9726716715000001, + "queue_waiting_mean": -0.428824454, + "scheduler_steps_per_s": 40.900000000000006 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.016591107009, + "request_hash": "2407501cda53bf9c761c448dc50ab4780b4a55958cc00ba1b19cd09de1cf73e1", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.016591107009/result.json" + }, + "target": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/confirm-1-anchor-0.016591107009/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.03238588118373942, + "batch_tokens.mean": -2.9133132020797436, + "decode_batch_size.mean": -0.0321242294926507, + "graph_full_share": 0.00025027553060685825, + "graph_none_share": -0.0009881332994778795, + "graph_padding_fraction": 0.06450969515475685, + "kv_usage_end_minus_start": 5.099829155752467e-06, + "kv_usage_max": 0.0, + "kv_usage_mean": -4.0493927196364344e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.4656521349077925, + "queue_running_mean": -0.06308746549999977, + "queue_waiting_mean": -0.0015971439000000004, + "scheduler_steps_per_s": 2.5 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.017126304009, + "request_hash": "93cd99ebf2a760cc440dce0cb7e51273686dcfbfb07650d27e9e16d2ff1d2eed", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.017126304009/result.json" + }, + "target": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/confirm-1-anchor-0.017126304009/result.json" + } + } + ], + "sanity": { + "action_pairs": 17, + "distinct_action_vectors": 17, + "invariants": { + "all_pair_hashes_match": true, + "all_values_finite": true, + "expected_action_pair_count": true, + "nonnegative_counters": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true, + "sufficient_repeat_pair_count": true + }, + "red_flags": [], + "repeat_pairs": 29, + "trials": 66 + } + }, + "5": { + "actions": [ + { + "delta_state": { + "batch_size.mean": 3.5280207937111703, + "batch_tokens.mean": 189.10013313046787, + "decode_batch_size.mean": 3.495993406872068, + "graph_full_share": -0.027843286420692226, + "graph_none_share": 0.04184100418410042, + "graph_padding_fraction": -0.10888597553063217, + "kv_usage_end_minus_start": 0.04832989960722345, + "kv_usage_max": 0.05576515049492037, + "kv_usage_mean": 0.04101485276318234, + "preemptions": 0.0, + "prefill_token_fraction": 0.8861120723274618, + "queue_running_mean": 3.4413866424000004, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -84.2 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.05698177434224494, + "batch_tokens.mean": 2.4183568541238287, + "decode_batch_size.mean": 0.056343521736047286, + "graph_full_share": -0.0005850648890149968, + "graph_none_share": 0.000531877171831785, + "graph_padding_fraction": 0.00021749194258890905, + "kv_usage_end_minus_start": 7.313298449929118e-05, + "kv_usage_max": 0.00030157213266412963, + "kv_usage_mean": 0.0016340770443056124, + "preemptions": 0.0, + "prefill_token_fraction": 8.473923694440888e-05, + "queue_running_mean": 0.1474536237999997, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.1278554259770317, + "batch_tokens.mean": 3.2897534051455466, + "decode_batch_size.mean": 0.127000801210718, + "graph_full_share": -0.0007834060357874151, + "graph_none_share": -0.0034719131131487613, + "graph_padding_fraction": 0.0008694893399684026, + "kv_usage_end_minus_start": 2.1174429976178466e-05, + "kv_usage_max": 0.03040170172083234, + "kv_usage_mean": 0.005773282024914478, + "preemptions": 0.0, + "prefill_token_fraction": -0.0001483422608830054, + "queue_running_mean": 0.5028872580000003, + "queue_waiting_mean": -0.0510258954, + "scheduler_steps_per_s": -0.7999999999999972 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.04195804195804198, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.958041958041958, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.06163060969068468, + "batch_tokens.mean": -2.443018058287123, + "decode_batch_size.mean": -0.06098694797067772, + "graph_full_share": 0.0005900232433398989, + "graph_none_share": -0.0005363847666726221, + "graph_padding_fraction": -0.00017399768727638983, + "kv_usage_end_minus_start": 4.7122279587474836e-05, + "kv_usage_max": 0.00011557359098801623, + "kv_usage_mean": -0.0004086366142161957, + "preemptions": 0.0, + "prefill_token_fraction": -6.356136875318441e-05, + "queue_running_mean": -0.0843407528000002, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.6000000000000014 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.024589582668510523, + "batch_tokens.mean": 0.8116866999964429, + "decode_batch_size.mean": 0.02437683934333279, + "graph_full_share": -0.00019501471474669962, + "graph_none_share": 0.00017728610431513858, + "graph_padding_fraction": 8.69181567518763e-05, + "kv_usage_end_minus_start": 0.00012295859486299854, + "kv_usage_max": 0.00030157213266412963, + "kv_usage_mean": 0.00032121484848791904, + "preemptions": 0.0, + "prefill_token_fraction": 0.0, + "queue_running_mean": 0.013309304399999888, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.20000000000000284 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 4.473159643294283, + "batch_tokens.mean": 189.17581745060323, + "decode_batch_size.mean": 4.408288162266131, + "graph_full_share": -0.046424200034971164, + "graph_none_share": 0.03890540304248993, + "graph_padding_fraction": 0.005662578660898298, + "kv_usage_end_minus_start": 0.08863131728902607, + "kv_usage_max": 0.08590208433411406, + "kv_usage_mean": 0.0354576817886873, + "preemptions": 0.0, + "prefill_token_fraction": 0.0007191134183018866, + "queue_running_mean": 3.8361017661999997, + "queue_waiting_mean": -1.0501423682, + "scheduler_steps_per_s": -7.799999999999997 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.06521739130434782, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.15579710144927536, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -1.4603817235396175, + "batch_tokens.mean": 220.88027761711976, + "decode_batch_size.mean": -1.5222672064777338, + "graph_full_share": -0.06998264893001727, + "graph_none_share": 0.055523423944476596, + "graph_padding_fraction": 0.0002698844942520852, + "kv_usage_end_minus_start": 0.03428564980259485, + "kv_usage_max": 0.03430172098005413, + "kv_usage_mean": 0.004527112114386889, + "preemptions": 0.0, + "prefill_token_fraction": 0.009687903063064107, + "queue_running_mean": 0.49739668660000014, + "queue_waiting_mean": -0.4414553072, + "scheduler_steps_per_s": -8.400000000000002 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.06521739130434782, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.15579710144927536, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.0, + "batch_tokens.mean": 0.0, + "decode_batch_size.mean": 0.0, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.0, + "kv_usage_end_minus_start": 0.0004933875406503141, + "kv_usage_max": 0.0005353228930036202, + "kv_usage_mean": 0.0001535070867022259, + "preemptions": 0.0, + "prefill_token_fraction": 0.0, + "queue_running_mean": -0.00963330859999978, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.02898550724637681, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.09057971014492754, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.5/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06159420289855073, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.06154364887467256, + "batch_tokens.mean": 0.0673666044072565, + "decode_batch_size.mean": 0.06098464514354429, + "graph_full_share": -0.0005590037311280494, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.005655833585839015, + "kv_usage_end_minus_start": -2.459171897339907e-06, + "kv_usage_max": 0.003445359565031225, + "kv_usage_mean": 0.0015597623110467189, + "preemptions": 0.0, + "prefill_token_fraction": 0.00030636946592198233, + "queue_running_mean": 0.12847152480000013, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -4.200000000000017 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.2421875, + "request_hash": "adb62bea7f7a12c1e33fa1572ec1d0e274100013ad90e14c6ef5c549b0e0d017", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "mns_increase", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.2421875/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.2421875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.23325211580663785, + "batch_tokens.mean": -0.2519983270912536, + "decode_batch_size.mean": -0.23134894714830168, + "graph_full_share": 0.001903168658336507, + "graph_none_share": -0.006795016987542469, + "graph_padding_fraction": -0.004536339467287626, + "kv_usage_end_minus_start": 5.075521287700724e-07, + "kv_usage_max": -2.1953909969063545e-05, + "kv_usage_mean": -0.0005894447334146543, + "preemptions": 0.0, + "prefill_token_fraction": -6.083338940097993e-05, + "queue_running_mean": -0.27085744499999986, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 16.200000000000017 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.4921875, + "request_hash": "cd227aee3be472a35e8c35af60a44a1411c4dc689cbd98746b3fb8a8d329b44c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.4921875/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.4921875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.6005778377124296, + "batch_tokens.mean": -4.162510512731277, + "decode_batch_size.mean": -0.5951764688823156, + "graph_full_share": 0.005401368830114217, + "graph_none_share": -0.01149694043268952, + "graph_padding_fraction": -0.004587093073780198, + "kv_usage_end_minus_start": 1.4354961987761428e-05, + "kv_usage_max": -0.0029570125473361575, + "kv_usage_mean": -0.0013514937493694472, + "preemptions": 0.0, + "prefill_token_fraction": -0.0010977628546262608, + "queue_running_mean": -0.5356197792000001, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 29.799999999999983 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.728937728937729, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.27106227106227104, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.35712385038833183, + "batch_tokens.mean": 111.90975236574697, + "decode_batch_size.mean": 0.32775366295797426, + "graph_full_share": -0.025012888955412538, + "graph_none_share": 0.026143790849673203, + "graph_padding_fraction": -0.2241759801940655, + "kv_usage_end_minus_start": 0.03485111013888609, + "kv_usage_max": 0.01867744710795105, + "kv_usage_mean": 0.006535245430020315, + "preemptions": 0.0, + "prefill_token_fraction": 0.8766852234122235, + "queue_running_mean": 2.085466555, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -65.39999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.7698209718670077, + "feasibility_transition": "true->false" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "mns_increase", + "source": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.23017902813299232, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.8613841701826859, + "batch_tokens.mean": 99.28363780953853, + "decode_batch_size.mean": -0.9000309262409161, + "graph_full_share": -0.013132607303010913, + "graph_none_share": 0.007356027303452698, + "graph_padding_fraction": -0.002483113089604908, + "kv_usage_end_minus_start": 0.013972653351394082, + "kv_usage_max": 0.013923980316157603, + "kv_usage_mean": 0.002289615839773029, + "preemptions": 0.0, + "prefill_token_fraction": 0.027119159852510122, + "queue_running_mean": 2.255338312400001, + "queue_waiting_mean": -1.3976082232, + "scheduler_steps_per_s": -8.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.032423208191126277, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06313993174061433, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.00950293278090264, + "batch_tokens.mean": -1.5782094981150578, + "decode_batch_size.mean": -0.008726077059269421, + "graph_full_share": 0.0005239259517987316, + "graph_none_share": -0.00023486335770291772, + "graph_padding_fraction": 8.79877867886172e-05, + "kv_usage_end_minus_start": 3.819849616093496e-05, + "kv_usage_max": 4.159303058193231e-05, + "kv_usage_mean": 6.884522888032929e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.00013165055575314888, + "queue_running_mean": 0.02412652539999982, + "queue_waiting_mean": -9.661300000000927e-05, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.08361774744027303, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.7391144064912778, + "batch_tokens.mean": -97.02946180778741, + "decode_batch_size.mean": -0.7271776858566685, + "graph_full_share": 0.015299949334160123, + "graph_none_share": -0.04103743978705223, + "graph_padding_fraction": 0.21027653696749515, + "kv_usage_end_minus_start": -0.006308261401860582, + "kv_usage_max": 0.004195788190260341, + "kv_usage_mean": -0.0005148594359863203, + "preemptions": 0.0, + "prefill_token_fraction": -0.7248910361224853, + "queue_running_mean": -0.7297395955999999, + "queue_waiting_mean": -3.4976107514, + "scheduler_steps_per_s": 59.80000000000001 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8216666666666667, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.17833333333333334, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.0033632286995510796, + "batch_tokens.mean": -0.0033632286995510796, + "decode_batch_size.mean": -0.003363228699551968, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.0030699089127894374, + "kv_usage_end_minus_start": 8.342535059435718e-07, + "kv_usage_max": 1.368725807104898e-05, + "kv_usage_mean": -6.535810642852545e-06, + "preemptions": 0.0, + "prefill_token_fraction": 9.850983893985865e-05, + "queue_running_mean": -0.009463884000000533, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.01835841147659245, + "batch_tokens.mean": 0.02343919454014376, + "decode_batch_size.mean": 0.01819951878212578, + "graph_full_share": -0.0012724026088667406, + "graph_none_share": 3.7831593920716316e-06, + "graph_padding_fraction": 0.0006955531292226758, + "kv_usage_end_minus_start": 1.2113828717907893e-06, + "kv_usage_max": 1.9874666237429395e-05, + "kv_usage_mean": 2.5481750126793087e-05, + "preemptions": 0.0, + "prefill_token_fraction": 3.285684502737918e-05, + "queue_running_mean": 0.018367366400000584, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "mns_increase", + "source": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.034252608017/result.json" + } + } + ], + "descriptive_full_outcome": { + "delta_pass_rate": { + "distinct_n": 10, + "max": 0.8216666666666667, + "min": -0.7698209718670077, + "n": 17 + }, + "feasibility_transitions": { + "false->false": 5, + "false->true": 2, + "true->false": 1, + "true->true": 9 + }, + "limitation": "Full outcomes may use different elapsed durations when a trial early-stopped; they are descriptive and are not a gate input.", + "negative": 4, + "positive": 5, + "zero": 8 + }, + "feature_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.35294117647058826, + "action_abs_median": 0.1278554259770317, + "action_delta": { + "distinct_n": 17, + "max": 4.473159643294283, + "min": -1.4603817235396175, + "n": 17 + }, + "action_signs": { + "consistency": 0.5, + "negative": 8, + "positive": 8, + "zero": 1 + }, + "effect_to_repeat_median": 1.9564724489442589, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.06534997517906493, + "repeat_abs_p95": 0.4689572472409891, + "repeat_delta": { + "distinct_n": 29, + "max": 0.9280844155844155, + "min": -0.373100622158411, + "n": 29 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.058823529411764705, + "action_abs_median": 2.443018058287123, + "action_delta": { + "distinct_n": 17, + "max": 220.88027761711976, + "min": -97.02946180778741, + "n": 17 + }, + "action_signs": { + "consistency": 0.625, + "negative": 6, + "positive": 10, + "zero": 1 + }, + "effect_to_repeat_median": 1.016963189345596, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 2.402267932489451, + "repeat_abs_p95": 212.7553678215874, + "repeat_delta": { + "distinct_n": 29, + "max": 100.79992784992785, + "min": -213.38441949985196, + "n": 29 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.35294117647058826, + "action_abs_median": 0.127000801210718, + "action_delta": { + "distinct_n": 17, + "max": 4.408288162266131, + "min": -1.5222672064777338, + "n": 17 + }, + "action_signs": { + "consistency": 0.5, + "negative": 8, + "positive": 8, + "zero": 1 + }, + "effect_to_repeat_median": 1.9625625182162536, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.06471172257286728, + "repeat_abs_p95": 0.5074569452309783, + "repeat_delta": { + "distinct_n": 29, + "max": 0.9185245310245311, + "min": -0.36846434553720986, + "n": 29 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.11764705882352941, + "action_abs_median": 0.0012724026088667406, + "action_delta": { + "distinct_n": 16, + "max": 0.015299949334160123, + "min": -0.06998264893001727, + "n": 17 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 10, + "positive": 5, + "zero": 2 + }, + "effect_to_repeat_median": 1.8902399214885985, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0006731434430105043, + "repeat_abs_p95": 0.029715971438244844, + "repeat_delta": { + "distinct_n": 28, + "max": 0.029898516783762652, + "min": -0.009559884559884546, + "n": 29 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.23529411764705882, + "action_abs_median": 0.0034719131131487613, + "action_delta": { + "distinct_n": 15, + "max": 0.055523423944476596, + "min": -0.04103743978705223, + "n": 17 + }, + "action_signs": { + "consistency": 0.5714285714285714, + "negative": 6, + "positive": 8, + "zero": 3 + }, + "effect_to_repeat_median": 3.2771966527196663, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0010594155557517443, + "repeat_abs_p95": 0.031925759090416374, + "repeat_delta": { + "distinct_n": 21, + "max": 0.02168109668109668, + "min": -0.03201970443349754, + "n": 29 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.002483113089604908, + "action_delta": { + "distinct_n": 17, + "max": 0.21027653696749515, + "min": -0.2241759801940655, + "n": 17 + }, + "action_signs": { + "consistency": 0.5625, + "negative": 7, + "positive": 9, + "zero": 1 + }, + "effect_to_repeat_median": 0.07674048944322485, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.032357274596769375, + "repeat_abs_p95": 0.24385677131445133, + "repeat_delta": { + "distinct_n": 29, + "max": 0.26242965569766435, + "min": -0.18173061839089436, + "n": 29 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.23529411764705882, + "action_abs_median": 7.313298449929118e-05, + "action_delta": { + "distinct_n": 17, + "max": 0.08863131728902607, + "min": -0.006308261401860582, + "n": 17 + }, + "action_signs": { + "consistency": 0.8823529411764706, + "negative": 2, + "positive": 15, + "zero": 0 + }, + "effect_to_repeat_median": 1.4692416585900228, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 4.977600796418624e-05, + "repeat_abs_p95": 0.03241204552355029, + "repeat_delta": { + "distinct_n": 14, + "max": 0.048360614955967995, + "min": -0.03483866016958903, + "n": 29 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.29411764705882354, + "action_abs_median": 0.0029570125473361575, + "action_delta": { + "distinct_n": 16, + "max": 0.08590208433411406, + "min": -0.0029570125473361575, + "n": 17 + }, + "action_signs": { + "consistency": 0.8823529411764706, + "negative": 2, + "positive": 15, + "zero": 0 + }, + "effect_to_repeat_median": null, + "effect_to_repeat_median_is_infinite": true, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.016674847306688556, + "repeat_delta": { + "distinct_n": 14, + "max": 0.004527588437235774, + "min": -0.018634334752970383, + "n": 29 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.11764705882352941, + "action_abs_median": 0.0013514937493694472, + "action_delta": { + "distinct_n": 17, + "max": 0.04101485276318234, + "min": -0.0013514937493694472, + "n": 17 + }, + "action_signs": { + "consistency": 0.7058823529411765, + "negative": 5, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 2.827343009345717, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0004780084145793828, + "repeat_abs_p95": 0.006592699934625479, + "repeat_delta": { + "distinct_n": 29, + "max": 0.004456117360644808, + "min": -0.006626073152505554, + "n": 29 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 17 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 17 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 29 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0001483422608830054, + "action_delta": { + "distinct_n": 16, + "max": 0.8861120723274618, + "min": -0.7248910361224853, + "n": 17 + }, + "action_signs": { + "consistency": 0.6, + "negative": 6, + "positive": 9, + "zero": 2 + }, + "effect_to_repeat_median": 0.0009890369063830554, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.1499865777764538, + "repeat_abs_p95": 0.9129144795909149, + "repeat_delta": { + "distinct_n": 28, + "max": 0.8754134325195487, + "min": -0.9129695381069503, + "n": 29 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.27085744499999986, + "action_delta": { + "distinct_n": 17, + "max": 3.8361017661999997, + "min": -0.7297395955999999, + "n": 17 + }, + "action_signs": { + "consistency": 0.6470588235294118, + "negative": 6, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 2.6248225517066213, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.10319076419999984, + "repeat_abs_p95": 6.572392109320001, + "repeat_delta": { + "distinct_n": 29, + "max": 0.07105187159999993, + "min": -6.5953128982, + "n": 29 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.23529411764705882, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 7, + "max": 0.0, + "min": -3.4976107514, + "n": 17 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 11 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.10560919299999999, + "repeat_delta": { + "distinct_n": 5, + "max": 0.0, + "min": -1.5032174162, + "n": 29 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 4.200000000000017, + "action_delta": { + "distinct_n": 15, + "max": 59.80000000000001, + "min": -84.2, + "n": 17 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 10, + "positive": 5, + "zero": 2 + }, + "effect_to_repeat_median": 2.333333333333365, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.799999999999983, + "repeat_abs_p95": 101.72, + "repeat_delta": { + "distinct_n": 24, + "max": 101.8, + "min": -31.799999999999997, + "n": 29 + } + } + }, + "horizon_s": 5.0, + "qualifying_features": [], + "repeats": [ + { + "delta_state": { + "batch_size.mean": -0.11342809216915306, + "batch_tokens.mean": -4.816915335040164, + "decode_batch_size.mean": -0.11215679350225116, + "graph_full_share": 0.0011653571113269257, + "graph_none_share": -0.0010594155557517443, + "graph_padding_fraction": -0.0002394644309091495, + "kv_usage_end_minus_start": 4.977600796418624e-05, + "kv_usage_max": -4.977600796418624e-05, + "kv_usage_mean": -0.002650866963514159, + "preemptions": 0.0, + "prefill_token_fraction": -0.00016947847485904166, + "queue_running_mean": -0.21751143979999998, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.2000000000000028 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp1_mns16/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.01603319151928506, + "batch_tokens.mean": 0.7965437220913429, + "decode_batch_size.mean": 0.015822228472979205, + "graph_full_share": -0.00019338279244751622, + "graph_none_share": 0.00017580253858866246, + "graph_padding_fraction": -2.1621324535013326e-05, + "kv_usage_end_minus_start": -4.982561036370736e-05, + "kv_usage_max": 0.0, + "kv_usage_mean": 0.0010437773534030942, + "preemptions": 0.0, + "prefill_token_fraction": 4.236776766430861e-05, + "queue_running_mean": 0.07105187159999993, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.19999999999999574 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/confirm-1-anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.012605042016806678, + "batch_tokens.mean": 0.012605042016787138, + "decode_batch_size.mean": 0.012605042016806678, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.00010847307953028382, + "kv_usage_end_minus_start": 4.982561036370736e-05, + "kv_usage_max": 0.0, + "kv_usage_mean": 0.00017122409391133464, + "preemptions": 0.0, + "prefill_token_fraction": -6.355303957206182e-05, + "queue_running_mean": 0.01412161399999956, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp1_mns32/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.06534997517906493, + "batch_tokens.mean": -2.426725054960656, + "decode_batch_size.mean": -0.06471172257286728, + "graph_full_share": 0.0005850648890149968, + "graph_none_share": -0.000531877171831785, + "graph_padding_fraction": -0.00034778464236925486, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -4.995504046367838e-05, + "kv_usage_mean": -0.0014804826259866288, + "preemptions": 0.0, + "prefill_token_fraction": -4.237146928010027e-05, + "queue_running_mean": -0.14321269680000004, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.24609375, + "request_hash": "7e48c6bfc00eeadd6011111170c31123fb117c9fa75c18ccc8d8d505fd7fcff3", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.24609375/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/confirm-1-anchor-0.24609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.060654008438818074, + "batch_tokens.mean": -2.402267932489451, + "decode_batch_size.mean": -0.06002109704641434, + "graph_full_share": 0.0005801687763713481, + "graph_none_share": -0.0005274261603375532, + "graph_padding_fraction": -0.00017401096415257163, + "kv_usage_end_minus_start": 4.995504046356736e-05, + "kv_usage_max": -4.995504046367838e-05, + "kv_usage_mean": -0.0013647646656208823, + "preemptions": 0.0, + "prefill_token_fraction": -6.355303957206182e-05, + "queue_running_mean": -0.09758093800000012, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.6000000000000014 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp1_mns64/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.9280844155844155, + "batch_tokens.mean": 100.79992784992785, + "decode_batch_size.mean": 0.9185245310245311, + "graph_full_share": -0.009559884559884546, + "graph_none_share": 0.02168109668109668, + "graph_padding_fraction": -0.18173061839089436, + "kv_usage_end_minus_start": 0.048360614955967995, + "kv_usage_max": 0.004527588437235774, + "kv_usage_mean": 0.004456117360644808, + "preemptions": 0.0, + "prefill_token_fraction": 0.8754134325195487, + "queue_running_mean": -0.277960427, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -31.799999999999997 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.2265625, + "request_hash": "a05848006335fc9f595432bab793a13d427244886e4f04e51ba74d495453ef90", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9920634920634921, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.2265625/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9920634920634921, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.2265625/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.013264036958066772, + "batch_tokens.mean": 0.31208244491826065, + "decode_batch_size.mean": 0.013175195451314625, + "graph_full_share": -8.884150675192437e-05, + "graph_none_share": 8.88415067519556e-05, + "graph_padding_fraction": 5.628150611948664e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": 6.337324494750524e-05, + "preemptions": 0.0, + "prefill_token_fraction": -2.779388529827731e-05, + "queue_running_mean": 0.007332840999999757, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": -0.20000000000000284 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.23046875, + "request_hash": "0f4c973153935c8abcd421eb631900670e31af0dc8a33bdf7e4d3d85e7ed7711", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9923076923076923, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.23046875/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9923076923076923, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.23046875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.01133760879523571, + "batch_tokens.mean": -0.4460948236371962, + "decode_batch_size.mean": -0.011211635364177575, + "graph_full_share": 0.00012597343105813508, + "graph_none_share": -0.0001259734310581767, + "graph_padding_fraction": 2.5032232616357952e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -2.595429314883324e-06, + "preemptions": 0.0, + "prefill_token_fraction": -2.465132225659694e-05, + "queue_running_mean": -0.0003634421999998416, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.20000000000000284 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.234375, + "request_hash": "14d2a74fffe25e7d9b3e7490c392eae0dc31bb1f3fe6566c9c0a567543bdd6ac", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9924242424242424, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.234375/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9924242424242424, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.234375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.03836578028285853, + "batch_tokens.mean": -2.3411943704830662, + "decode_batch_size.mean": -0.037743352121442975, + "graph_full_share": 0.0005705591479649996, + "graph_none_share": -0.0005705591479650066, + "graph_padding_fraction": 8.67237650413492e-05, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.00012192636323195283, + "preemptions": 0.0, + "prefill_token_fraction": -0.00016953030936628366, + "queue_running_mean": -0.0205579666000002, + "queue_waiting_mean": -0.0408831314, + "scheduler_steps_per_s": 0.6000000000000014 + }, + "descriptive_full_outcome": { + "delta_pass_rate": -0.006993006993006978, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.25, + "request_hash": "26629995f38f2c013d5aa5e4b9d7344311ab1f951c9d83e68212c67ca8076fda", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 1 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.958041958041958, + "mns": 8, + "trial_id": "tp1_mns8/anchor-0.25/result.json" + }, + "target": { + "cell": "tp1_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.951048951048951, + "mns": 8, + "trial_id": "tp1_mns8/confirm-1-anchor-0.25/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.2758963500935341, + "batch_tokens.mean": -16.135882715276434, + "decode_batch_size.mean": -0.2723042461451284, + "graph_full_share": 0.0035921039484052564, + "graph_none_share": -0.0048721071863581, + "graph_padding_fraction": 0.10945908199689197, + "kv_usage_end_minus_start": -0.004480240988461115, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0010388280544342795, + "preemptions": 0.0, + "prefill_token_fraction": -0.7696270176402451, + "queue_running_mean": -0.36681181659999984, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 28.400000000000006 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/confirm-1-anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.006385050186775487, + "batch_tokens.mean": -0.006810792236799035, + "decode_batch_size.mean": -0.006341827643625564, + "graph_full_share": 4.322254314959029e-05, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.0016266700040674653, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -1.7558597020827724e-05, + "preemptions": 0.0, + "prefill_token_fraction": 3.0428082455902228e-05, + "queue_running_mean": -0.0059703380000000195, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.4000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/anchor-0.5/result.json" + }, + "target": { + "cell": "tp2_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp2_mns16/confirm-1-anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.006361323155215537, + "batch_tokens.mean": -0.006361323155216425, + "decode_batch_size.mean": -0.006361323155215537, + "graph_full_share": 0.0, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.0018169528035430205, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -2.673005121892058e-05, + "preemptions": 0.0, + "prefill_token_fraction": 0.0001139335897381577, + "queue_running_mean": -0.008856016399999334, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.0 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp2_mns32/confirm-1-anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.36334379104298975, + "batch_tokens.mean": -99.4815933746712, + "decode_batch_size.mean": 0.38174161358586733, + "graph_full_share": 0.015964732518547464, + "graph_none_share": -0.019464720194647202, + "graph_padding_fraction": 0.26242965569766435, + "kv_usage_end_minus_start": -0.024225165055486664, + "kv_usage_max": -0.013153357937349175, + "kv_usage_mean": -0.006611009430699675, + "preemptions": 0.0, + "prefill_token_fraction": -0.8730961888774497, + "queue_running_mean": -2.3594406332, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 55.999999999999986 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.7385786802030457, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.75390625, + "request_hash": "2677a6a12ff0e104d6246dd121d0667ff4fe3d25f8d68699547906420246ebdd", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.20558375634517767, + "mns": 32, + "trial_id": "tp2_mns32/anchor-0.75390625/result.json" + }, + "target": { + "cell": "tp2_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.9441624365482234, + "mns": 32, + "trial_id": "tp2_mns32/confirm-1-anchor-0.75390625/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.008984926990108555, + "batch_tokens.mean": -0.009617608016100565, + "decode_batch_size.mean": -0.00892069541386542, + "graph_full_share": 6.42315762428014e-05, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.00023488823078426924, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -3.459821231118886e-05, + "preemptions": 0.0, + "prefill_token_fraction": 3.0382556995986465e-05, + "queue_running_mean": -0.01180635799999985, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.5999999999999943 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.5, + "request_hash": "3160017dbd8249f711eb115f1e72ddbb37eab3a3333a48762cc3104e94bd2f15", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.5/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp2_mns64/confirm-1-anchor-0.5/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.36810601285071254, + "batch_tokens.mean": -111.9216322174668, + "decode_batch_size.mean": -0.3386389524789051, + "graph_full_share": 0.025109761896862537, + "graph_none_share": -0.026143790849673203, + "graph_padding_fraction": 0.22131844466419573, + "kv_usage_end_minus_start": -0.03483866016958903, + "kv_usage_max": -0.018634334752970383, + "kv_usage_mean": -0.006551218139822743, + "preemptions": 0.0, + "prefill_token_fraction": -0.8766852234122235, + "queue_running_mean": -2.0965550011999996, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 65.8 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.6624040920716113, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.75, + "request_hash": "4f4c792e90f5c1bf33978fea26debcadfce5f8b872a31be86a88a1bf6b09a21c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.23017902813299232, + "mns": 64, + "trial_id": "tp2_mns64/anchor-0.75/result.json" + }, + "target": { + "cell": "tp2_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.8925831202046036, + "mns": 64, + "trial_id": "tp2_mns64/confirm-1-anchor-0.75/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.017777364465908274, + "batch_tokens.mean": -0.019283121886095156, + "decode_batch_size.mean": -0.017624495692285524, + "graph_full_share": 0.0001528687736229717, + "graph_none_share": -4.586063208690105e-05, + "graph_padding_fraction": -0.0033762430254597187, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -3.866481934330576e-05, + "preemptions": 0.0, + "prefill_token_fraction": -3.0428082455902228e-05, + "queue_running_mean": -0.01706728199999974, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.200000000000017 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.4921875, + "request_hash": "cd227aee3be472a35e8c35af60a44a1411c4dc689cbd98746b3fb8a8d329b44c", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.4921875/result.json" + }, + "target": { + "cell": "tp2_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp2_mns8/confirm-1-anchor-0.4921875/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.373100622158411, + "batch_tokens.mean": -19.752101280210574, + "decode_batch_size.mean": -0.36846434553720986, + "graph_full_share": 0.004636276621201252, + "graph_none_share": -0.006318796362766213, + "graph_padding_fraction": 0.13466522306023426, + "kv_usage_end_minus_start": -0.004491915827622761, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0010760794618353137, + "preemptions": 0.0, + "prefill_token_fraction": -0.7698627933477863, + "queue_running_mean": -0.3758000022000001, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 24.799999999999983 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.6190476190476191, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.49609375, + "request_hash": "a78992ce59bb3b857a7ef8844e0a690ea9dd529a659cf5edd599b63fef8b6a80", + "study_sha256": "9474f0d0b53579f1db852ca68abfb0b96ba43ae4e17738118bf8e3209eb09ece", + "tp": 2 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.27106227106227104, + "mns": 8, + "trial_id": "tp2_mns8/anchor-0.49609375/result.json" + }, + "target": { + "cell": "tp2_mns8", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.8901098901098901, + "mns": 8, + "trial_id": "tp2_mns8/confirm-1-anchor-0.49609375/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.035790047012957515, + "batch_tokens.mean": -0.037904196766425535, + "decode_batch_size.mean": -0.03550338263960562, + "graph_full_share": -0.0008241600733860377, + "graph_none_share": 0.0, + "graph_padding_fraction": -0.002233730654183863, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -1.0202155715566263e-05, + "kv_usage_mean": -4.688019669552533e-05, + "preemptions": 0.0, + "prefill_token_fraction": 0.0, + "queue_running_mean": -0.04172862040000069, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.1999999999999886 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.2469565217391304, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033182214016, + "request_hash": "e6b624a0212807df3733c54b4a2160fba24f19ca4776b14f34918d83428bba7f", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7530434782608696, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033182214016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.033182214016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.31521561989075764, + "batch_tokens.mean": -113.69135049600723, + "decode_batch_size.mean": -0.2948443636019542, + "graph_full_share": 0.013747312308444393, + "graph_none_share": -0.02466367713004484, + "graph_padding_fraction": 0.2327910644295707, + "kv_usage_end_minus_start": -0.014762519320332457, + "kv_usage_max": -0.002734177731754772, + "kv_usage_mean": -0.003886883143649442, + "preemptions": 0.0, + "prefill_token_fraction": -0.8857303498661139, + "queue_running_mean": -3.9262134145999994, + "queue_waiting_mean": -1.5032174162, + "scheduler_steps_per_s": 81.39999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.34982935153583616, + "feasibility_transition": "false->false" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.06313993174061433, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.4129692832764505, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.0565950637806516, + "batch_tokens.mean": -1.2259074371774261, + "decode_batch_size.mean": -0.05612473435097076, + "graph_full_share": -0.0006731434430105043, + "graph_none_share": -0.0011210762331838565, + "graph_padding_fraction": 0.03352610055520758, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -8.342130161142786e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.1499865777764538, + "queue_running_mean": -0.07804562620000066, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 1.799999999999983 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.017452685323813277, + "batch_tokens.mean": -0.018160486007140264, + "decode_batch_size.mean": -0.01735671234980174, + "graph_full_share": 8.877500095971769e-05, + "graph_none_share": 0.0, + "graph_padding_fraction": 0.0015546681503931792, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -5.101077857783132e-06, + "kv_usage_mean": -2.2530897596876806e-05, + "preemptions": 0.0, + "prefill_token_fraction": 5.993471374250714e-05, + "queue_running_mean": -0.021152762000000713, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 0.4000000000000057 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.2469565217391304, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033182214016, + "request_hash": "e6b624a0212807df3733c54b4a2160fba24f19ca4776b14f34918d83428bba7f", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns16", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.7530434782608696, + "mns": 16, + "trial_id": "tp4_mns16/anchor-0.033182214016/result.json" + }, + "target": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "trial_id": "tp4_mns16/confirm-2-anchor-0.033182214016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.16723437281046838, + "batch_tokens.mean": -213.3770035894838, + "decode_batch_size.mean": 0.2293820267109341, + "graph_full_share": 0.029853079087214685, + "graph_none_share": -0.03201970443349754, + "graph_padding_fraction": 0.24386359763484056, + "kv_usage_end_minus_start": -0.02873466386876411, + "kv_usage_max": -0.01666487968508279, + "kv_usage_mean": -0.006565235690514186, + "preemptions": 0.0, + "prefill_token_fraction": -0.9129695381069503, + "queue_running_mean": -6.573865443400001, + "queue_waiting_mean": -0.105609193, + "scheduler_steps_per_s": 101.60000000000001 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.9692832764505119, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.06700882459564284, + "batch_tokens.mean": -1.237066033004087, + "decode_batch_size.mean": -0.06643525070889744, + "graph_full_share": -0.0005748153886646623, + "graph_none_share": -0.0011210762331838565, + "graph_padding_fraction": 0.032357274596769375, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": -5.1056616682254585e-06, + "kv_usage_mean": -0.00010377856500988665, + "preemptions": 0.0, + "prefill_token_fraction": -0.15007350008784556, + "queue_running_mean": -0.10319076419999984, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.1999999999999886 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.16017120245497818, + "batch_tokens.mean": -213.38441949985196, + "decode_batch_size.mean": 0.22236668550970418, + "graph_full_share": 0.029898516783762652, + "graph_none_share": -0.03201970443349754, + "graph_padding_fraction": 0.2419788945317641, + "kv_usage_end_minus_start": -0.02873466386876411, + "kv_usage_max": -0.016659774023414564, + "kv_usage_mean": -0.006562301403937485, + "preemptions": 0.0, + "prefill_token_fraction": -0.9129575557004137, + "queue_running_mean": -6.570182108200001, + "queue_waiting_mean": -0.105609193, + "scheduler_steps_per_s": 101.8 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.9692832764505119, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns32", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.030716723549488054, + "mns": 32, + "trial_id": "tp4_mns32/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns32", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 32, + "trial_id": "tp4_mns32/confirm-2-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.5328616639627084, + "batch_tokens.mean": -211.42080067637045, + "decode_batch_size.mean": 0.5912671663277198, + "graph_full_share": 0.027680346411352086, + "graph_none_share": -0.03178484107579462, + "graph_padding_fraction": 0.23688813872052464, + "kv_usage_end_minus_start": -0.028772123554492213, + "kv_usage_max": -0.016661043117286778, + "kv_usage_mean": -0.006272675400456023, + "preemptions": 0.0, + "prefill_token_fraction": -0.912693798885484, + "queue_running_mean": -6.2424369058, + "queue_waiting_mean": -0.10551258, + "scheduler_steps_per_s": 89.39999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8430034129692833, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 0.9573378839590444, + "mns": 64, + "trial_id": "tp4_mns64/confirm-1-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.08536723607223529, + "batch_tokens.mean": -1.2605052275442308, + "decode_batch_size.mean": -0.08463476949102322, + "graph_full_share": 0.0006975872202020783, + "graph_none_share": -0.0011248593925759281, + "graph_padding_fraction": 0.034272917448908424, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.00010935706758585708, + "preemptions": 0.0, + "prefill_token_fraction": -0.15010635693287294, + "queue_running_mean": -0.10249850960000018, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 2.799999999999983 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.034252608017, + "request_hash": "21c1f3b7d972b19c952a894863a70d1e85bd0e24a7e7b6b0fcc088360d0e01bd", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.034252608017/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/confirm-1-anchor-0.034252608017/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": 0.1540235762873472, + "batch_tokens.mean": -211.8229141697429, + "decode_batch_size.mean": 0.21558506586552006, + "graph_full_share": 0.029510309964790093, + "graph_none_share": -0.03178484107579462, + "graph_padding_fraction": 0.24384653183386754, + "kv_usage_end_minus_start": -0.028772123554492213, + "kv_usage_max": -0.016681492387759067, + "kv_usage_mean": -0.006626073152505554, + "preemptions": 0.0, + "prefill_token_fraction": -0.9128498654266666, + "queue_running_mean": -6.5953128982, + "queue_waiting_mean": -0.10551258, + "scheduler_steps_per_s": 101.8 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.8856655290102389, + "feasibility_transition": "false->true" + }, + "group": { + "anchor": 0.033717411016, + "request_hash": "472a874ecc0e844b5b36a715bbc309b63e73074d39327cd734869f76721d71db", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns64", + "early_stopped": true, + "full_feasible": false, + "full_pass_rate": 0.11433447098976109, + "mns": 64, + "trial_id": "tp4_mns64/anchor-0.033717411016/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "trial_id": "tp4_mns64/confirm-2-anchor-0.033717411016/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.3497477888336098, + "batch_tokens.mean": -19.68794568822554, + "decode_batch_size.mean": -0.3465001381978996, + "graph_full_share": 0.004242675511332172, + "graph_none_share": -0.004954394693200663, + "graph_padding_fraction": 0.14768569288462857, + "kv_usage_end_minus_start": -0.0007955733482928329, + "kv_usage_max": 0.0, + "kv_usage_mean": -0.0004780084145793828, + "preemptions": 0.0, + "prefill_token_fraction": -0.8110422816759224, + "queue_running_mean": -0.41782893040000024, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 28.19999999999999 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.016591107009, + "request_hash": "2407501cda53bf9c761c448dc50ab4780b4a55958cc00ba1b19cd09de1cf73e1", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.016591107009/result.json" + }, + "target": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/confirm-1-anchor-0.016591107009/result.json" + } + }, + { + "delta_state": { + "batch_size.mean": -0.060037523452157515, + "batch_tokens.mean": -3.365563298490128, + "decode_batch_size.mean": -0.05956848030018724, + "graph_full_share": 0.0004467077637808625, + "graph_none_share": -0.000982757080318056, + "graph_padding_fraction": 0.07118473038433726, + "kv_usage_end_minus_start": 0.0, + "kv_usage_max": 0.0, + "kv_usage_mean": -6.376841981488128e-05, + "preemptions": 0.0, + "prefill_token_fraction": -0.5189516489461117, + "queue_running_mean": -0.1074558633999998, + "queue_waiting_mean": 0.0, + "scheduler_steps_per_s": 5.200000000000017 + }, + "descriptive_full_outcome": { + "delta_pass_rate": 0.0, + "feasibility_transition": "true->true" + }, + "group": { + "anchor": 0.017126304009, + "request_hash": "93cd99ebf2a760cc440dce0cb7e51273686dcfbfb07650d27e9e16d2ff1d2eed", + "study_sha256": "76e9d642ba9570917cfeeea848ed92f27904e87760eedabd48805b84e82eca26", + "tp": 4 + }, + "kind": "same_config_repeat", + "source": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/anchor-0.017126304009/result.json" + }, + "target": { + "cell": "tp4_mns8", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 8, + "trial_id": "tp4_mns8/confirm-1-anchor-0.017126304009/result.json" + } + } + ], + "sanity": { + "action_pairs": 17, + "distinct_action_vectors": 17, + "invariants": { + "all_pair_hashes_match": true, + "all_values_finite": true, + "expected_action_pair_count": true, + "nonnegative_counters": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true, + "sufficient_repeat_pair_count": true + }, + "red_flags": [], + "repeat_pairs": 29, + "trials": 66 + } + } + }, + "provenance": { + "analysis_script": "/home/gahow/phd/aituner/runs/intervention-response-v0/analyze_phase6.py", + "analysis_script_sha256": "f1b40c350f22e797b08b129df7b55386b407bd5ae7f466d36987e94d768ded2f", + "phase6_metrics": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/metrics.json", + "phase6_metrics_sha256": "290ba7fcb8727291166de7e4d47afdc84e230052495c81dd087db0ace9f93a16", + "raw_root": "/home/gahow/phd/aituner/runs/opprof-phase6/phase6/solo-authoritative/cells", + "streams": [ + { + "bytes": 12745297, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp1_mns16/opprof/opprof-v1-dp0-pid2781473-1783876972093434233.jsonl", + "sha256": "265932302b079d1ad81febc4a345f87ac2a3e00118e63bf5873b92d61146d3b8" + }, + { + "bytes": 20018398, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp1_mns32/opprof/opprof-v1-dp0-pid2786356-1783877278912197439.jsonl", + "sha256": "552bb3961893ab20760e95a6148a24387df544370ca2cb3af052fa87a7ef79f6" + }, + { + "bytes": 19825716, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp1_mns64/opprof/opprof-v1-dp0-pid2795014-1783877830485564578.jsonl", + "sha256": "f07f2beed8ffcd695665edba931cd5824f10f4ef678064ac09d5a0d9b96acf50" + }, + { + "bytes": 38381207, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp1_mns8/opprof/opprof-v1-dp0-pid2770271-1783876258461072221.jsonl", + "sha256": "f3758f599bd58952d039eae3cc7a1b6abe3372e82fe15aff38d8f27bedf0cbef" + }, + { + "bytes": 42681418, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp2_mns16/opprof/opprof-v1-dp0-pid2755628-1783875389707263062.jsonl", + "sha256": "e56faa0ceb0f750ad781b04a0fa86388a8fd4acdff4503fb40064bbb43b8a4de" + }, + { + "bytes": 20701713, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp2_mns32/opprof/opprof-v1-dp0-pid2730662-1783873935616837232.jsonl", + "sha256": "ab1070348c976491ff5ad4905ab172678f692598810ca72a0dbab824f3bb491a" + }, + { + "bytes": 24186851, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp2_mns64/opprof/opprof-v1-dp0-pid2736153-1783874256706634444.jsonl", + "sha256": "c8d7582674a9ff3cba8363a93672b96977082f6f2eb595aee0ea41012e85978e" + }, + { + "bytes": 25542356, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp2_mns8/opprof/opprof-v1-dp0-pid2750298-1783875076865595283.jsonl", + "sha256": "ad006affdccb0c9cf2ae97938810a7f232ed0517086d3e20772aa461930f8518" + }, + { + "bytes": 50625291, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp4_mns16/opprof/opprof-v1-dp0-pid2741574-1783874579483097755.jsonl", + "sha256": "4443a0f159f23242445e48c846e45e29b3647012734edeff2678bcf75b008ada" + }, + { + "bytes": 34980126, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp4_mns32/opprof/opprof-v1-dp0-pid2712385-1783873000008589576.jsonl", + "sha256": "445b190a5392df7aa05555de33a550d6401d739bc9000494af650671db4c63c9" + }, + { + "bytes": 34066572, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp4_mns64/opprof/opprof-v1-dp0-pid2721119-1783873467764669604.jsonl", + "sha256": "ddc7bf89342fe7fececc68b6c6be988816abd1ff37dfc63c8c2a55f7b9a96f44" + }, + { + "bytes": 52957710, + "path": "runs/opprof-phase6/phase6/solo-authoritative/cells/tp4_mns8/opprof/opprof-v1-dp0-pid2762505-1783875811578126796.jsonl", + "sha256": "ffdbfe671cecc166ea718ff0c31f7a15f788dcc2da64d6b4ee10fc16d489a849" + } + ] + }, + "sanity": { + "red_flags": [], + "stream_bytes": { + "distinct_n": 12, + "max": 52957710.0, + "min": 12745297.0, + "n": 12 + }, + "stream_count": 12 + }, + "schema": "intervention-response-audit-v0", + "stable_qualifying_features": [], + "status": "COMPLETE" +} diff --git a/runs/intervention-response-v0/test_analysis.py b/runs/intervention-response-v0/test_analysis.py new file mode 100644 index 0000000..74ada8c --- /dev/null +++ b/runs/intervention-response-v0/test_analysis.py @@ -0,0 +1,57 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import math +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_v0", HERE / "analyze_phase6.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def pair(module, delta: dict[str, float]) -> dict[str, object]: + state = {feature: 0.0 for feature in module.ALL_FEATURES} + state.update(delta) + return {"delta_state": state} + + +def main() -> None: + module = load_module() + assert module.numeric([0.0, 1.0, 1.0]) == { + "n": 3, + "min": 0.0, + "max": 1.0, + "distinct_n": 2, + } + assert math.isclose(module.quantile([0.0, 10.0], 0.95), 9.5) + + actions = [ + pair(module, {"queue_waiting_mean": -1.0 - 0.1 * index}) + for index in range(8) + ] + repeats = [ + pair(module, {"queue_waiting_mean": 0.01 * ((index % 3) - 1)}) + for index in range(20) + ] + stats = module.response_statistics(actions, repeats) + waiting = stats["queue_waiting_mean"] + assert waiting["qualifies"] + assert waiting["action_signs"]["negative"] == 8 + assert waiting["action_signs"]["consistency"] == 1.0 + assert waiting["effect_to_repeat_median"] > 2.0 + assert not stats["kv_usage_mean"]["qualifies"] + print("intervention response v0 analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v0/test_p1_analysis.py b/runs/intervention-response-v0/test_p1_analysis.py new file mode 100644 index 0000000..8a99899 --- /dev/null +++ b/runs/intervention-response-v0/test_p1_analysis.py @@ -0,0 +1,35 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_p1", HERE / "analyze_p1.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def main() -> None: + module = load_module() + values = [-2.0, -1.0, 1.0, 2.0] + labels = [0, 0, 1, 1] + threshold, direction, balanced = module._fit_threshold(values, labels) + assert direction == 1 + assert -1.0 < threshold < 1.0 + assert balanced == 1.0 + assert module._balanced_accuracy(labels, labels) == 1.0 + assert module._balanced_accuracy(labels, [1, 1, 0, 0]) == 0.0 + print("intervention response P1 confirmation analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v2/analyze_existing.py b/runs/intervention-response-v2/analyze_existing.py new file mode 100644 index 0000000..3d3fe85 --- /dev/null +++ b/runs/intervention-response-v2/analyze_existing.py @@ -0,0 +1,467 @@ +#!/usr/bin/env python3 +"""Audit telemetry responses over every uncensored replay decile. + +This corrective analysis keeps the frozen P1 pairs and thresholds, but replaces +the absolute 5/10-second cutoff with cumulative and non-overlapping 10%-of-trace +windows. It deliberately reports every common decile instead of selecting the +best-looking horizon. +""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +from pathlib import Path +from statistics import median +from typing import Any, Iterable, Mapping + + +HERE = Path(__file__).resolve().parent +P1_PATH = HERE.parent / "intervention-response-v0" / "analyze_p1.py" +SCHEMA = "intervention-response-phase-aware-existing-v2" +DECILE_FRACTION = 0.1 +MAX_DECILES = 10 + + +def _load_p1(): + spec = importlib.util.spec_from_file_location("intervention_response_p1", P1_PATH) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +P1 = _load_p1() + + +def numeric(values: Iterable[float | int]) -> dict[str, Any]: + finite = [float(value) for value in values] + result = P1.V0.numeric(finite) + result["median"] = median(finite) + return result + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def trial_directories(run_root: Path) -> list[Path]: + result = [] + for cell in sorted((run_root / "cells").iterdir()): + if not cell.is_dir(): + continue + for candidate in sorted(cell.iterdir()): + if candidate.is_dir() and P1.RUN_PATTERN.match(candidate.name): + result.append(candidate) + if not result: + raise ValueError("P1 run root contains no measured trial directories") + return result + + +def load_metadata(run_root: Path) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]: + metadata = [] + streams = [] + for cell in sorted((run_root / "cells").iterdir()): + if not cell.is_dir(): + continue + stream_paths = sorted((cell / "opprof").glob("*.jsonl")) + if len(stream_paths) != 1: + raise ValueError(f"{cell}: expected one Layer-1 stream") + stream_path = stream_paths[0] + streams.append( + { + "cell": cell.name, + "path": str(stream_path.resolve()), + "sha256": sha256_file(stream_path), + "bytes": stream_path.stat().st_size, + } + ) + for run_dir in trial_directories(run_root): + match = P1.RUN_PATTERN.match(run_dir.name) + assert match is not None + level, replicate_text = match.groups() + result_path = run_dir / "result.json" + requests_path = run_dir / "requests.jsonl" + result = json.loads(result_path.read_text(encoding="utf-8")) + selected = int(result["selection"]["count"]) + offered = float(result["selection"]["offered_req_s"]) + if selected <= 0 or offered <= 0.0: + raise ValueError(f"{result_path}: invalid selected count or offered rate") + metadata.append( + { + "trial_id": str(result_path.relative_to(run_root)), + "cell": str(result["cell"]), + "tp": int(result["tp"]), + "mns": int(result["mns"]), + "level": level, + "replicate": int(replicate_text), + "elapsed_s": float(result["interval"]["elapsed_s"]), + "trace_duration_s": round(selected / offered, 9), + "early_stopped": bool(result["early_stopped"]), + "request_count": selected, + "result_sha256": sha256_file(result_path), + "requests_sha256": sha256_file(requests_path), + } + ) + return metadata, streams + + +def common_decile_fractions( + *, trace_duration_s: float, minimum_elapsed_s: float +) -> tuple[float, ...]: + if trace_duration_s <= 0.0 or minimum_elapsed_s <= 0.0: + raise ValueError("trace duration and elapsed time must be positive") + supported = min( + MAX_DECILES, + int(math.floor((minimum_elapsed_s / trace_duration_s) * 10.0 + 1e-12)), + ) + return tuple( + round(index * DECILE_FRACTION, 10) for index in range(1, supported + 1) + ) + + +def _trial_record( + *, + run_root: Path, + run_dir: Path, + result: Mapping[str, Any], + state: dict[str, float], + outcome: dict[str, float], +) -> dict[str, Any]: + match = P1.RUN_PATTERN.match(run_dir.name) + assert match is not None + level, replicate_text = match.groups() + result_path = run_dir / "result.json" + requests_path = run_dir / "requests.jsonl" + return { + "trial_id": str(result_path.relative_to(run_root)), + "cell": str(result["cell"]), + "tp": int(result["tp"]), + "mns": int(result["mns"]), + "level": level, + "replicate": int(replicate_text), + "offered_rate_per_gpu": float( + result["selection"]["offered_req_s_per_gpu"] + ), + "request_hash": str(result["selection"]["request_id_order_sha256"]), + "request_count": int(result["selection"]["count"]), + "result_sha256": sha256_file(result_path), + "requests_sha256": sha256_file(requests_path), + "full_pass_rate": float(result["pass_rate"]), + "full_feasible": bool(result["feasible"]), + "early_stopped": bool(result["early_stopped"]), + "state": state, + "outcome": outcome, + } + + +def load_interval_trials( + run_root: Path, + intervals_s: tuple[tuple[float, float], ...], +) -> tuple[dict[tuple[float, float], list[dict[str, Any]]], list[dict[str, Any]]]: + by_interval = {interval: [] for interval in intervals_s} + stream_provenance = [] + for cell in sorted((run_root / "cells").iterdir()): + if not cell.is_dir(): + continue + stream_paths = sorted((cell / "opprof").glob("*.jsonl")) + if len(stream_paths) != 1: + raise ValueError(f"{cell}: expected one Layer-1 stream") + stream_path = stream_paths[0] + stream = P1.load_jsonl(stream_path) + stream_provenance.append( + { + "cell": cell.name, + "path": str(stream_path.resolve()), + "sha256": sha256_file(stream_path), + "bytes": stream_path.stat().st_size, + } + ) + for run_dir in sorted(cell.iterdir()): + if not run_dir.is_dir() or P1.RUN_PATTERN.match(run_dir.name) is None: + continue + result_path = run_dir / "result.json" + requests_path = run_dir / "requests.jsonl" + result = json.loads(result_path.read_text(encoding="utf-8")) + requests = P1.load_jsonl(requests_path) + start_ns = int(result["interval"]["start_mono_ns"]) + elapsed_s = float(result["interval"]["elapsed_s"]) + for interval in intervals_s: + start_s, end_s = interval + if start_s < 0.0 or end_s <= start_s: + raise ValueError(f"invalid analysis interval: {interval}") + if elapsed_s + 1e-9 < end_s: + raise ValueError( + f"{result_path}: elapsed {elapsed_s} shorter than {end_s}s" + ) + state = P1.V0.flatten_state( + P1.summarize_engine( + stream, + start_ns=start_ns + int(start_s * 1e9), + end_ns=start_ns + int(end_s * 1e9), + request_count=int(result["selection"]["count"]), + ) + ) + outcome = P1._prefix_outcome(result, requests, end_s) + by_interval[interval].append( + _trial_record( + run_root=run_root, + run_dir=run_dir, + result=result, + state=state, + outcome=outcome, + ) + ) + return by_interval, stream_provenance + + +def coverage(trials: list[dict[str, Any]]) -> dict[str, Any]: + admitted = [float(trial["outcome"]["admitted_fraction"]) for trial in trials] + completed = [ + float(trial["outcome"]["admitted_fraction"]) + * float(trial["outcome"]["completed_over_admitted"]) + for trial in trials + ] + return { + "admitted_fraction_of_total": numeric(admitted), + "completed_fraction_of_total": numeric(completed), + } + + +def slim_window_analysis( + trials: list[dict[str, Any]], *, start_s: float, end_s: float, fraction: float +) -> dict[str, Any]: + analysis = P1.analyze_horizon(trials, end_s) + return { + "start_s": start_s, + "end_s": end_s, + "end_fraction": fraction, + "coverage_at_end": coverage(trials), + "action_pairs": len(analysis["actions"]), + "repeat_pairs": len(analysis["repeats"]), + "response_statistics": analysis["response_statistics"], + "qualifying_response_features": analysis["qualifying_response_features"], + "efficacy": analysis["efficacy"], + "sanity": analysis["sanity"], + } + + +def _pearson(left: list[float], right: list[float]) -> float | None: + if len(left) != len(right) or not left: + raise ValueError("Pearson inputs must be non-empty and have equal length") + left_mean = sum(left) / len(left) + right_mean = sum(right) / len(right) + numerator = sum( + (x - left_mean) * (y - right_mean) + for x, y in zip(left, right, strict=True) + ) + left_ss = sum((x - left_mean) ** 2 for x in left) + right_ss = sum((y - right_mean) ** 2 for y in right) + if left_ss == 0.0 or right_ss == 0.0: + return None + return numerator / math.sqrt(left_ss * right_ss) + + +def trajectory_summary( + block_trials: list[tuple[tuple[float, float], list[dict[str, Any]]]] +) -> dict[str, Any]: + if not block_trials: + raise ValueError("trajectory requires at least one block") + identities = [] + states_by_block = [] + for interval, trials in block_trials: + ordered = sorted( + trials, + key=lambda trial: (trial["cell"], trial["level"], trial["replicate"]), + ) + current_identities = [ + (trial["cell"], trial["level"], trial["replicate"]) for trial in ordered + ] + if identities and current_identities != identities: + raise ValueError("trajectory blocks do not contain identical trials") + identities = current_identities + states_by_block.append((interval, [trial["state"] for trial in ordered])) + + features = {} + for feature in P1.V0.ALL_FEATURES: + block_values = [ + [float(state[feature]) for state in states] + for _interval, states in states_by_block + ] + first = block_values[0] + last = block_values[-1] + delta = [right - left for left, right in zip(first, last, strict=True)] + features[feature] = { + "block_medians": [median(values) for values in block_values], + "first_to_last_delta": numeric(delta), + "first_to_last_abs_delta": numeric(abs(value) for value in delta), + "first_to_last_pearson": _pearson(first, last), + "changed_trials": sum(abs(value) > 1e-12 for value in delta), + } + return { + "trial_count": len(identities), + "blocks": [ + {"start_s": interval[0], "end_s": interval[1]} + for interval, _states in states_by_block + ], + "features": features, + } + + +def audit(*, run_root: Path, manifest_path: Path, output_path: Path) -> dict[str, Any]: + metadata, metadata_streams = load_metadata(run_root) + durations = [float(item["trace_duration_s"]) for item in metadata] + elapsed = [float(item["elapsed_s"]) for item in metadata] + duration = median(durations) + deciles = common_decile_fractions( + trace_duration_s=duration, minimum_elapsed_s=min(elapsed) + ) + if not deciles: + raise ValueError("no complete replay decile is shared by all trials") + cumulative_intervals = tuple( + (0.0, round(duration * fraction, 9)) for fraction in deciles + ) + block_intervals = tuple( + ( + round(duration * (fraction - DECILE_FRACTION), 9), + round(duration * fraction, 9), + ) + for fraction in deciles + ) + all_intervals = tuple(dict.fromkeys([*cumulative_intervals, *block_intervals])) + trials_by_interval, streams = load_interval_trials(run_root, all_intervals) + manifest_validation = P1.validate_manifest( + trials_by_interval[cumulative_intervals[0]], manifest_path + ) + + cumulative = [] + blocks = [] + for fraction, cumulative_interval, block_interval in zip( + deciles, cumulative_intervals, block_intervals, strict=True + ): + cumulative.append( + slim_window_analysis( + trials_by_interval[cumulative_interval], + start_s=cumulative_interval[0], + end_s=cumulative_interval[1], + fraction=fraction, + ) + ) + blocks.append( + slim_window_analysis( + trials_by_interval[block_interval], + start_s=block_interval[0], + end_s=block_interval[1], + fraction=fraction, + ) + ) + + invariants = { + "expected_trial_count": len(metadata) == 36, + "trace_duration_consistent": max(durations) - min(durations) <= 1e-9, + "all_intervals_uncensored": all( + item["elapsed_s"] + 1e-9 >= cumulative_intervals[-1][1] + for item in metadata + ), + "stream_provenance_consistent": metadata_streams == streams, + "manifest_trials_match": ( + manifest_validation["expected_trials"] + == manifest_validation["matched_trials"] + == len(metadata) + ), + "all_window_sanity_pass": all( + not item["sanity"]["red_flags"] for item in [*cumulative, *blocks] + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + complete_full_trajectory = min(elapsed) + 1e-9 >= duration + if red_flags: + decision = "STOP_DATA_INVALID" + elif not complete_full_trajectory: + decision = "REQUIRES_UNCENSORED_PHASE_AWARE_PILOT" + else: + decision = "FULL_TRAJECTORY_AVAILABLE" + + payload = { + "schema": SCHEMA, + "status": "COMPLETE", + "decision": decision, + "claim_boundary": ( + "Post-hoc corrective audit over every common replay decile. It can " + "diagnose horizon sensitivity but cannot establish a held-out tuning claim." + ), + "design": { + "decile_fraction": DECILE_FRACTION, + "available_deciles": list(deciles), + "trace_duration_s": duration, + "maximum_common_end_s": cumulative_intervals[-1][1], + "maximum_common_fraction": deciles[-1], + "select_best_horizon": False, + "cumulative_and_nonoverlapping_blocks": True, + }, + "cumulative": cumulative, + "blocks": blocks, + "trajectory": trajectory_summary( + [(interval, trials_by_interval[interval]) for interval in block_intervals] + ), + "provenance": { + "analysis_script": str(Path(__file__).resolve()), + "analysis_script_sha256": sha256_file(Path(__file__).resolve()), + "p1_analysis_script": str(P1_PATH.resolve()), + "p1_analysis_script_sha256": sha256_file(P1_PATH), + "run_root": str(run_root.resolve()), + "manifest": str(manifest_path.resolve()), + "manifest_sha256": sha256_file(manifest_path), + "manifest_validation": manifest_validation, + "streams": streams, + "trial_inputs": metadata, + }, + "sanity": { + "trials": len(metadata), + "elapsed_s": numeric(elapsed), + "trace_duration_s": numeric(durations), + "early_stopped": sum(bool(item["early_stopped"]) for item in metadata), + "request_count": numeric(item["request_count"] for item in metadata), + "stream_bytes": numeric(item["bytes"] for item in streams), + "invariants": invariants, + "red_flags": red_flags, + }, + } + output_path.parent.mkdir(parents=True, exist_ok=True) + output_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = audit( + run_root=args.run_root, + manifest_path=args.manifest, + output_path=args.output, + ) + print( + json.dumps( + { + "decision": payload["decision"], + "design": payload["design"], + "sanity": payload["sanity"], + }, + indent=2, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v2/analyze_pilot.py b/runs/intervention-response-v2/analyze_pilot.py new file mode 100644 index 0000000..0586680 --- /dev/null +++ b/runs/intervention-response-v2/analyze_pilot.py @@ -0,0 +1,651 @@ +#!/usr/bin/env python3 +"""Analyze the uncensored 300-second phase-aware matched pilot.""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import math +from collections import defaultdict +from pathlib import Path +from typing import Any, Iterable, Mapping + + +HERE = Path(__file__).resolve().parent +P1_PATH = HERE.parent / "intervention-response-v0" / "analyze_p1.py" +SCHEMA = "intervention-response-phase-aware-pilot-analysis-v3" +EXPECTED_ACTION_PAIRS = 6 +EXPECTED_REPEAT_PAIRS = 8 +MIN_EFFICACY_CLASS = 2 +MAX_LAYER1_GAP_S = 1.0 + + +def _load_p1(): + spec = importlib.util.spec_from_file_location("intervention_response_p1", P1_PATH) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +P1 = _load_p1() + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def numeric(values: Iterable[float | int]) -> dict[str, Any]: + return P1.V0.numeric(values) + + +def _trial_record( + *, + run_root: Path, + session: Mapping[str, Any], + level: str, + result: Mapping[str, Any], + result_path: Path, + requests_path: Path, + state: dict[str, float], + outcome: dict[str, float], + telemetry_coverage: dict[str, float], +) -> dict[str, Any]: + return { + "trial_id": str(result_path.relative_to(run_root)), + "cell": str(result["cell"]), + "tp": int(result["tp"]), + "mns": int(result["mns"]), + "level": level, + "replicate": int(session["replicate"]), + "offered_rate_per_gpu": float( + result["selection"]["offered_req_s_per_gpu"] + ), + "request_hash": str(result["selection"]["request_id_order_sha256"]), + "request_count": int(result["selection"]["count"]), + "result_sha256": sha256_file(result_path), + "requests_sha256": sha256_file(requests_path), + "full_pass_rate": float(result["pass_rate"]), + "full_feasible": bool(result["feasible"]), + "early_stopped": bool(result["early_stopped"]), + "state": state, + "outcome": outcome, + "telemetry_coverage": telemetry_coverage, + } + + +def telemetry_coverage( + records: list[dict[str, Any]], *, start_ns: int, end_ns: int +) -> dict[str, float]: + layer1 = [record for record in records if "step_index" in record] + timestamps = [int(record["submit_mono_ns"]) for record in layer1] + if timestamps != sorted(timestamps): + raise ValueError("Layer-1 timestamps are not monotonic") + selected = [timestamp for timestamp in timestamps if start_ns <= timestamp <= end_ns] + if not selected: + raise ValueError("Layer-1 coverage interval contains no records") + internal_gaps = [ + (right - left) / 1e9 + for left, right in zip(selected, selected[1:], strict=False) + ] + coverage = { + "start_gap_s": (selected[0] - start_ns) / 1e9, + "end_gap_s": (end_ns - selected[-1]) / 1e9, + "max_internal_gap_s": max(internal_gaps, default=0.0), + } + if any(value > MAX_LAYER1_GAP_S for value in coverage.values()): + raise ValueError(f"Layer-1 coverage gap exceeds {MAX_LAYER1_GAP_S}s: {coverage}") + return coverage + + +def validate_result_against_manifest( + *, + result: Mapping[str, Any], + selection: Mapping[str, Any], + session: Mapping[str, Any], + level: str, + expected_duration_s: float, +) -> None: + identity = f"{session['session']}:{level}" + if int(result["mns"]) != int(session["mns"]) or int(result["tp"]) != 4: + raise ValueError(f"config mismatch: {identity}") + if bool(result["early_stopped"]): + raise ValueError(f"early-stopped measured result: {identity}") + if result.get("slo_early_stop_disabled") is not True: + raise ValueError(f"SLO early stop was enabled: {identity}") + if float(result["interval"]["elapsed_s"]) + 1e-9 < expected_duration_s: + raise ValueError(f"result does not cover full arrival window: {identity}") + if int(result["selection"]["count"]) != int(selection["selected_count"]): + raise ValueError(f"selection count mismatch: {identity}") + for result_key, manifest_key in ( + ("request_id_order_sha256", "request_id_order_sha256"), + ("arrival_order_sha256", "arrival_order_sha256"), + ("raw_length_order_sha256", "input_length_order_sha256"), + ): + if result["selection"][result_key] != selection[manifest_key]: + raise ValueError(f"selection hash mismatch {result_key}: {identity}") + if int(result["observed_count"]) != int(selection["selected_count"]): + raise ValueError(f"request accounting mismatch: {identity}") + + +def load_interval_trials( + *, + run_root: Path, + manifest: Mapping[str, Any], + intervals_s: tuple[tuple[float, float], ...], +) -> tuple[dict[tuple[float, float], list[dict[str, Any]]], list[dict[str, Any]]]: + by_interval = {interval: [] for interval in intervals_s} + streams = [] + duration_s = float(manifest["engine"]["duration_s"]) + for session in manifest["sessions"]: + session_root = run_root / "sessions" / str(session["session"]) + stream_paths = sorted((session_root / "opprof").glob("*.jsonl")) + if len(stream_paths) != 1: + raise ValueError(f"{session_root}: expected one Layer-1 stream") + stream_path = stream_paths[0] + stream = P1.load_jsonl(stream_path) + streams.append( + { + "session": str(session["session"]), + "path": str(stream_path.resolve()), + "sha256": sha256_file(stream_path), + "bytes": stream_path.stat().st_size, + } + ) + repetition = manifest["repetitions"][str(session["replicate"])] + for level, selection in repetition["selections"].items(): + result_path = session_root / level / "result.json" + requests_path = session_root / level / "requests.jsonl" + result = json.loads(result_path.read_text(encoding="utf-8")) + requests = P1.load_jsonl(requests_path) + validate_result_against_manifest( + result=result, + selection=selection, + session=session, + level=level, + expected_duration_s=duration_s, + ) + start_ns = int(result["interval"]["start_mono_ns"]) + for interval in intervals_s: + start_s, end_s = interval + interval_start_ns = start_ns + int(start_s * 1e9) + interval_end_ns = start_ns + int(end_s * 1e9) + coverage = telemetry_coverage( + stream, start_ns=interval_start_ns, end_ns=interval_end_ns + ) + state = P1.V0.flatten_state( + P1.summarize_engine( + stream, + start_ns=interval_start_ns, + end_ns=interval_end_ns, + request_count=int(result["selection"]["count"]), + ) + ) + outcome = P1._prefix_outcome(result, requests, end_s) + by_interval[interval].append( + _trial_record( + run_root=run_root, + session=session, + level=level, + result=result, + result_path=result_path, + requests_path=requests_path, + state=state, + outcome=outcome, + telemetry_coverage=coverage, + ) + ) + return by_interval, streams + + +def analyze_window( + trials: list[dict[str, Any]], *, start_s: float, end_s: float, fraction: float +) -> dict[str, Any]: + actions, repeats = P1.build_pairs(trials) + response = P1.V0.response_statistics(actions, repeats) + response_qualifying = sorted( + feature for feature, item in response.items() if item["qualifies"] + ) + labels = [int(pair["full_action_efficacy"]) for pair in actions] + cross_validation_possible = all( + set( + int(pair["full_action_efficacy"]) + for pair in actions + if pair["group"]["replicate"] != held_out + ) + == {0, 1} + for held_out in (1, 2, 3) + ) + if cross_validation_possible: + outcome_cv = P1.one_feature_leave_repeat_out( + actions, delta_key="delta_outcome", features=P1.OUTCOME_FEATURES + ) + telemetry_cv = P1.one_feature_leave_repeat_out( + actions, delta_key="delta_state", features=P1.V0.GATE_FEATURES + ) + outcome_best = float(outcome_cv["best_balanced_accuracy"]) + efficacy_qualifying = sorted( + feature + for feature, item in telemetry_cv["features"].items() + if item["balanced_accuracy"] >= P1.MIN_EFFICACY_BALANCED_ACCURACY + and item["balanced_accuracy"] + >= outcome_best + P1.MIN_EFFICACY_DELTA_OVER_OUTCOME + ) + else: + unavailable = { + "status": "UNAVAILABLE", + "reason": "each leave-one-repetition-out train fold needs both classes", + } + outcome_cv = unavailable + telemetry_cv = unavailable + efficacy_qualifying = [] + transitions = defaultdict(int) + for pair in actions: + transitions[pair["full_feasibility_transition"]] += 1 + admitted = [float(trial["outcome"]["admitted_fraction"]) for trial in trials] + completed = [ + float(trial["outcome"]["admitted_fraction"]) + * float(trial["outcome"]["completed_over_admitted"]) + for trial in trials + ] + state_vectors = { + tuple(round(float(trial["state"][feature]), 12) for feature in P1.V0.ALL_FEATURES) + for trial in trials + } + per_cell_vectors: dict[str, set[tuple[float, ...]]] = defaultdict(set) + for trial in trials: + per_cell_vectors[str(trial["cell"])].add( + tuple( + round(float(trial["state"][feature]), 12) + for feature in P1.V0.ALL_FEATURES + ) + ) + ratio_features = ( + "prefill_token_fraction", + "kv_usage_mean", + "kv_usage_max", + "graph_none_share", + "graph_full_share", + "graph_padding_fraction", + ) + nonnegative_features = tuple( + feature + for feature in P1.V0.ALL_FEATURES + if feature != "kv_usage_end_minus_start" + ) + action_metadata = [ + { + "group": pair["group"], + "source": pair["source"], + "target": pair["target"], + "full_action_efficacy": pair["full_action_efficacy"], + "full_feasibility_transition": pair["full_feasibility_transition"], + "delta_state": pair["delta_state"], + "delta_outcome": pair["delta_outcome"], + } + for pair in actions + ] + invariants = { + "expected_action_pair_count": len(actions) == EXPECTED_ACTION_PAIRS, + "expected_repeat_pair_count": len(repeats) == EXPECTED_REPEAT_PAIRS, + "finite_deltas": all( + math.isfinite(value) + for pair in [*actions, *repeats] + for key in ("delta_state", "delta_outcome") + for value in pair[key].values() + ), + "all_results_uncensored": all(not trial["early_stopped"] for trial in trials), + "state_vectors_not_all_identical": len(state_vectors) > 1, + "per_cell_state_vectors_not_all_identical": all( + len(vectors) > 1 for vectors in per_cell_vectors.values() + ), + "ratios_bounded": all( + 0.0 <= float(trial["state"][feature]) <= 1.0 + for trial in trials + for feature in ratio_features + ), + "nonnegative_counters": all( + float(trial["state"][feature]) >= 0.0 + for trial in trials + for feature in nonnegative_features + ), + "layer1_boundary_and_internal_gaps_bounded": all( + value <= MAX_LAYER1_GAP_S + for trial in trials + for value in trial["telemetry_coverage"].values() + ), + } + return { + "start_s": start_s, + "end_s": end_s, + "end_fraction": fraction, + "coverage_at_end": { + "admitted_fraction_of_total": numeric(admitted), + "completed_fraction_of_total": numeric(completed), + }, + "action_pairs": len(actions), + "repeat_pairs": len(repeats), + "actions": action_metadata, + "trial_sanity": [ + { + "trial_id": trial["trial_id"], + "cell": trial["cell"], + "level": trial["level"], + "replicate": trial["replicate"], + "admitted_fraction": trial["outcome"]["admitted_fraction"], + "completed_fraction": ( + trial["outcome"]["admitted_fraction"] + * trial["outcome"]["completed_over_admitted"] + ), + "telemetry_coverage": trial["telemetry_coverage"], + } + for trial in trials + ], + "response_statistics": response, + "qualifying_response_features": response_qualifying, + "efficacy": { + "labels": numeric(labels), + "positive": sum(labels), + "negative": len(labels) - sum(labels), + "label_balance_sufficient": ( + sum(labels) >= MIN_EFFICACY_CLASS + and len(labels) - sum(labels) >= MIN_EFFICACY_CLASS + ), + "cross_validation_possible": cross_validation_possible, + "transitions": dict(sorted(transitions.items())), + "outcome_delta": outcome_cv, + "telemetry_delta": telemetry_cv, + "telemetry_qualifying_features": efficacy_qualifying, + }, + "sanity": { + "trials": len(trials), + "invariants": invariants, + "red_flags": [name for name, passed in invariants.items() if not passed], + }, + } + + +def stable_adjacent_features(windows: list[dict[str, Any]]) -> dict[str, list[str]]: + result = {} + for left, right in zip(windows, windows[1:], strict=False): + key = f"{left['end_fraction']:.2f}->{right['end_fraction']:.2f}" + result[key] = sorted( + set(left["qualifying_response_features"]) + & set(right["qualifying_response_features"]) + ) + return result + + +def stable_adjacent_efficacy_features( + windows: list[dict[str, Any]], +) -> dict[str, list[str]]: + eligible = [window for window in windows if window["end_fraction"] >= 0.25] + result = {} + for left, right in zip(eligible, eligible[1:], strict=False): + key = f"{left['end_fraction']:.2f}->{right['end_fraction']:.2f}" + result[key] = sorted( + set(left["efficacy"]["telemetry_qualifying_features"]) + & set(right["efficacy"]["telemetry_qualifying_features"]) + ) + return result + + +def consistent_load_regimes( + windows: list[dict[str, Any]], stable: dict[str, list[str]] +) -> dict[str, Any]: + by_end = {float(window["end_fraction"]): window for window in windows} + result = {} + for transition, features in stable.items(): + _left_text, right_text = transition.split("->") + window = by_end[float(right_text)] + for feature in features: + deltas_by_level: dict[str, list[float]] = defaultdict(list) + all_deltas = [] + for action in window["actions"]: + value = float(action["delta_state"][feature]) + deltas_by_level[str(action["group"]["level"])].append(value) + all_deltas.append(value) + positive = sum(value > 1e-12 for value in all_deltas) + negative = sum(value < -1e-12 for value in all_deltas) + direction = 1 if positive >= negative else -1 + consistent = [] + for level, values in sorted(deltas_by_level.items()): + matching = sum(direction * value > 1e-12 for value in values) + nonzero = sum(abs(value) > 1e-12 for value in values) + if nonzero and matching / nonzero >= 2.0 / 3.0: + consistent.append(level) + result[f"{transition}:{feature}"] = { + "direction": direction, + "consistent_load_regimes": consistent, + "passes_two_regimes": len(consistent) >= 2, + } + return result + + +def mechanism_gate( + stable: Mapping[str, list[str]], load_consistency: Mapping[str, Mapping[str, Any]] +) -> dict[str, Any]: + by_transition = {} + for transition, features in stable.items(): + qualifying = sorted( + feature + for feature in features + if load_consistency[f"{transition}:{feature}"]["passes_two_regimes"] + ) + by_transition[transition] = qualifying + passing_transitions = sorted( + transition + for transition, features in by_transition.items() + if len(features) >= 2 + ) + return { + "minimum_features": 2, + "by_transition": by_transition, + "passing_transitions": passing_transitions, + "passes": bool(passing_transitions), + } + + +def controller_gate( + run_root: Path, manifest: Mapping[str, Any] +) -> dict[str, Any]: + path = run_root / "controller-state.json" + state = json.loads(path.read_text(encoding="utf-8")) + expected_sessions = {str(session["session"]) for session in manifest["sessions"]} + actual_sessions = set(state.get("sessions", {})) + session_invariants = [ + passed + for session in state.get("sessions", {}).values() + for passed in session.get("validation", {}).get("invariants", {}).values() + ] + invariants = { + "controller_complete": state.get("status") == "complete", + "completed_session_count": int(state.get("completed_sessions", -1)) + == len(expected_sessions), + "exact_session_set": actual_sessions == expected_sessions, + "all_sessions_complete": all( + session.get("status") == "complete" + for session in state.get("sessions", {}).values() + ), + "no_controller_failures": not state.get("failures"), + "under_h20_hour_cap": float(state.get("gpu_hours_total", math.inf)) + <= float(manifest["budget"]["hard_cap_h20_hours"]), + "all_stream_validation_invariants_pass": bool(session_invariants) + and all(session_invariants), + } + return { + "path": str(path.resolve()), + "sha256": sha256_file(path), + "gpu_hours_total": float(state.get("gpu_hours_total", math.nan)), + "invariants": invariants, + "red_flags": [name for name, passed in invariants.items() if not passed], + } + + +def cumulative_coverage_gate(windows: list[dict[str, Any]]) -> dict[str, Any]: + trajectories: dict[str, list[tuple[float, float]]] = defaultdict(list) + trial_sets = [] + for window in windows: + trial_sets.append({str(trial["trial_id"]) for trial in window["trial_sanity"]}) + for trial in window["trial_sanity"]: + trajectories[str(trial["trial_id"])].append( + ( + float(trial["admitted_fraction"]), + float(trial["completed_fraction"]), + ) + ) + invariants = { + "same_trials_at_every_checkpoint": bool(trial_sets) + and all(trial_set == trial_sets[0] for trial_set in trial_sets[1:]), + "admitted_fraction_monotonic": all( + all(right[0] + 1e-12 >= left[0] for left, right in zip(values, values[1:])) + for values in trajectories.values() + ), + "completed_fraction_monotonic": all( + all(right[1] + 1e-12 >= left[1] for left, right in zip(values, values[1:])) + for values in trajectories.values() + ), + } + return { + "invariants": invariants, + "red_flags": [name for name, passed in invariants.items() if not passed], + } + + +def audit(*, run_root: Path, manifest_path: Path, output_path: Path) -> dict[str, Any]: + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + if manifest.get("schema") != "intervention-response-phase-aware-pilot-manifest-v3": + raise ValueError("unexpected phase-aware pilot manifest schema") + fractions = [float(value) for value in manifest["checkpoints"]["fractions"]] + seconds = [float(value) for value in manifest["checkpoints"]["seconds"]] + cumulative_intervals = tuple((0.0, end_s) for end_s in seconds) + quarter_intervals = ((0.0, 75.0), (75.0, 150.0), (150.0, 225.0), (225.0, 300.0)) + intervals = tuple(dict.fromkeys([*cumulative_intervals, *quarter_intervals])) + trials_by_interval, streams = load_interval_trials( + run_root=run_root, manifest=manifest, intervals_s=intervals + ) + cumulative = [ + analyze_window( + trials_by_interval[interval], + start_s=interval[0], + end_s=interval[1], + fraction=fraction, + ) + for fraction, interval in zip(fractions, cumulative_intervals, strict=True) + ] + quarter_blocks = [ + analyze_window( + trials_by_interval[interval], + start_s=interval[0], + end_s=interval[1], + fraction=interval[1] / 300.0, + ) + for interval in quarter_intervals + ] + stable = stable_adjacent_features(cumulative) + load_consistency = consistent_load_regimes(cumulative, stable) + mechanism = mechanism_gate(stable, load_consistency) + mechanism_features = sorted( + { + feature + for transition in mechanism["passing_transitions"] + for feature in mechanism["by_transition"][transition] + } + ) + full = cumulative[-1] + efficacy_stable = stable_adjacent_efficacy_features(cumulative) + efficacy_candidates = sorted( + {feature for features in efficacy_stable.values() for feature in features} + ) + efficacy_features = sorted(set(efficacy_candidates) & set(mechanism_features)) + controller = controller_gate(run_root, manifest) + coverage = cumulative_coverage_gate(cumulative) + red_flags = sorted( + { + flag + for window in [*cumulative, *quarter_blocks] + for flag in window["sanity"]["red_flags"] + } + | set(controller["red_flags"]) + | set(coverage["red_flags"]) + ) + if red_flags: + decision = "STOP_DATA_INVALID" + elif not mechanism["passes"]: + decision = "STOP_NO_PHASE_STABLE_RESPONSE" + elif not full["efficacy"]["label_balance_sufficient"]: + decision = "MECHANISM_ONLY_NO_LABEL_BALANCE" + elif not efficacy_features: + decision = "STOP_NO_INCREMENTAL_TUNING_SIGNAL" + else: + decision = "OPEN_E2E_POLICY_TEST" + payload = { + "schema": SCHEMA, + "status": "COMPLETE", + "decision": decision, + "claim_boundary": "Development mechanism pilot; not a held-out paper claim.", + "mechanism_features": mechanism_features, + "mechanism_gate": mechanism, + "stable_adjacent_features": stable, + "load_consistency": load_consistency, + "stable_incremental_efficacy_features": efficacy_features, + "stable_incremental_efficacy_candidates": efficacy_candidates, + "stable_adjacent_efficacy_features": efficacy_stable, + "cumulative": cumulative, + "quarter_blocks": quarter_blocks, + "provenance": { + "analysis_script": str(Path(__file__).resolve()), + "analysis_script_sha256": sha256_file(Path(__file__).resolve()), + "manifest": str(manifest_path.resolve()), + "manifest_sha256": sha256_file(manifest_path), + "run_root": str(run_root.resolve()), + "streams": streams, + }, + "sanity": { + "streams": len(streams), + "stream_bytes": numeric(item["bytes"] for item in streams), + "controller": controller, + "cumulative_coverage": coverage, + "red_flags": red_flags, + }, + } + output_path.parent.mkdir(parents=True, exist_ok=True) + output_path.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + return payload + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--run-root", type=Path, required=True) + parser.add_argument("--manifest", type=Path, required=True) + parser.add_argument("--output", type=Path, required=True) + args = parser.parse_args() + payload = audit( + run_root=args.run_root, + manifest_path=args.manifest, + output_path=args.output, + ) + print( + json.dumps( + { + "decision": payload["decision"], + "mechanism_features": payload["mechanism_features"], + "stable_incremental_efficacy_features": payload[ + "stable_incremental_efficacy_features" + ], + "sanity": payload["sanity"], + }, + indent=2, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v2/existing-phase-audit.json b/runs/intervention-response-v2/existing-phase-audit.json new file mode 100644 index 0000000..3884878 --- /dev/null +++ b/runs/intervention-response-v2/existing-phase-audit.json @@ -0,0 +1,9915 @@ +{ + "blocks": [ + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 21, + "max": 0.12295081967213115, + "median": 0.08985507246376812, + "min": 0.04918032786885246, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 30, + "max": 0.09361702127659574, + "median": 0.050437283852088316, + "min": 0.005586592178770949, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.4166666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 0 + ], + "threshold": 0.051724137931034475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.09339080459770116, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.018181818181818188, + "train_balanced_accuracy": 0.625 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.4166666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 0 + ], + "threshold": -0.051724137931034475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.09339080459770113, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.01818181818181816, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.015623848188639142, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0010320465723873684, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.009580791853053022, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0002360795908454838, + "train_balanced_accuracy": 0.625 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.0072977563346891365, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.021249613758603417, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.007412010668000825, + "train_balanced_accuracy": 0.875 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": -0.0008523871757378475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.001080823830574982, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.006279735096096474, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.6666666666666666, + "best_balanced_accuracy": 0.6666666666666666, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.5847120251326046, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.1503494070398912, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.759913509761732, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0003962349313431858, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0006270962078854048, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.003423657452078254, + "train_balanced_accuracy": 0.75 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00016329402239600305, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00013516820682281123, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.00013516820682281123, + "train_balanced_accuracy": 0.625 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": -0.0003659931906908054, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0003153433269442174, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": 0.002846241748737599, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 1.2024471918333335, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.18325025108333315, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 1.2024471918333335, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -1.6539956596666667, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.009266072583333333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -1.6539956596666667, + "train_balanced_accuracy": 0.875 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 1.5833333333333357, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 4.416666666666664, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -19.583333333333332, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.1, + "end_s": 6.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.587816632458785, + "action_delta": { + "distinct_n": 12, + "max": 10.944163150492265, + "min": -0.9368137011239002, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.1464621283214008, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 4.013437734353163, + "repeat_abs_p95": 8.899704585700528, + "repeat_delta": { + "distinct_n": 24, + "max": 9.092698412698415, + "min": -9.933716475095787, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 7.901570082144033, + "action_delta": { + "distinct_n": 12, + "max": 564.520698546648, + "min": -8.58386322115848, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.05293318721992956, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 149.27440604160446, + "repeat_abs_p95": 426.8141407023969, + "repeat_delta": { + "distinct_n": 24, + "max": 449.2116931216931, + "min": -584.6987867177522, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.582158519407685, + "action_delta": { + "distinct_n": 12, + "max": 10.735314111579934, + "min": -0.9288831363334129, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.14646507698937813, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.9747257938484823, + "repeat_abs_p95": 8.770067452024344, + "repeat_delta": { + "distinct_n": 24, + "max": 8.954735449735448, + "min": -9.735983397190292, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.005042559238173261, + "action_delta": { + "distinct_n": 12, + "max": 0.008374384236453203, + "min": -0.10829817158931077, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.12202178454078347, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.04132507369196753, + "repeat_abs_p95": 0.08139543508103343, + "repeat_delta": { + "distinct_n": 24, + "max": 0.12107279693486583, + "min": -0.0822222222222222, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.008375994890060249, + "action_delta": { + "distinct_n": 12, + "max": 0.09274496015002344, + "min": -0.01986863711001642, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 9, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.22398385207837773, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.037395530134597704, + "repeat_abs_p95": 0.07586812708364432, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07631077509373656, + "min": -0.10941890166028097, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0011703901376772582, + "action_delta": { + "distinct_n": 12, + "max": 0.006703036588479568, + "min": -0.0015820428864207082, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.29946486376659237, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.003908271985422232, + "repeat_abs_p95": 0.008187883039017584, + "repeat_delta": { + "distinct_n": 24, + "max": 0.007141591578301508, + "min": -0.009201187757062074, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.006820888350695331, + "action_delta": { + "distinct_n": 12, + "max": 0.140893666867539, + "min": -0.005054735620165451, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 0.6661896437498689, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010238658638254572, + "repeat_abs_p95": 0.1277225497052652, + "repeat_delta": { + "distinct_n": 24, + "max": 0.13402937356379263, + "min": -0.1168947946847837, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.007604372340016818, + "action_delta": { + "distinct_n": 12, + "max": 0.06684057734774362, + "min": 0.00016629145416469093, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 0.3770466591391387, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.020168252802926, + "repeat_abs_p95": 0.11084274153262068, + "repeat_delta": { + "distinct_n": 24, + "max": 0.12803476870816266, + "min": -0.11234888600259774, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0014831270777658032, + "action_delta": { + "distinct_n": 12, + "max": 0.02895016189128262, + "min": -0.00029784868681948656, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.12120157645159224, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.01223686292857884, + "repeat_abs_p95": 0.07958153755957757, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08820147130614941, + "min": -0.08208138515366167, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0028294297587345696, + "action_delta": { + "distinct_n": 12, + "max": 0.013691652669458043, + "min": -0.002949832419021914, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.44362661948582977, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.006377953067861264, + "repeat_abs_p95": 0.031272369907012425, + "repeat_delta": { + "distinct_n": 24, + "max": 0.036461564751038456, + "min": -0.018166035027340932, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.50741128125, + "action_delta": { + "distinct_n": 12, + "max": 15.614046583666667, + "min": -0.30207624649999953, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.12962585804860988, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.9144294887499997, + "repeat_abs_p95": 7.101594725466666, + "repeat_delta": { + "distinct_n": 24, + "max": 6.513803891666667, + "min": -13.5536487565, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.24895167858333334, + "action_delta": { + "distinct_n": 10, + "max": 0.0, + "min": -8.040727011166666, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 9, + "positive": 0, + "zero": 3 + }, + "effect_to_repeat_median": 1.1132189691456902, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.22363226416666668, + "repeat_abs_p95": 2.296808863758333, + "repeat_delta": { + "distinct_n": 16, + "max": 2.3311656403333334, + "min": -5.974843045, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 3.249999999999993, + "action_delta": { + "distinct_n": 12, + "max": 10.5, + "min": -30.999999999999993, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.1065573770491801, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 30.5, + "repeat_abs_p95": 59.92499999999999, + "repeat_delta": { + "distinct_n": 24, + "max": 59.5, + "min": -64.5, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 0.0 + }, + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 22, + "max": 0.20535714285714285, + "median": 0.18235294117647058, + "min": 0.1564245810055866, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 31, + "max": 0.180327868852459, + "median": 0.14411764705882352, + "min": 0.08933333333333335, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.012666666666666666, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.008666666666666666, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.008666666666666666, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.03401360544217691, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": 0.006172839506172811, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.04871948779511809, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.11875000000000002, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.07738095238095238, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.07738095238095238, + "train_balanced_accuracy": 0.75 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.03401360544217688, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": -0.006172839506172839, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.04871948779511806, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.028440662518539606, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.02030446551608167, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.008882265987138677, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": -0.0020290170113650557, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.04992359808650447, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.009324194656287296, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.01753702182516767, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.01753702182516767, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.018086205842943556, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.010556909058063094, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.003084423316020539, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.027367186208561683, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.22344269117369397, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.22344269117369397, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.5979430550644496, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.006014036688642874, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.003996367458148475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 0.004817288488210095, + "train_balanced_accuracy": 0.875 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": 0.0009690245761390937, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.014778075272949286, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00021729561459569828, + "train_balanced_accuracy": 0.75 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.0004249745076460587, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0008306228170546492, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": -0.005838187880452539, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": -0.014251387166666962, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 9.969867595499998, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.36342276333333334, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.37299224008333337, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.09157276941666666, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.08574448658333333, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.75, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.75, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 6.833333333333329, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.2, + "end_s": 12.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.836467208595514, + "action_delta": { + "distinct_n": 12, + "max": 38.01660606060606, + "min": -3.7894183532722323, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.6942297063022367, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 1.2048853585521353, + "repeat_abs_p95": 9.90189234803657, + "repeat_delta": { + "distinct_n": 24, + "max": 6.213858257127487, + "min": -19.9836254925941, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 68.40350288667949, + "action_delta": { + "distinct_n": 12, + "max": 1287.087393939394, + "min": -109.74922437881139, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.6828036858921454, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 40.648534026959354, + "repeat_abs_p95": 252.72760702584105, + "repeat_delta": { + "distinct_n": 24, + "max": 53.09534549057537, + "min": -841.3287810843865, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.79935614604261, + "action_delta": { + "distinct_n": 12, + "max": 37.563515151515155, + "min": -3.7555281535670666, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.6727103099461282, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.1882620709449565, + "repeat_abs_p95": 9.833929419939027, + "repeat_delta": { + "distinct_n": 24, + "max": 6.226902904787521, + "min": -19.76212800652263, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.023223408379118127, + "action_delta": { + "distinct_n": 12, + "max": 0.03622200353327565, + "min": -0.23381818181818181, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 2.5685579123999758, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.009041419026218855, + "repeat_abs_p95": 0.07343409055233649, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08839516238619383, + "min": -0.027874161461947766, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.0196424445317687, + "action_delta": { + "distinct_n": 12, + "max": 0.22624242424242422, + "min": -0.043130466400287726, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 1.9837484419794449, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.009901681138647232, + "repeat_abs_p95": 0.07568864235997094, + "repeat_delta": { + "distinct_n": 24, + "max": 0.02793199167244964, + "min": -0.10324092947411331, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0019741621395090708, + "action_delta": { + "distinct_n": 12, + "max": 0.013535143477830873, + "min": -0.0002595396932579229, + "n": 12 + }, + "action_signs": { + "consistency": 0.9166666666666666, + "negative": 1, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 0.6824854579006991, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.002892606892433317, + "repeat_abs_p95": 0.013935039231725855, + "repeat_delta": { + "distinct_n": 24, + "max": 0.016710257140637542, + "min": -0.0048630254047383765, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.016292713617491184, + "action_delta": { + "distinct_n": 12, + "max": 0.06869972057736717, + "min": -0.13905774767384216, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 1.2163146605755397, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.013395146951350356, + "repeat_abs_p95": 0.10929110116632139, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07238485363173153, + "min": -0.2557698071735438, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.01447565921273003, + "action_delta": { + "distinct_n": 12, + "max": 0.10932093798799158, + "min": 0.0002683613903166071, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 1.4049345895249048, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010303439975540174, + "repeat_abs_p95": 0.07329653312019176, + "repeat_delta": { + "distinct_n": 24, + "max": 0.05015486062543706, + "min": -0.12688580277750028, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.005335907884240179, + "action_delta": { + "distinct_n": 12, + "max": 0.06835893039525985, + "min": -0.0020715594873930866, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 1.4198198744789212, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.003758158327089537, + "repeat_abs_p95": 0.054511554710704666, + "repeat_delta": { + "distinct_n": 24, + "max": 0.05530066409194215, + "min": -0.1166293577215588, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00639806071748561, + "action_delta": { + "distinct_n": 12, + "max": 0.01743846124119386, + "min": -0.012883247545976029, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.5860208014218505, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.010917804797990316, + "repeat_abs_p95": 0.038061500221448455, + "repeat_delta": { + "distinct_n": 24, + "max": 0.006259881301210357, + "min": -0.03966667403951463, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.8894945159166667, + "action_delta": { + "distinct_n": 12, + "max": 39.57822979983333, + "min": -2.9770223346666658, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.4702668338949774, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.8914676770833334, + "repeat_abs_p95": 10.096755181974991, + "repeat_delta": { + "distinct_n": 21, + "max": 6.583703286333334, + "min": -20.7125852155, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.79726731775, + "action_delta": { + "distinct_n": 11, + "max": 0.0, + "min": -30.713414816833335, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 10, + "positive": 0, + "zero": 2 + }, + "effect_to_repeat_median": 17.026052234021265, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.04682631691666666, + "repeat_abs_p95": 3.5462989908416667, + "repeat_delta": { + "distinct_n": 17, + "max": 3.561578917166667, + "min": -17.805020274833332, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 13.083333333333334, + "action_delta": { + "distinct_n": 12, + "max": 23.66666666666667, + "min": -40.5, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 1.2661290322580652, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 10.333333333333329, + "repeat_abs_p95": 55.25833333333328, + "repeat_delta": { + "distinct_n": 24, + "max": 71.83333333333333, + "min": -19.333333333333343, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 6.0 + }, + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 22, + "max": 0.30654761904761907, + "median": 0.2795403407264545, + "min": 0.24313725490196078, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 30, + "max": 0.2681564245810056, + "median": 0.22518087736972248, + "min": 0.12849162011173185, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "completed_pass_rate", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.02825325884543762, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.02266666666666667, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.040253258845437616, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.03272660415517559, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.022522522522522515, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.07358436332279389, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.14270516717325227, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.12142857142857144, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.2182462927143778, + "train_balanced_accuracy": 0.75 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.03272660415517557, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.022522522522522515, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0735843633227939, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.37412821213021774, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.009857117327656123, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.008882265987138677, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": -0.004552157689988692, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.007167514027906635, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.006851562902858294, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.017755539496041216, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.007733676747496546, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.5040019049192779, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.007216190772090309, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.007216190772090309, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.24525856553394837, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "graph_padding_fraction", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 1.367157621094477, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": -0.45324800586584313, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.7719319083901095, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.0006822319057190087, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.0006574027141346845, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.0007737011910822953, + "train_balanced_accuracy": 0.75 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.0014242600332016828, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": -0.00020601061133531957, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.0014242600332016828, + "train_balanced_accuracy": 0.75 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 0.0001890219308638419, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": 6.874128372463817e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 0 + ], + "threshold": 0.0012836451412676642, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 2.090120409833334, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": -0.4494322117499996, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 1.383383161750001, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.015561036916666665, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.015561036916666665, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.35703936641666667, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -5.666666666666664, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": 3.0, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -2.6666666666666607, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.3, + "end_s": 18.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 1.4376624963208817, + "action_delta": { + "distinct_n": 12, + "max": 44.66904761904762, + "min": -3.1358226741876534, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.5185498088911062, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.9467338429752983, + "repeat_abs_p95": 8.569947817628815, + "repeat_delta": { + "distinct_n": 24, + "max": 4.8797843665768195, + "min": -24.305454429328734, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 29.54235812446782, + "action_delta": { + "distinct_n": 12, + "max": 981.3419420915683, + "min": -40.25177564153491, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.44186157650879665, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 66.85885285135149, + "repeat_abs_p95": 331.06991793371697, + "repeat_delta": { + "distinct_n": 24, + "max": 347.23329948092976, + "min": -494.75378424985627, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 1.426379390321979, + "action_delta": { + "distinct_n": 12, + "max": 44.29932712215321, + "min": -3.1155318979899898, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.5217762546171796, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.9373121613602792, + "repeat_abs_p95": 8.519190735455247, + "repeat_delta": { + "distinct_n": 24, + "max": 4.8432704402515725, + "min": -24.101136871686787, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.010376892235975155, + "action_delta": { + "distinct_n": 12, + "max": 0.020290776197664284, + "min": -0.19708796988318666, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.5862533874020509, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0177003535654775, + "repeat_abs_p95": 0.05622040303840937, + "repeat_delta": { + "distinct_n": 24, + "max": 0.14549402823018454, + "min": -0.05811329271044907, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.015566450509702876, + "action_delta": { + "distinct_n": 12, + "max": 0.1830537562974035, + "min": -0.028434098673234293, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.9711084270914145, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.016029570000052673, + "repeat_abs_p95": 0.06874268847661695, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07111825772963212, + "min": -0.1536373507057546, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.0019204725324846257, + "action_delta": { + "distinct_n": 12, + "max": 0.011531473650189953, + "min": 1.5710198019351293e-05, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 1.109916935840515, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0017302849163485332, + "repeat_abs_p95": 0.011201032721421313, + "repeat_delta": { + "distinct_n": 24, + "max": 0.012042080910355725, + "min": -0.011611080181771623, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.012368563885621575, + "action_delta": { + "distinct_n": 12, + "max": 0.1693202894846253, + "min": -0.027238473582144707, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 1.1201833133796462, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.01104155341174029, + "repeat_abs_p95": 0.14212743371007872, + "repeat_delta": { + "distinct_n": 24, + "max": 0.1535118393445899, + "min": -0.2053651713457888, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.005959053027072225, + "action_delta": { + "distinct_n": 12, + "max": 0.22224967428314468, + "min": 3.590298430544703e-05, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 0.49652693165856604, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.012001469904496409, + "repeat_abs_p95": 0.06026674766281165, + "repeat_delta": { + "distinct_n": 24, + "max": 0.06029573383954434, + "min": -0.267359376561095, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.0057476210577813765, + "action_delta": { + "distinct_n": 12, + "max": 0.18534326857338257, + "min": -0.015438462146793161, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.6334301280804308, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.009073804359762887, + "repeat_abs_p95": 0.0458855138590948, + "repeat_delta": { + "distinct_n": 24, + "max": 0.033340684091908934, + "min": -0.24888103806759748, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.003319188436037235, + "action_delta": { + "distinct_n": 12, + "max": 0.013152637562827363, + "min": -0.02492780421592411, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.1361252231813899, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.02438334614602866, + "repeat_abs_p95": 0.07693546674256392, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08157979537248183, + "min": -0.04610629729936333, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 1.8914151395833336, + "action_delta": { + "distinct_n": 12, + "max": 45.162770761333334, + "min": -2.3232337786666672, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 1.438574132498461, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.3147846168333333, + "repeat_abs_p95": 6.971346508883332, + "repeat_delta": { + "distinct_n": 21, + "max": 4.9605220578333356, + "min": -20.271844710166665, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 1.12754089475, + "action_delta": { + "distinct_n": 10, + "max": 0.0, + "min": -47.07583785533333, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 9, + "positive": 0, + "zero": 3 + }, + "effect_to_repeat_median": 16.245384499464624, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.06940684566666666, + "repeat_abs_p95": 5.181811176816667, + "repeat_delta": { + "distinct_n": 18, + "max": 0.6568425336666667, + "min": -7.1538662623333344, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 7.166666666666668, + "action_delta": { + "distinct_n": 12, + "max": 34.0, + "min": -47.33333333333333, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.31617647058823534, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 22.666666666666668, + "repeat_abs_p95": 55.666666666666664, + "repeat_delta": { + "distinct_n": 24, + "max": 85.33333333333333, + "min": -55.66666666666667, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 12.0 + } + ], + "claim_boundary": "Post-hoc corrective audit over every common replay decile. It can diagnose horizon sensitivity but cannot establish a held-out tuning claim.", + "cumulative": [ + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 21, + "max": 0.12295081967213115, + "median": 0.08985507246376812, + "min": 0.04918032786885246, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 30, + "max": 0.09361702127659574, + "median": 0.050437283852088316, + "min": 0.005586592178770949, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.4166666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 0 + ], + "threshold": 0.051724137931034475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.09339080459770116, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.018181818181818188, + "train_balanced_accuracy": 0.625 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.4166666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 0 + ], + "threshold": -0.051724137931034475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.09339080459770113, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.01818181818181816, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.000601708027350864, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.015623848188639142, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0010320465723873684, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 0 + ], + "threshold": 0.009580791853053022, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0002360795908454838, + "train_balanced_accuracy": 0.625 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.0072977563346891365, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.021249613758603417, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.007412010668000825, + "train_balanced_accuracy": 0.875 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": -0.0008523871757378475, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.001080823830574982, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.006279735096096474, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.6666666666666666, + "best_balanced_accuracy": 0.6666666666666666, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.5847120251326046, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.1503494070398912, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.759913509761732, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.0003962349313431858, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0006270962078854048, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 0.003423657452078254, + "train_balanced_accuracy": 0.75 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00016329402239600305, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00013516820682281123, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.00013516820682281123, + "train_balanced_accuracy": 0.625 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": -0.0003659931906908054, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0003153433269442174, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 0 + ], + "threshold": 0.002846241748737599, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 1.2024471918333335, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.18325025108333315, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 1.2024471918333335, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -1.6539956596666667, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.009266072583333333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -1.6539956596666667, + "train_balanced_accuracy": 0.875 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": 1.5833333333333357, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 4.416666666666664, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -19.583333333333332, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.1, + "end_s": 6.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.587816632458785, + "action_delta": { + "distinct_n": 12, + "max": 10.944163150492265, + "min": -0.9368137011239002, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.1464621283214008, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 4.013437734353163, + "repeat_abs_p95": 8.899704585700528, + "repeat_delta": { + "distinct_n": 24, + "max": 9.092698412698415, + "min": -9.933716475095787, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 7.901570082144033, + "action_delta": { + "distinct_n": 12, + "max": 564.520698546648, + "min": -8.58386322115848, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.05293318721992956, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 149.27440604160446, + "repeat_abs_p95": 426.8141407023969, + "repeat_delta": { + "distinct_n": 24, + "max": 449.2116931216931, + "min": -584.6987867177522, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.582158519407685, + "action_delta": { + "distinct_n": 12, + "max": 10.735314111579934, + "min": -0.9288831363334129, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.14646507698937813, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.9747257938484823, + "repeat_abs_p95": 8.770067452024344, + "repeat_delta": { + "distinct_n": 24, + "max": 8.954735449735448, + "min": -9.735983397190292, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.005042559238173261, + "action_delta": { + "distinct_n": 12, + "max": 0.008374384236453203, + "min": -0.10829817158931077, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.12202178454078347, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.04132507369196753, + "repeat_abs_p95": 0.08139543508103343, + "repeat_delta": { + "distinct_n": 24, + "max": 0.12107279693486583, + "min": -0.0822222222222222, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.008375994890060249, + "action_delta": { + "distinct_n": 12, + "max": 0.09274496015002344, + "min": -0.01986863711001642, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 9, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.22398385207837773, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.037395530134597704, + "repeat_abs_p95": 0.07586812708364432, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07631077509373656, + "min": -0.10941890166028097, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0011703901376772582, + "action_delta": { + "distinct_n": 12, + "max": 0.006703036588479568, + "min": -0.0015820428864207082, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.29946486376659237, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.003908271985422232, + "repeat_abs_p95": 0.008187883039017584, + "repeat_delta": { + "distinct_n": 24, + "max": 0.007141591578301508, + "min": -0.009201187757062074, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.006820888350695331, + "action_delta": { + "distinct_n": 12, + "max": 0.140893666867539, + "min": -0.005054735620165451, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 0.6661896437498689, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010238658638254572, + "repeat_abs_p95": 0.1277225497052652, + "repeat_delta": { + "distinct_n": 24, + "max": 0.13402937356379263, + "min": -0.1168947946847837, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.007604372340016818, + "action_delta": { + "distinct_n": 12, + "max": 0.06684057734774362, + "min": 0.00016629145416469093, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 0.3770466591391387, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.020168252802926, + "repeat_abs_p95": 0.11084274153262068, + "repeat_delta": { + "distinct_n": 24, + "max": 0.12803476870816266, + "min": -0.11234888600259774, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0014831270777658032, + "action_delta": { + "distinct_n": 12, + "max": 0.02895016189128262, + "min": -0.00029784868681948656, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.12120157645159224, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.01223686292857884, + "repeat_abs_p95": 0.07958153755957757, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08820147130614941, + "min": -0.08208138515366167, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0028294297587345696, + "action_delta": { + "distinct_n": 12, + "max": 0.013691652669458043, + "min": -0.002949832419021914, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.44362661948582977, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.006377953067861264, + "repeat_abs_p95": 0.031272369907012425, + "repeat_delta": { + "distinct_n": 24, + "max": 0.036461564751038456, + "min": -0.018166035027340932, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.50741128125, + "action_delta": { + "distinct_n": 12, + "max": 15.614046583666667, + "min": -0.30207624649999953, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.12962585804860988, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.9144294887499997, + "repeat_abs_p95": 7.101594725466666, + "repeat_delta": { + "distinct_n": 24, + "max": 6.513803891666667, + "min": -13.5536487565, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.24895167858333334, + "action_delta": { + "distinct_n": 10, + "max": 0.0, + "min": -8.040727011166666, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 9, + "positive": 0, + "zero": 3 + }, + "effect_to_repeat_median": 1.1132189691456902, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.22363226416666668, + "repeat_abs_p95": 2.296808863758333, + "repeat_delta": { + "distinct_n": 16, + "max": 2.3311656403333334, + "min": -5.974843045, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 3.249999999999993, + "action_delta": { + "distinct_n": 12, + "max": 10.5, + "min": -30.999999999999993, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.1065573770491801, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 30.5, + "repeat_abs_p95": 59.92499999999999, + "repeat_delta": { + "distinct_n": 24, + "max": 59.5, + "min": -64.5, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 0.0 + }, + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 22, + "max": 0.20535714285714285, + "median": 0.18235294117647058, + "min": 0.1564245810055866, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 31, + "max": 0.180327868852459, + "median": 0.14411764705882352, + "min": 0.08933333333333335, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.012666666666666666, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.008666666666666666, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.008666666666666666, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.03401360544217691, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": 0.006172839506172811, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.04871948779511809, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.11875000000000002, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.07738095238095238, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.07738095238095238, + "train_balanced_accuracy": 0.75 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.03401360544217688, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": -0.006172839506172839, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.04871948779511806, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.028440662518539606, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.02030446551608167, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.008882265987138677, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.4166666666666667, + "balanced_accuracy": 0.41666666666666663, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 0, + 1 + ], + "threshold": -0.0020290170113650557, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.04992359808650447, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.009324194656287296, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.01753702182516767, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.01753702182516767, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 1 + ], + "threshold": -0.018086205842943556, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.010556909058063094, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.003084423316020539, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.027367186208561683, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.6666666666666666, + "best_balanced_accuracy": 0.6666666666666666, + "best_feature": "scheduler_steps_per_s", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": -0.17161348298791945, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.30619421565755833, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.6654132739143297, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.005113976430465315, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -5.156781096974641e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": 0.0022397228564162543, + "train_balanced_accuracy": 0.75 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -1.7899149433659497e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -2.0776132762541973e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00013983916927168966, + "train_balanced_accuracy": 0.75 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 3.837762108915843e-05, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": -0.00021675516061736566, + "train_balanced_accuracy": 0.625 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.837762108915843e-05, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": -0.0830968510000003, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.12225680983333342, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.2734091407916659, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.2763032289583333, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.08217400995833332, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -1.0689820661666667, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.2916666666666643, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.5833333333333286, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 6.291666666666657, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.2, + "end_s": 12.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.6805455173736323, + "action_delta": { + "distinct_n": 12, + "max": 21.022824302134648, + "min": -1.6199557480463245, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.2914327913831549, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 2.335171392840622, + "repeat_abs_p95": 10.342426400114645, + "repeat_delta": { + "distinct_n": 24, + "max": 10.582610461920808, + "min": -14.50489117143556, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 30.73978867286374, + "action_delta": { + "distinct_n": 12, + "max": 837.5693088520675, + "min": -38.259099896431565, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.5085540691379722, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 60.44546792235762, + "repeat_abs_p95": 370.0653796793133, + "repeat_delta": { + "distinct_n": 24, + "max": 385.95091643367505, + "min": -701.2529783740349, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.6733471921553615, + "action_delta": { + "distinct_n": 12, + "max": 20.720637408568443, + "min": -1.606656623670089, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.2892646285380025, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.327789593766041, + "repeat_abs_p95": 10.226231015894982, + "repeat_delta": { + "distinct_n": 24, + "max": 10.460062847993882, + "min": -14.289496209342136, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.008498282584807226, + "action_delta": { + "distinct_n": 12, + "max": 0.024842293569343754, + "min": -0.15582922824302126, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.517404626039514, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.01642482915133081, + "repeat_abs_p95": 0.0799421671518786, + "repeat_delta": { + "distinct_n": 24, + "max": 0.11509974496034647, + "min": -0.08092575333954644, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.01137367331008028, + "action_delta": { + "distinct_n": 12, + "max": 0.14248395282878046, + "min": -0.030463233217211186, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 8, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.6710383267094245, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.016949364674672823, + "repeat_abs_p95": 0.07611571753021439, + "repeat_delta": { + "distinct_n": 24, + "max": 0.0761583865032141, + "min": -0.11305593403905953, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.0022211647342633674, + "action_delta": { + "distinct_n": 12, + "max": 0.007755344475118115, + "min": -0.0006496983398026016, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 1.0130418206479903, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0021925696343341517, + "repeat_abs_p95": 0.006616310818721587, + "repeat_delta": { + "distinct_n": 24, + "max": 0.007943398472261458, + "min": -0.005259691066565701, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.0021058619896323627, + "action_delta": { + "distinct_n": 12, + "max": 0.06906439560383515, + "min": -0.0026342182586551743, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.17774345954986717, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.011847760783802841, + "repeat_abs_p95": 0.0656650541454329, + "repeat_delta": { + "distinct_n": 24, + "max": 0.020281746428214564, + "min": -0.10385652912378862, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.011216828476763308, + "action_delta": { + "distinct_n": 12, + "max": 0.08922043547542868, + "min": 0.0002683613903166071, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 1.0844557892569866, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010343278709820436, + "repeat_abs_p95": 0.07465434288403082, + "repeat_delta": { + "distinct_n": 24, + "max": 0.05015486062543706, + "min": -0.12688580277750028, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0041456974144474384, + "action_delta": { + "distinct_n": 12, + "max": 0.047957700246711504, + "min": -0.00037567596362376825, + "n": 12 + }, + "action_signs": { + "consistency": 0.75, + "negative": 3, + "positive": 9, + "zero": 0 + }, + "effect_to_repeat_median": 0.8378282569881378, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.004948147045494235, + "repeat_abs_p95": 0.06813512586166338, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07105816994418354, + "min": -0.09865864218773521, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0003854997463232923, + "action_delta": { + "distinct_n": 12, + "max": 0.012211161699714035, + "min": -0.0029429188195312372, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.04836213965959768, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.007971106097386826, + "repeat_abs_p95": 0.020160934134987728, + "repeat_delta": { + "distinct_n": 24, + "max": 0.011632967170489006, + "min": -0.023705586156158254, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.5689573993333328, + "action_delta": { + "distinct_n": 12, + "max": 27.59238343825, + "min": -0.6469868719999994, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.26512988706199775, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.1459572349166667, + "repeat_abs_p95": 7.856052630799997, + "repeat_delta": { + "distinct_n": 24, + "max": 6.557133349166667, + "min": -17.1293622325, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.40246922141666663, + "action_delta": { + "distinct_n": 11, + "max": 0.0, + "min": -19.377070914, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 10, + "positive": 0, + "zero": 2 + }, + "effect_to_repeat_median": 4.226054832868273, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.09523521045833332, + "repeat_abs_p95": 2.510103133283333, + "repeat_delta": { + "distinct_n": 18, + "max": 2.268429387666668, + "min": -11.883906298916667, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 7.958333333333329, + "action_delta": { + "distinct_n": 12, + "max": 11.75, + "min": -35.16666666666667, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 7, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.38047808764940205, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 20.91666666666667, + "repeat_abs_p95": 51.31666666666662, + "repeat_delta": { + "distinct_n": 24, + "max": 65.66666666666667, + "min": -34.58333333333333, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 0.0 + }, + { + "action_pairs": 12, + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 22, + "max": 0.30654761904761907, + "median": 0.2795403407264545, + "min": 0.24313725490196078, + "n": 36 + }, + "completed_fraction_of_total": { + "distinct_n": 30, + "max": 0.2681564245810056, + "median": 0.22518087736972248, + "min": 0.12849162011173185, + "n": 36 + } + }, + "efficacy": { + "feasibility_transitions": { + "false->false": 3, + "false->true": 6, + "true->true": 3 + }, + "minimum_balanced_accuracy": 0.75, + "minimum_delta_over_best_outcome": 0.15, + "outcome_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "completed_pass_rate", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.02825325884543762, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.02266666666666667, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.040253258845437616, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.03272660415517559, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.022522522522522515, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.07358436332279389, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.14270516717325227, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.12142857142857144, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 0.2182462927143778, + "train_balanced_accuracy": 0.75 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.03272660415517557, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": -0.022522522522522515, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.0735843633227939, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": 0.37412821213021774, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.009857117327656123, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": 0.008882265987138677, + "train_balanced_accuracy": 0.625 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666667, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 0, + 1 + ], + "threshold": -0.004552157689988692, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.007167514027906635, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.006851562902858294, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 0 + ], + "threshold": -0.017755539496041216, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.007733676747496546, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.5040019049192779, + "train_balanced_accuracy": 0.75 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.007216190772090309, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.007216190772090309, + "train_balanced_accuracy": 0.875 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.24525856553394837, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_delta": { + "best_accuracy": 0.75, + "best_balanced_accuracy": 0.75, + "best_feature": "scheduler_steps_per_s", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.44507332514357945, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": -0.24667983170250252, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.5943046676188155, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333334, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.004225031845271454, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": 0.0004360066659713927, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 0.0034270719814960106, + "train_balanced_accuracy": 0.875 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.00013357357159209972, + "train_balanced_accuracy": 0.625 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.00013492200088597965, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": -0.00018506887864057613, + "train_balanced_accuracy": 0.625 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5833333333333334, + "balanced_accuracy": 0.5833333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 0, + 1 + ], + "threshold": 0.0008649236257326498, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": 0.0017369622931825424, + "train_balanced_accuracy": 0.875 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 1, + 1, + 1 + ], + "threshold": 0.0005824031404205732, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 11.957967036527776, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 0 + ], + "threshold": -0.23078758088888907, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 1, + 0 + ], + "threshold": 18.57988346463889, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.18938916494444447, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 1, + 1, + 1 + ], + "threshold": -0.12479013208333334, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 1, + 0, + 1, + 1 + ], + "threshold": -0.23008102869444447, + "train_balanced_accuracy": 0.75 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.75, + "balanced_accuracy": 0.75, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 4.277777777777775, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 3.888888888888893, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 0, + 1, + 1 + ], + "test_predictions": [ + 0, + 0, + 0, + 1 + ], + "threshold": 4.277777777777775, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 12 + }, + "negative": 6, + "positive": 6 + }, + "telemetry_qualifying_features": [] + }, + "end_fraction": 0.3, + "end_s": 18.0, + "qualifying_response_features": [], + "repeat_pairs": 24, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.6066330085464959, + "action_delta": { + "distinct_n": 12, + "max": 28.73606563075747, + "min": -1.1485291194593508, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.4934987464973646, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 1.2292493402508278, + "repeat_abs_p95": 8.067582719584564, + "repeat_delta": { + "distinct_n": 24, + "max": 8.686808262759335, + "min": -14.918642465494425, + "n": 24 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 10.317307738579501, + "action_delta": { + "distinct_n": 12, + "max": 801.7642402753155, + "min": -13.613525901316592, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.17112975179373358, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 60.28938644763055, + "repeat_abs_p95": 340.3503302119946, + "repeat_delta": { + "distinct_n": 24, + "max": 374.64907917730886, + "min": -561.6272830402723, + "n": 24 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.6035604856985604, + "action_delta": { + "distinct_n": 12, + "max": 28.411725240727222, + "min": -1.1421295113155576, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.4962266942146829, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.2162999144045272, + "repeat_abs_p95": 7.9623825938693935, + "repeat_delta": { + "distinct_n": 24, + "max": 8.570025092978447, + "min": -14.737105312913595, + "n": 24 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.25, + "action_abs_median": 0.006690006870484877, + "action_delta": { + "distinct_n": 12, + "max": 0.013223015997172571, + "min": -0.1490562102409011, + "n": 12 + }, + "action_signs": { + "consistency": 0.5833333333333334, + "negative": 5, + "positive": 7, + "zero": 0 + }, + "effect_to_repeat_median": 0.6364709072202164, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010511096099747042, + "repeat_abs_p95": 0.06858287350501925, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08329551900170162, + "min": -0.0738002419680065, + "n": 24 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.01055037975803224, + "action_delta": { + "distinct_n": 12, + "max": 0.13907776271422118, + "min": -0.018167596600093858, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 8, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.9080980815382388, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.01161810598714273, + "repeat_abs_p95": 0.06969738634332051, + "repeat_delta": { + "distinct_n": 24, + "max": 0.07468521754716137, + "min": -0.08558328606541879, + "n": 24 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.0024714849986972203, + "action_delta": { + "distinct_n": 12, + "max": 0.006494393749041705, + "min": 0.000227090832439927, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 1.373420385875363, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0017995109320603211, + "repeat_abs_p95": 0.005547351274885883, + "repeat_delta": { + "distinct_n": 24, + "max": 0.005694038006766439, + "min": -0.0035699364880986415, + "n": 24 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.010270226283016048, + "action_delta": { + "distinct_n": 12, + "max": 0.23522057920397632, + "min": -0.003022494939389775, + "n": 12 + }, + "action_signs": { + "consistency": 0.9166666666666666, + "negative": 1, + "positive": 11, + "zero": 0 + }, + "effect_to_repeat_median": 1.07628737944467, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.009542271403679525, + "repeat_abs_p95": 0.08564791687481275, + "repeat_delta": { + "distinct_n": 24, + "max": 0.08662204016385266, + "min": -0.31416724947547214, + "n": 24 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.015846588112988946, + "action_delta": { + "distinct_n": 12, + "max": 0.22224967428314468, + "min": 0.0002683613903166071, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 12, + "zero": 0 + }, + "effect_to_repeat_median": 1.3242319261149558, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.011966625936500275, + "repeat_abs_p95": 0.10502297931861322, + "repeat_delta": { + "distinct_n": 24, + "max": 0.05015486062543706, + "min": -0.14511939254670791, + "n": 24 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.08333333333333333, + "action_abs_median": 0.004416664987538066, + "action_delta": { + "distinct_n": 12, + "max": 0.06671902028938567, + "min": -0.00038309683389285525, + "n": 12 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 2, + "positive": 10, + "zero": 0 + }, + "effect_to_repeat_median": 0.8499259269415633, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.005196529306302399, + "repeat_abs_p95": 0.03531051450496546, + "repeat_delta": { + "distinct_n": 24, + "max": 0.028898615902990022, + "min": -0.05744492686343953, + "n": 24 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 12 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 12 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 24 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.001539018865967745, + "action_delta": { + "distinct_n": 12, + "max": 0.003050797976300035, + "min": -0.0015887887661509836, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.15591584638446954, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.009870830333516656, + "repeat_abs_p95": 0.021489430916148855, + "repeat_delta": { + "distinct_n": 24, + "max": 0.021795755533912664, + "min": -0.02021322592113628, + "n": 24 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.5940786813055561, + "action_delta": { + "distinct_n": 12, + "max": 33.446711043166665, + "min": -0.6256704651111118, + "n": 12 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 8, + "zero": 0 + }, + "effect_to_repeat_median": 0.30447578852990603, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.9511524518055556, + "repeat_abs_p95": 4.689088691175, + "repeat_delta": { + "distinct_n": 24, + "max": 3.8720077572777782, + "min": -11.837926215166668, + "n": 24 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.4166666666666667, + "action_abs_median": 0.9777240376944445, + "action_delta": { + "distinct_n": 11, + "max": 0.0, + "min": -28.609993227777778, + "n": 12 + }, + "action_signs": { + "consistency": 1.0, + "negative": 10, + "positive": 0, + "zero": 2 + }, + "effect_to_repeat_median": 8.318675069172114, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.11753362519444445, + "repeat_abs_p95": 2.0560520443777768, + "repeat_delta": { + "distinct_n": 20, + "max": 1.6669221998333335, + "min": -9.630657473833335, + "n": 24 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 5.527777777777779, + "action_delta": { + "distinct_n": 12, + "max": 9.166666666666671, + "min": -33.27777777777777, + "n": 12 + }, + "action_signs": { + "consistency": 0.5, + "negative": 6, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.3338926174496646, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 16.55555555555555, + "repeat_abs_p95": 37.27222222222221, + "repeat_delta": { + "distinct_n": 24, + "max": 31.888888888888886, + "min": -41.611111111111114, + "n": 24 + } + } + }, + "sanity": { + "action_pairs": 12, + "invariants": { + "efficacy_label_balance": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "matched_action_request_hashes": true, + "probabilities_bounded": true + }, + "red_flags": [], + "repeat_pairs": 24, + "trials": 36 + }, + "start_s": 0.0 + } + ], + "decision": "REQUIRES_UNCENSORED_PHASE_AWARE_PILOT", + "design": { + "available_deciles": [ + 0.1, + 0.2, + 0.3 + ], + "cumulative_and_nonoverlapping_blocks": true, + "decile_fraction": 0.1, + "maximum_common_end_s": 18.0, + "maximum_common_fraction": 0.3, + "select_best_horizon": false, + "trace_duration_s": 60.0 + }, + "provenance": { + "analysis_script": "/home/gahow/phd/aituner/runs/intervention-response-v2/analyze_existing.py", + "analysis_script_sha256": "c7eb2cd2d00f2b3afa62e22e7b2f1cce6b87ec51ebb7921ef7ffb9d13d28c50a", + "manifest": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/pilot-manifest.json", + "manifest_sha256": "86b249a087ab1bcd51cd8de3d1da6bf34afd41af977eb58c93229b941dbf5c8a", + "manifest_validation": { + "expected_trials": 36, + "matched_trials": 36, + "schema": "fidelity-prefix-pilot-manifest-v1" + }, + "p1_analysis_script": "/home/gahow/phd/aituner/runs/intervention-response-v0/analyze_p1.py", + "p1_analysis_script_sha256": "902adcba58bcb73f97c455fdea3710f6cb28fa8d8f4ea5bc4915143ed106b17a", + "run_root": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b", + "streams": [ + { + "bytes": 24501686, + "cell": "tp1_mns64", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns64/opprof/opprof-v1-dp0-pid234511-1784005631788937266.jsonl", + "sha256": "a813076afe677e5c13010b2ec44f46998794ca44cdf30ee81be836e3623f0068" + }, + { + "bytes": 25524027, + "cell": "tp1_mns8", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp1_mns8/opprof/opprof-v1-dp0-pid227734-1784005177936445360.jsonl", + "sha256": "1708759d2e77c42bff1940d43214fb2d927c3eebeda62d53a3fb7d04817e9b22" + }, + { + "bytes": 29431988, + "cell": "tp2_mns64", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns64/opprof/opprof-v1-dp0-pid247791-1784006495467408663.jsonl", + "sha256": "65355de883f2eb5d31be161d82b26dddc53c41a0a35a0fdb807777bc5b21c5ec" + }, + { + "bytes": 26036993, + "cell": "tp2_mns8", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp2_mns8/opprof/opprof-v1-dp0-pid241594-1784006096655191299.jsonl", + "sha256": "485e203d37d45fdd3c4de768b95da2050478f111f4d739b129d4f4ebd4737b70" + }, + { + "bytes": 17449143, + "cell": "tp4_mns16", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns16/opprof/opprof-v1-dp0-pid256090-1784007028237607411.jsonl", + "sha256": "59e98014d01bfc68d9f97112ecd782cb03571ad437382ed67416c7b85ae48c56" + }, + { + "bytes": 22674678, + "cell": "tp4_mns64", + "path": "/home/gahow/phd/replayserve/runs/fidelity_p1_frontier_committed_20260714/real/p1b/cells/tp4_mns64/opprof/opprof-v1-dp0-pid262293-1784007376383569515.jsonl", + "sha256": "03e4f3cca3023a78e859b663703b9468dedf89e58320f0292e2973070e7616c8" + } + ], + "trial_inputs": [ + { + "cell": "tp1_mns64", + "early_stopped": true, + "elapsed_s": 19.448401608, + "level": "high", + "mns": 64, + "replicate": 1, + "request_count": 179, + "requests_sha256": "e970dcac20144cb1e13a35997377f6695a9c57d788a88bad7f18cf6c64e35cec", + "result_sha256": "3233686b60b920dcd6cc2f45193a81f3af2d46f781ef42c4d117630612084feb", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/high-rep1/result.json" + }, + { + "cell": "tp1_mns64", + "early_stopped": true, + "elapsed_s": 56.362951594, + "level": "high", + "mns": 64, + "replicate": 2, + "request_count": 179, + "requests_sha256": "ec06878e2aefdc4853d972197cedc29ab94df7147abb2701e0d88b5dabf29728", + "result_sha256": "8215477a014f061c55e94d879a25bbf1676ee2ef1c5e81efadd23d20d337e07a", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/high-rep2/result.json" + }, + { + "cell": "tp1_mns64", + "early_stopped": true, + "elapsed_s": 59.320339642, + "level": "high", + "mns": 64, + "replicate": 3, + "request_count": 179, + "requests_sha256": "18d72d1fe16136359e28a1909e395ffd553dbe3edd2c0fd93a9cc270a5ec6d60", + "result_sha256": "480a3b50acaf9b9c1902e2669cf3aa7723a6703a00778da269b5b9757990ceef", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/high-rep3/result.json" + }, + { + "cell": "tp1_mns64", + "early_stopped": false, + "elapsed_s": 60.653813767, + "level": "low", + "mns": 64, + "replicate": 1, + "request_count": 122, + "requests_sha256": "44d729c1ace9b09fe2ea4c4e110bd0f95fee69f83edc78ee0f65256922dcd2c2", + "result_sha256": "d2c08e71c5d4e0dd99662bab8457bd1ad0a4aba6bda57794dbfce62e36720f62", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/low-rep1/result.json" + }, + { + "cell": "tp1_mns64", + "early_stopped": false, + "elapsed_s": 61.821918535, + "level": "low", + "mns": 64, + "replicate": 2, + "request_count": 122, + "requests_sha256": "97de3cced64fd88a5c2dcd087ca6c29d8b09a0a43d7bf0e4e844ef246462465b", + "result_sha256": "f194db2f7abb995171cd24626e4e515c000481f81283c929b9a5b63f110f1506", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/low-rep2/result.json" + }, + { + "cell": "tp1_mns64", + "early_stopped": false, + "elapsed_s": 59.372640014, + "level": "low", + "mns": 64, + "replicate": 3, + "request_count": 122, + "requests_sha256": "30f4623100c58580cfa4efe9a1998c63581c8e44511c556ef39cf5bd777883bf", + "result_sha256": "0be457aa2fc9ce95bdb92a6eb25544f5f61a9739932ee5afa40535d26456e348", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns64/low-rep3/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": true, + "elapsed_s": 24.256301395, + "level": "high", + "mns": 8, + "replicate": 1, + "request_count": 179, + "requests_sha256": "5f5bfa4a16846190f488d3369eed89ae895804145cea59197e6b544b703223fc", + "result_sha256": "f65e3dc91d17f5caab1f4aecf97d2c464b99f107231af2cfd96f5484880136ee", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/high-rep1/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": true, + "elapsed_s": 57.560632316, + "level": "high", + "mns": 8, + "replicate": 2, + "request_count": 179, + "requests_sha256": "e802d09057de349e5164c18314fcad09dc88c9673834fd6e5b100a57dd12fbf4", + "result_sha256": "4835a985dd6bb8dc50b53737baa980ac95a49e22f0b8a7326e1fb416c9b4e798", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/high-rep2/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": true, + "elapsed_s": 46.401123902, + "level": "high", + "mns": 8, + "replicate": 3, + "request_count": 179, + "requests_sha256": "cf1226185845cd8aa4e7ead0cb5c95c77eda62b84851a2ab579f984bbb88e8ac", + "result_sha256": "95d5103f2678eeb99c258870324566932e2f2853c3b6e0ee94bc6c538f6ed865", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/high-rep3/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": false, + "elapsed_s": 60.653530104, + "level": "low", + "mns": 8, + "replicate": 1, + "request_count": 122, + "requests_sha256": "0edc6095b16e5361a88b2765cc5d0f267ed16ff1d883e02fc3c58f57e59c1018", + "result_sha256": "8bc433109d148c21b7d158df95ec5d1fa4fa619243114563068e1e98c4241a29", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/low-rep1/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": false, + "elapsed_s": 62.156958699, + "level": "low", + "mns": 8, + "replicate": 2, + "request_count": 122, + "requests_sha256": "ca58a878401ac44cca0a9a595eab41659dafec2e923f7f2474e4f82ea30b2f3e", + "result_sha256": "42a56af136d1d3bdfc16d9cea092c2db0eaa4db2c85dceb21958c5e418122900", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/low-rep2/result.json" + }, + { + "cell": "tp1_mns8", + "early_stopped": false, + "elapsed_s": 59.499655201, + "level": "low", + "mns": 8, + "replicate": 3, + "request_count": 122, + "requests_sha256": "8a72240f84bbeff9eabd63e7d1f7a4c19910660fc5e4335013a1deaf89408127", + "result_sha256": "0214f32cabbcdc1bad6c331c3945c505cad136fcc9a8336aeb6314b27474e279", + "tp": 1, + "trace_duration_s": 60.0, + "trial_id": "cells/tp1_mns8/low-rep3/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 60.842664607, + "level": "high", + "mns": 64, + "replicate": 1, + "request_count": 345, + "requests_sha256": "19537ec05f1e40543d850964db12234561954c209d55cd8cd0a1da716ab105fc", + "result_sha256": "49523b3269db649aa70072058666da3309ecabb93242cd504b6413f5de0f5d54", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/high-rep1/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 61.33597184, + "level": "high", + "mns": 64, + "replicate": 2, + "request_count": 345, + "requests_sha256": "d95e0dcc3d3f25af11b4d3da0ca133f647fd602753b804337a7131cc56e03059", + "result_sha256": "f37a42d9ea734adcb24892f86cbb19e2c51cee0baf8a6a63c5d51c4820bd28f2", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/high-rep2/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 60.925907661, + "level": "high", + "mns": 64, + "replicate": 3, + "request_count": 345, + "requests_sha256": "ccba3d0a18a7db8501b040855bda473ee55c519619d4b9fa2a105bda06300829", + "result_sha256": "55447419925f30397d19863e149890241c58863d0adaafd511a67757905cb58e", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/high-rep3/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 60.73903317, + "level": "low", + "mns": 64, + "replicate": 1, + "request_count": 235, + "requests_sha256": "8685f475baf247f7ce52a6a2df1183cd77a6f580cd57a3a87c407b4e96c0895f", + "result_sha256": "7a8f4e462d496bd56b73b0a686653cbe499a238df62a8c9a50db687d051f1c68", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/low-rep1/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 61.049384964, + "level": "low", + "mns": 64, + "replicate": 2, + "request_count": 235, + "requests_sha256": "016a76ab3dc015700f9ecb9f60a20fcc3b456c45270472fdd1d04e84ebd7021a", + "result_sha256": "368869326f44a0032985eff9d3abfa2306b93c25370caa05cee255a8c0208c0e", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/low-rep2/result.json" + }, + { + "cell": "tp2_mns64", + "early_stopped": false, + "elapsed_s": 60.611333104, + "level": "low", + "mns": 64, + "replicate": 3, + "request_count": 235, + "requests_sha256": "5455d1cdeabd6e2937130ed8fc8d86c1aaa6f04cbf26c3cb69ad064be5f67d73", + "result_sha256": "2de5c4524f57b487b94b4fdba35defc2a566236fc3e31f7260b7a9f85bb7007c", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns64/low-rep3/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": true, + "elapsed_s": 25.452006333, + "level": "high", + "mns": 8, + "replicate": 1, + "request_count": 336, + "requests_sha256": "5d41a316f329e2efcc0f728cdd078377eb72fb9258738636db704bcfcca290aa", + "result_sha256": "466b3991190a232c984ba1b54fb6793cac1893dce196ee96f7238d70d49fcb74", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/high-rep1/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": true, + "elapsed_s": 24.770382696, + "level": "high", + "mns": 8, + "replicate": 2, + "request_count": 336, + "requests_sha256": "1b7d2cdcd639c00c443ae1b98287a7cc57323eb80c7f0e1bfc72f1ab5bf48760", + "result_sha256": "e9156d6871cc8cb3a20dce629cc2443df7473f3836b46eb0944d491ea5c4d44c", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/high-rep2/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": true, + "elapsed_s": 25.69871505, + "level": "high", + "mns": 8, + "replicate": 3, + "request_count": 336, + "requests_sha256": "5a68fda7c29477e5b3e66d70ad2b05b77be6c200cc1beea89d663a5729f31a2b", + "result_sha256": "f2116797b35057949acbf1f57898befac9ff18b94c808462f0969bdcf2446dd6", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/high-rep3/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": false, + "elapsed_s": 60.605998599, + "level": "low", + "mns": 8, + "replicate": 1, + "request_count": 229, + "requests_sha256": "455a1c21388700f0e3234c49d15ba3b0426b0d3b551c096b297793afa0d4602f", + "result_sha256": "13de25a73cc8c752245f169cfca1d64220f697c15b835a646ba969a1005047fb", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/low-rep1/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": false, + "elapsed_s": 61.925251312, + "level": "low", + "mns": 8, + "replicate": 2, + "request_count": 229, + "requests_sha256": "cae4b764ebb8714a6f5d2009bb79497420b388986a992c51410f1681810c14c6", + "result_sha256": "92f92a576e990946c4767a2107d2ce0e6ba1b066337911399360c199fae3f5e4", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/low-rep2/result.json" + }, + { + "cell": "tp2_mns8", + "early_stopped": false, + "elapsed_s": 60.612475588, + "level": "low", + "mns": 8, + "replicate": 3, + "request_count": 229, + "requests_sha256": "cc8c01e4560411cd5b84cdc9bebdda9a86b88615ee1c39a996affe669d26c167", + "result_sha256": "c9a7d026bd89cc027c45b8f1899d56e21d53d4234e95db365a9492f670a78d36", + "tp": 2, + "trace_duration_s": 60.0, + "trial_id": "cells/tp2_mns8/low-rep3/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 25.468414615, + "level": "high", + "mns": 16, + "replicate": 1, + "request_count": 750, + "requests_sha256": "8695422a0f1e0e1c097a08a7f4c082f1ba2e3bb26b58caadae0afc7438aeca90", + "result_sha256": "c54a2bb5c1b78bdef3fa09c39f5974215370d6fc7c2b6d403a3d376d53a1d9eb", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/high-rep1/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 24.508063881, + "level": "high", + "mns": 16, + "replicate": 2, + "request_count": 750, + "requests_sha256": "aa0482b10414d911af227e2df172c19ecca0add2e0024dc88f85ffbd063854bb", + "result_sha256": "91e2491de4ca0868b667e7d126b755f03bdd78378681b8070a2bbc0c38fb9055", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/high-rep2/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 27.456202024, + "level": "high", + "mns": 16, + "replicate": 3, + "request_count": 750, + "requests_sha256": "3a23ffbe0e724648226152ef5433cc5825ad640c37cf47997c2f091210eceeec", + "result_sha256": "136dde0451f806841ff3996b081fe0bffac0a76f60c3d78b7516f2689e106008", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/high-rep3/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 48.952560958, + "level": "low", + "mns": 16, + "replicate": 1, + "request_count": 510, + "requests_sha256": "d7eaaefd35cc906f759dcaa7ac3c8543249a1a02f28572bec953e41e4379a29f", + "result_sha256": "11d9ec9d2a3358aed988f7b4b1014af087d61c4b5a72eb9076ae19bb5cc481cd", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/low-rep1/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 49.912113702, + "level": "low", + "mns": 16, + "replicate": 2, + "request_count": 510, + "requests_sha256": "7fdb90f679bba1747fbacc92fac2045c820e09965c755b20f6dafd4fd9e26a7a", + "result_sha256": "800e2aa6e9f9efef4be12ee778be6817364914d63ae0624775d02729089768d6", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/low-rep2/result.json" + }, + { + "cell": "tp4_mns16", + "early_stopped": true, + "elapsed_s": 48.923086251, + "level": "low", + "mns": 16, + "replicate": 3, + "request_count": 510, + "requests_sha256": "b592fe254476957265fc9139aefdea77be4b5540f3c1b8a429c10ea6ccab9a6a", + "result_sha256": "cc1b1f7f5280c2892d0c6deac9df9f34ce126b0284e6ad893ffbf0ddbf27969f", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns16/low-rep3/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 61.434823743, + "level": "high", + "mns": 64, + "replicate": 1, + "request_count": 750, + "requests_sha256": "b4aeaf5b3175bd85fdac08a2ef657fb2278c6c4b36ea32ea417275ab6d55812d", + "result_sha256": "111a1042b614d05772888665a7cfbc86925b2fbeb52ffe19edb62a5bcfe2f513", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/high-rep1/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 61.130504059, + "level": "high", + "mns": 64, + "replicate": 2, + "request_count": 750, + "requests_sha256": "5d82b6885f5c654ae1ae126696df4bafe04f5be1c1109db24c7856b0456d9443", + "result_sha256": "6ed574c18723e265ab1327c5e43d3c76d37b3fc5711d06c77a51718e4ba16601", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/high-rep2/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 61.22613867, + "level": "high", + "mns": 64, + "replicate": 3, + "request_count": 750, + "requests_sha256": "5c5861d716181f06dc3f951ae0d6ee479b1b61fc21ebf24541c5b45053fcc167", + "result_sha256": "35b8f0465f6b0c197f71b5ecaf4e771e6613f86998ae44f8701ab9e4e1e440b4", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/high-rep3/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 61.123849552, + "level": "low", + "mns": 64, + "replicate": 1, + "request_count": 510, + "requests_sha256": "931d372ded1e3aaa9c67f06653fac8f79b4ee4f369028c1148163580386681ce", + "result_sha256": "325ce57f33133670a6761b5dffb831ed3b2f464755ff743b56976b4a974c1215", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/low-rep1/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 60.86041954, + "level": "low", + "mns": 64, + "replicate": 2, + "request_count": 510, + "requests_sha256": "72c8f70a240f79edcaa4eb0a2fe9bade8ac07093339c5b2d09c49ddd1cf27d16", + "result_sha256": "ff8b2283d02d5967893ca6225b561ae11011da6149d1f6cff7b4275c00170e1d", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/low-rep2/result.json" + }, + { + "cell": "tp4_mns64", + "early_stopped": false, + "elapsed_s": 60.47213624, + "level": "low", + "mns": 64, + "replicate": 3, + "request_count": 510, + "requests_sha256": "8b2c779d2097c5e4038e94e1ac9b986e29f55d9acfd67eeb366c4e0e9270f6e0", + "result_sha256": "f026b69356701894fab184e0f2f7504a30b537982e823373899c9ab123a33891", + "tp": 4, + "trace_duration_s": 60.0, + "trial_id": "cells/tp4_mns64/low-rep3/result.json" + } + ] + }, + "sanity": { + "early_stopped": 15, + "elapsed_s": { + "distinct_n": 36, + "max": 62.156958699, + "median": 60.5390674195, + "min": 19.448401608, + "n": 36 + }, + "invariants": { + "all_intervals_uncensored": true, + "all_window_sanity_pass": true, + "expected_trial_count": true, + "manifest_trials_match": true, + "stream_provenance_consistent": true, + "trace_duration_consistent": true + }, + "red_flags": [], + "request_count": { + "distinct_n": 8, + "max": 750.0, + "median": 285.5, + "min": 122.0, + "n": 36 + }, + "stream_bytes": { + "distinct_n": 6, + "max": 29431988.0, + "median": 25012856.5, + "min": 17449143.0, + "n": 6 + }, + "trace_duration_s": { + "distinct_n": 1, + "max": 60.0, + "median": 60.0, + "min": 60.0, + "n": 36 + }, + "trials": 36 + }, + "schema": "intervention-response-phase-aware-existing-v2", + "status": "COMPLETE", + "trajectory": { + "blocks": [ + { + "end_s": 6.0, + "start_s": 0.0 + }, + { + "end_s": 12.0, + "start_s": 6.0 + }, + { + "end_s": 18.0, + "start_s": 12.0 + } + ], + "features": { + "batch_size.mean": { + "block_medians": [ + 6.414992146729492, + 7.937718404518684, + 7.938679599969923 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 44.717667092379735, + "median": 2.4047228794539084, + "min": 0.12661717921526972, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 44.717667092379735, + "median": 1.715194194802789, + "min": -6.818300947246247, + "n": 36 + }, + "first_to_last_pearson": 0.7466021228318502 + }, + "batch_tokens.mean": { + "block_medians": [ + 182.10234474017744, + 172.7298136645963, + 202.28603044732077 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 908.0537249893571, + "median": 75.02808983931541, + "min": 7.692264317468613, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 908.0537249893571, + "median": 21.47667280168706, + "min": -316.4167321876521, + "n": 36 + }, + "first_to_last_pearson": 0.7254256632404311 + }, + "decode_batch_size.mean": { + "block_medians": [ + 6.354735422986181, + 7.873722548380083, + 7.875311697892343 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 44.34555129842486, + "median": 2.4157231102242664, + "min": 0.15471898197242862, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 44.34555129842486, + "median": 1.7131997250467412, + "min": -6.730278183383879, + "n": 36 + }, + "first_to_last_pearson": 0.7458594110784589 + }, + "graph_full_share": { + "block_medians": [ + 0.9440933791043857, + 0.9375353988798221, + 0.9361517143943923 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.20131971051511277, + "median": 0.027226282812290892, + "min": 0.00023443249531129862, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.08227563742558686, + "median": -0.0017697738257967388, + "min": -0.20131971051511277, + "n": 36 + }, + "first_to_last_pearson": 0.6826036350375567 + }, + "graph_none_share": { + "block_medians": [ + 0.05080178541907753, + 0.05845283871165627, + 0.06118872118872119 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.19429544487015749, + "median": 0.01965769475771631, + "min": 0.0028422907600285444, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.19429544487015749, + "median": 0.0040749575702726006, + "min": -0.0846718334643753, + "n": 36 + }, + "first_to_last_pearson": 0.65716602522967 + }, + "graph_padding_fraction": { + "block_medians": [ + 0.005497687293445758, + 0.0075248839288535735, + 0.006051201250183472 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.014587933688096199, + "median": 0.002744262881772697, + "min": 4.7975018144524526e-05, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.014587933688096199, + "median": -0.0005530753213689365, + "min": -0.008320005718079814, + "n": 36 + }, + "first_to_last_pearson": 0.3098612628425477 + }, + "kv_usage_end_minus_start": { + "block_medians": [ + 0.023398338975546773, + -0.001777370408002521, + 0.0038338466257342163 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.262813467878909, + "median": 0.017891096563429187, + "min": 0.0015150201237521532, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.07218503346987704, + "median": -0.01752170592552832, + "min": -0.262813467878909, + "n": 36 + }, + "first_to_last_pearson": -0.012340699297270389 + }, + "kv_usage_max": { + "block_medians": [ + 0.03336296364308622, + 0.04022944403944284, + 0.030052448221737305 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.29553401938255563, + "median": 0.011325832339947661, + "min": 0.0002397506593143639, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.29553401938255563, + "median": 0.00083657676867116, + "min": -0.09986012588670201, + "n": 36 + }, + "first_to_last_pearson": 0.4134682913789121 + }, + "kv_usage_mean": { + "block_medians": [ + 0.016731908775651918, + 0.02129250131291721, + 0.01914443130373699 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.271100737119201, + "median": 0.009396606492664827, + "min": 0.0007061033361193698, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.271100737119201, + "median": 0.002361521286958815, + "min": -0.06598177225454588, + "n": 36 + }, + "first_to_last_pearson": 0.10554346287202872 + }, + "preemptions": { + "block_medians": [ + 0.0, + 0.0, + 0.0 + ], + "changed_trials": 0, + "first_to_last_abs_delta": { + "distinct_n": 1, + "max": 0.0, + "median": 0.0, + "min": 0.0, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 1, + "max": 0.0, + "median": 0.0, + "min": 0.0, + "n": 36 + }, + "first_to_last_pearson": null + }, + "prefill_token_fraction": { + "block_medians": [ + 0.9617744444596128, + 0.955926466887596, + 0.9488462859261657 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 0.0574568246654229, + "median": 0.013436705979377572, + "min": 0.00024055263508171443, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 0.008129115874074722, + "median": -0.013436705979377572, + "min": -0.0574568246654229, + "n": 36 + }, + "first_to_last_pearson": 0.23303776424770067 + }, + "queue_running_mean": { + "block_medians": [ + 6.85322045575, + 8.0, + 8.0 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 44.04634992016667, + "median": 2.4438409930833327, + "min": 0.5856487165000002, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 44.04634992016667, + "median": 1.8893361798333332, + "min": -4.201522480666667, + "n": 36 + }, + "first_to_last_pearson": 0.8338308544566243 + }, + "queue_waiting_mean": { + "block_medians": [ + 0.0, + 0.020209624833333332, + 0.008993971666666666 + ], + "changed_trials": 23, + "first_to_last_abs_delta": { + "distinct_n": 24, + "max": 39.884282579499995, + "median": 0.008993971666666666, + "min": 0.0, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 24, + "max": 39.884282579499995, + "median": 0.00026988233333333, + "min": -0.7860144426666666, + "n": 36 + }, + "first_to_last_pearson": 0.8749679224022264 + }, + "scheduler_steps_per_s": { + "block_medians": [ + 75.83333333333334, + 78.16666666666667, + 73.58333333333334 + ], + "changed_trials": 36, + "first_to_last_abs_delta": { + "distinct_n": 36, + "max": 76.5, + "median": 28.250000000000004, + "min": 2.1666666666666714, + "n": 36 + }, + "first_to_last_delta": { + "distinct_n": 36, + "max": 73.33333333333333, + "median": -3.1666666666666714, + "min": -76.5, + "n": 36 + }, + "first_to_last_pearson": 0.40161538799446195 + } + }, + "trial_count": 36 + } +} diff --git a/runs/intervention-response-v2/failed-three-load-controller-state-v2.json b/runs/intervention-response-v2/failed-three-load-controller-state-v2.json new file mode 100644 index 0000000..94a6dd4 --- /dev/null +++ b/runs/intervention-response-v2/failed-three-load-controller-state-v2.json @@ -0,0 +1,49 @@ +{ + "completed_sessions": 0, + "failures": [ + { + "failure": "TimeoutError('client timeout: tp4_mns16 high')", + "session": "rep1-mns16" + } + ], + "gpu_hours_total": 1.3580546813540988, + "hard_cap_h20_hours": 8.0, + "schema": "intervention-response-phase-aware-pilot-state-v2", + "sessions": { + "rep1-mns16": { + "burnin": { + "elapsed_s": 64.469448031, + "feasible": false, + "pass_rate": 0.5137254901960784 + }, + "failure": "TimeoutError('client timeout: tp4_mns16 high')", + "gpu_hours": 1.3580546813540988, + "mns": 16, + "replicate": 1, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 301.114789132, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + }, + { + "early_stopped": false, + "elapsed_s": 313.040655133, + "feasible": false, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 0.5631372549019608, + "selected_count": 2550 + } + ], + "started_at": 1784021277.1191542, + "status": "failed" + } + }, + "started_at": 1784021276.8224819, + "status": "failed" +} diff --git a/runs/intervention-response-v2/pilot-audit-v3.json b/runs/intervention-response-v2/pilot-audit-v3.json new file mode 100644 index 0000000..8a775e7 --- /dev/null +++ b/runs/intervention-response-v2/pilot-audit-v3.json @@ -0,0 +1,16451 @@ +{ + "claim_boundary": "Development mechanism pilot; not a held-out paper claim.", + "cumulative": [ + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.01629688095037396, + "tpot_mean_over_slo": -0.00616777464948251, + "ttft_max_over_slo_max": -0.04632652748841792, + "ttft_mean_over_slo_max": -0.002870656998408553 + }, + "delta_state": { + "batch_size.mean": -0.07015726376607656, + "batch_tokens.mean": -1.516200383603163, + "decode_batch_size.mean": -0.06904842848992132, + "graph_full_share": 0.0002648406614842447, + "graph_none_share": -0.002304071470614101, + "graph_padding_fraction": 0.0016774700470183774, + "kv_usage_end_minus_start": 1.8935114667439912e-05, + "kv_usage_max": 0.000835313152050321, + "kv_usage_mean": -0.0002023458881830597, + "preemptions": 0.0, + "prefill_token_fraction": -6.6475857588566e-06, + "queue_running_mean": -0.2196172835333332, + "queue_waiting_mean": -0.06376780166666667, + "scheduler_steps_per_s": 1.6666666666666714 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.000784313725490196, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.009216589861751112, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.29719366945920317, + "tpot_mean_over_slo": 0.03216772663644496, + "ttft_max_over_slo_max": -0.26695654933185625, + "ttft_mean_over_slo_max": -0.0567957568681583 + }, + "delta_state": { + "batch_size.mean": -1.148782542917619, + "batch_tokens.mean": -26.133196704277452, + "decode_batch_size.mean": -1.1386565766713446, + "graph_full_share": 0.012748972070391162, + "graph_none_share": -0.017569378163986613, + "graph_padding_fraction": 0.0029836610976256795, + "kv_usage_end_minus_start": 6.887268460797902e-05, + "kv_usage_max": 0.023285407613873432, + "kv_usage_mean": 0.0014582718656800941, + "preemptions": 0.0, + "prefill_token_fraction": -0.00016554148424863246, + "queue_running_mean": 1.499029866199999, + "queue_waiting_mean": -2.5166945028333334, + "scheduler_steps_per_s": 9.266666666666666 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.051139209137158914, + "tpot_mean_over_slo": -0.01140034356948863, + "ttft_max_over_slo_max": -0.016116498833677426, + "ttft_mean_over_slo_max": -0.0032859918483368646 + }, + "delta_state": { + "batch_size.mean": -0.38049431384111454, + "batch_tokens.mean": -9.40007307911111, + "decode_batch_size.mean": -0.37729230010560766, + "graph_full_share": 0.003814274093207559, + "graph_none_share": -0.008561445619684624, + "graph_padding_fraction": 0.0018195892219365715, + "kv_usage_end_minus_start": 3.2185675094664745e-05, + "kv_usage_max": 0.0022799714777140823, + "kv_usage_mean": -0.00029136261801412065, + "preemptions": 0.0, + "prefill_token_fraction": 1.739291503155549e-06, + "queue_running_mean": -0.4335179698999996, + "queue_waiting_mean": -0.0903444594, + "scheduler_steps_per_s": 5.733333333333334 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.007450980392156863, + "completed_over_admitted": 0.09411764705882353, + "completed_pass_rate": 0.0875576036866359, + "outstanding_over_admitted": -0.09411764705882353, + "tpot_max_over_slo": 0.23737520567053472, + "tpot_mean_over_slo": 0.05593989653787507, + "ttft_max_over_slo_max": -0.35981755234145874, + "ttft_mean_over_slo_max": -0.14934518008110995 + }, + "delta_state": { + "batch_size.mean": 1.556060197394805, + "batch_tokens.mean": 36.05112475298853, + "decode_batch_size.mean": 1.538767142928517, + "graph_full_share": -0.0035723945897613296, + "graph_none_share": -0.005640562586105874, + "graph_padding_fraction": 0.005841856698433652, + "kv_usage_end_minus_start": 0.0004291753466407755, + "kv_usage_max": 0.022672590909591506, + "kv_usage_mean": 0.004255843424978539, + "preemptions": 0.0, + "prefill_token_fraction": 0.00010886369563634091, + "queue_running_mean": 4.158758860866666, + "queue_waiting_mean": -8.736845286133335, + "scheduler_steps_per_s": -2.8666666666666742 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.006756756756756799, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.006756756756756757, + "tpot_max_over_slo": -0.041394997635621655, + "tpot_mean_over_slo": -0.01348315882501705, + "ttft_max_over_slo_max": -0.017940445327743276, + "ttft_mean_over_slo_max": -0.0005297892694696611 + }, + "delta_state": { + "batch_size.mean": -0.31669151594137457, + "batch_tokens.mean": -6.478151733400836, + "decode_batch_size.mean": -0.31462821691596865, + "graph_full_share": 0.0030818985907783425, + "graph_none_share": -0.003947353186013504, + "graph_padding_fraction": 0.0005071155621097952, + "kv_usage_end_minus_start": 1.1754955765441721e-05, + "kv_usage_max": -0.0005344961965740858, + "kv_usage_mean": -0.00033430782337439107, + "preemptions": 0.0, + "prefill_token_fraction": -1.9980466968050337e-05, + "queue_running_mean": -0.4079099217666666, + "queue_waiting_mean": -0.0088438692, + "scheduler_steps_per_s": 8.866666666666674 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.00462962962962965, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.004629629629629629, + "tpot_max_over_slo": 0.08779462519483128, + "tpot_mean_over_slo": -0.008887102176446698, + "ttft_max_over_slo_max": -0.1334900961664971, + "ttft_mean_over_slo_max": -0.024482643441387678 + }, + "delta_state": { + "batch_size.mean": -0.9935947147364157, + "batch_tokens.mean": -21.599624508861467, + "decode_batch_size.mean": -0.9860765047065438, + "graph_full_share": 0.01304670412975728, + "graph_none_share": -0.016958392743085346, + "graph_padding_fraction": 0.0017066418645361164, + "kv_usage_end_minus_start": -0.0004999534145386786, + "kv_usage_max": 0.01138222251747023, + "kv_usage_mean": -0.0002145679209778744, + "preemptions": 0.0, + "prefill_token_fraction": 4.875021747730024e-06, + "queue_running_mean": -0.33937291983333395, + "queue_waiting_mean": -1.0697946031666667, + "scheduler_steps_per_s": 9.5 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.10055555555555555, + "min": 0.08222222222222222, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 9, + "max": 0.09450980392156863, + "min": 0.08055555555555556, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "ttft_mean_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.0037254901960784314, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.000392156862745098, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.000392156862745098, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.002314814814814825, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": 0.0056931931931932245, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.047058823529411764, + "train_balanced_accuracy": 0.75 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.04377880184331795, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004608294930875556, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.004608294930875556, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.0023148148148148147, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": -0.005693193193193193, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.047058823529411764, + "train_balanced_accuracy": 0.75 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0694669171659951, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.05204575307260262, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.14425720740384682, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.010143722872967664, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.011185130500731874, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.012999975993481225, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.07571527074712019, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.08990831182745751, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.1566415384101371, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.013884317644862271, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.013676650219898116, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.03004087435824758, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.6120694630062742, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.6503523608112562, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.7348593572192978, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0011068787133229558, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0016920559557772469, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0024016251597811255, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.00025296526949599753, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00027443787217613274, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0006279629887485172, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.3071566254427864e-06, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -8.862820055632881e-07, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 5.530149356974823e-05, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.3736414208000003, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.3736414208000003, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.6397062913333329, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5800695312833334, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5667812024166667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -1.3035194811166666, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 1.43333333333333, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 9.06666666666667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.6000000000000014, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.1, + "end_s": 30.0, + "qualifying_response_features": [], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.6870445142887651, + "action_delta": { + "distinct_n": 6, + "max": 1.556060197394805, + "min": -1.148782542917619, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.3035211077938369, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 2.263580675764507, + "repeat_abs_p95": 4.395096149234066, + "repeat_delta": { + "distinct_n": 8, + "max": 3.5883262763875994, + "min": -4.829510696151395, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 15.499848793986288, + "action_delta": { + "distinct_n": 6, + "max": 36.05112475298853, + "min": -26.133196704277452, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.22257219949412319, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 69.63964425573081, + "repeat_abs_p95": 109.27564134554108, + "repeat_delta": { + "distinct_n": 8, + "max": 90.40754990562004, + "min": -119.43538289011397, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.6816844024060758, + "action_delta": { + "distinct_n": 6, + "max": 1.538767142928517, + "min": -1.1386565766713446, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.3037217743723986, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.2444370470793134, + "repeat_abs_p95": 4.356596129476966, + "repeat_delta": { + "distinct_n": 8, + "max": 3.5576095550798925, + "min": -4.7868196695369285, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0036933343414844444, + "action_delta": { + "distinct_n": 6, + "max": 0.01304670412975728, + "min": -0.0035723945897613296, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.20058285425846942, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.01841301119748373, + "repeat_abs_p95": 0.03154665462224765, + "repeat_delta": { + "distinct_n": 8, + "max": 0.035311983128397895, + "min": -0.024553901682254353, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.007101004102895249, + "action_delta": { + "distinct_n": 6, + "max": -0.002304071470614101, + "min": -0.017569378163986613, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.43631188200171145, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.01627506468610771, + "repeat_abs_p95": 0.023219157547472946, + "repeat_delta": { + "distinct_n": 8, + "max": 0.01731224262368622, + "min": -0.024172434833284892, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.001763115543236344, + "action_delta": { + "distinct_n": 6, + "max": 0.005841856698433652, + "min": 0.0005071155621097952, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.3596998263590112, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0012966946888252496, + "repeat_abs_p95": 0.004289292768312434, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005183048738787855, + "min": -0.0026294602517152265, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 5.052917985132188e-05, + "action_delta": { + "distinct_n": 6, + "max": 0.0004291753466407755, + "min": -0.0004999534145386786, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.005664167631863443, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.008920848240273282, + "repeat_abs_p95": 0.01385760527350104, + "repeat_delta": { + "distinct_n": 8, + "max": 0.007292667470716885, + "min": -0.013864756025266267, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.0068310969975921565, + "action_delta": { + "distinct_n": 6, + "max": 0.023285407613873432, + "min": -0.0005344961965740858, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 2.5270151274703703, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0027032275839323217, + "repeat_abs_p95": 0.00993555673664627, + "repeat_delta": { + "distinct_n": 8, + "max": 0.006112144565506239, + "min": -0.011994317136490906, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00031283522069425586, + "action_delta": { + "distinct_n": 6, + "max": 0.004255843424978539, + "min": -0.00033430782337439107, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 2, + "zero": 0 + }, + "effect_to_repeat_median": 0.09690654635110718, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0032282155589448643, + "repeat_abs_p95": 0.007211323545202779, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005822121070260689, + "min": -0.007959355647094676, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 1.3314026363453468e-05, + "action_delta": { + "distinct_n": 6, + "max": 0.00010886369563634091, + "min": -0.00016554148424863246, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.003330724358187011, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.003997336594584067, + "repeat_abs_p95": 0.008763072555415058, + "repeat_delta": { + "distinct_n": 8, + "max": 0.00594070169016081, + "min": -0.00877067447087998, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4207139458333331, + "action_delta": { + "distinct_n": 6, + "max": 4.158758860866666, + "min": -0.4335179698999996, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 2, + "zero": 0 + }, + "effect_to_repeat_median": 0.15311155459778686, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 2.7477609181, + "repeat_abs_p95": 6.218551537058331, + "repeat_delta": { + "distinct_n": 8, + "max": 4.289906706066667, + "min": -7.2570525999, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.5800695312833334, + "action_delta": { + "distinct_n": 6, + "max": -0.0088438692, + "min": -8.736845286133335, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 41.55250132555395, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.013959918483333333, + "repeat_abs_p95": 7.160673893926665, + "repeat_delta": { + "distinct_n": 8, + "max": 6.219170523933334, + "min": -7.667637247, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 7.300000000000004, + "action_delta": { + "distinct_n": 6, + "max": 9.5, + "min": -2.8666666666666742, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.3724489795918368, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 19.60000000000001, + "repeat_abs_p95": 26.37000000000001, + "repeat_delta": { + "distinct_n": 8, + "max": 27.466666666666683, + "min": -21.63333333333334, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 0.08944444444444444, + "cell": "tp4_mns16", + "completed_fraction": 0.08499999999999999, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.060456432, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.09058823529411765, + "cell": "tp4_mns16", + "completed_fraction": 0.08509803921568629, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.057208547, + "max_internal_gap_s": 0.183419024, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.08944444444444444, + "cell": "tp4_mns64", + "completed_fraction": 0.08499999999999999, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.058728784, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.09058823529411765, + "cell": "tp4_mns64", + "completed_fraction": 0.08509803921568629, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.05828864, + "max_internal_gap_s": 0.184687818, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.10055555555555555, + "cell": "tp4_mns64", + "completed_fraction": 0.09444444444444444, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.003924877, + "max_internal_gap_s": 0.185686919, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.1, + "cell": "tp4_mns64", + "completed_fraction": 0.09450980392156863, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.00433309, + "max_internal_gap_s": 0.17460317, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.10055555555555555, + "cell": "tp4_mns16", + "completed_fraction": 0.09444444444444444, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.005845828, + "max_internal_gap_s": 0.174851445, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.1, + "cell": "tp4_mns16", + "completed_fraction": 0.08509803921568627, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.00492273, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.08222222222222222, + "cell": "tp4_mns16", + "completed_fraction": 0.08055555555555556, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.00040958, + "max_internal_gap_s": 0.611972913, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.08470588235294117, + "cell": "tp4_mns16", + "completed_fraction": 0.08156862745098038, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.008592672, + "max_internal_gap_s": 0.169483992, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.08222222222222222, + "cell": "tp4_mns64", + "completed_fraction": 0.08111111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.000396531, + "max_internal_gap_s": 0.612067602, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.08470588235294117, + "cell": "tp4_mns64", + "completed_fraction": 0.08196078431372548, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.009495956, + "max_internal_gap_s": 0.18072421, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.07082632707989239, + "tpot_mean_over_slo": -0.00490180963817316, + "ttft_max_over_slo_max": -0.04153848900265682, + "ttft_mean_over_slo_max": -0.006891098180069163 + }, + "delta_state": { + "batch_size.mean": -0.2638104929979219, + "batch_tokens.mean": -5.845648825359461, + "decode_batch_size.mean": -0.26164909366742517, + "graph_full_share": 0.002180359902912321, + "graph_none_share": -0.006323059607193614, + "graph_padding_fraction": 0.0018335461795120685, + "kv_usage_end_minus_start": 1.026766711520466e-05, + "kv_usage_max": 0.004459034805309736, + "kv_usage_mean": -0.00014393587383104527, + "preemptions": 0.0, + "prefill_token_fraction": -2.725586323304441e-05, + "queue_running_mean": -0.18631020918666685, + "queue_waiting_mean": -0.20975605228, + "scheduler_steps_per_s": 4.9466666666666725 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.00196078431372549, + "completed_over_admitted": 0.01983471074380172, + "completed_pass_rate": 0.008665511265164683, + "outstanding_over_admitted": -0.01983471074380165, + "tpot_max_over_slo": 0.2753294892902449, + "tpot_mean_over_slo": 0.04011748202807386, + "ttft_max_over_slo_max": -0.2852395010025551, + "ttft_mean_over_slo_max": -0.06592333248276636 + }, + "delta_state": { + "batch_size.mean": -0.8231037548470876, + "batch_tokens.mean": -19.74330932592642, + "decode_batch_size.mean": -0.8152070406366647, + "graph_full_share": 0.00839719842479858, + "graph_none_share": -0.018146448159773024, + "graph_padding_fraction": 0.005340401585168801, + "kv_usage_end_minus_start": -0.004438767356134465, + "kv_usage_max": 0.027975123151637415, + "kv_usage_mean": 0.002385274257794787, + "preemptions": 0.0, + "prefill_token_fraction": -0.00025035500653580733, + "queue_running_mean": 2.133982110813333, + "queue_waiting_mean": -3.33810440732, + "scheduler_steps_per_s": 6.320000000000007 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.004796163069544446, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.004796163069544366, + "tpot_max_over_slo": 0.051139209137158914, + "tpot_mean_over_slo": -0.0074211636478002385, + "ttft_max_over_slo_max": -0.03542824399967988, + "ttft_mean_over_slo_max": -0.0033018737683347885 + }, + "delta_state": { + "batch_size.mean": -0.2685691864197386, + "batch_tokens.mean": -6.959429182398708, + "decode_batch_size.mean": -0.26598172047692614, + "graph_full_share": 0.0028314394428595735, + "graph_none_share": -0.007209322700149463, + "graph_padding_fraction": 0.0012877200176958283, + "kv_usage_end_minus_start": -0.0011333814702583211, + "kv_usage_max": 0.0022799714777140823, + "kv_usage_mean": -0.0001951815494633533, + "preemptions": 0.0, + "prefill_token_fraction": -0.00014483809619469934, + "queue_running_mean": -0.2747586459066662, + "queue_waiting_mean": -0.08017956545333334, + "scheduler_steps_per_s": 5.586666666666659 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.02980392156862745, + "completed_over_admitted": 0.018394648829431426, + "completed_pass_rate": 0.13356766256590513, + "outstanding_over_admitted": -0.01839464882943144, + "tpot_max_over_slo": 0.23737520567053472, + "tpot_mean_over_slo": 0.010490282370595683, + "ttft_max_over_slo_max": -0.38453158099825185, + "ttft_mean_over_slo_max": -0.19098879755079737 + }, + "delta_state": { + "batch_size.mean": -1.413753104794269, + "batch_tokens.mean": -30.311859850911844, + "decode_batch_size.mean": -1.4061952013981784, + "graph_full_share": 0.021360307900858633, + "graph_none_share": -0.03011422871811248, + "graph_padding_fraction": 0.007512227776197788, + "kv_usage_end_minus_start": 0.0021360977816522464, + "kv_usage_max": 0.022672590909591506, + "kv_usage_mean": 0.0010254846876152635, + "preemptions": 0.0, + "prefill_token_fraction": 0.0003313271687591657, + "queue_running_mean": 0.7838954335199997, + "queue_waiting_mean": -9.07047515872, + "scheduler_steps_per_s": 8.959999999999994 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0024154589371980784, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0024154589371980714, + "tpot_max_over_slo": -0.006984168036628269, + "tpot_mean_over_slo": -0.01111313281352852, + "ttft_max_over_slo_max": -0.017940445327743276, + "ttft_mean_over_slo_max": -0.0018007220652844075 + }, + "delta_state": { + "batch_size.mean": -0.3420664768336481, + "batch_tokens.mean": -6.950334498514508, + "decode_batch_size.mean": -0.3395672760130166, + "graph_full_share": 0.0029764291623558536, + "graph_none_share": -0.005728618501492012, + "graph_padding_fraction": 0.0005459147599019987, + "kv_usage_end_minus_start": -0.003587734700701506, + "kv_usage_max": 3.276805657914661e-05, + "kv_usage_mean": -0.0002979372353175526, + "preemptions": 0.0, + "prefill_token_fraction": -6.130606240972725e-05, + "queue_running_mean": -0.4039724217999998, + "queue_waiting_mean": -0.030365675946666667, + "scheduler_steps_per_s": 7.826666666666668 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.013628620102214661, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.013628620102214647, + "tpot_max_over_slo": 0.08779462519483128, + "tpot_mean_over_slo": -0.005083324937856265, + "ttft_max_over_slo_max": -0.17233173300822574, + "ttft_mean_over_slo_max": -0.03508650576710582 + }, + "delta_state": { + "batch_size.mean": -0.5655282844513536, + "batch_tokens.mean": -11.412062377336781, + "decode_batch_size.mean": -0.562525695471205, + "graph_full_share": 0.009956297228293054, + "graph_none_share": -0.01796699078506772, + "graph_padding_fraction": 0.0038365149210898572, + "kv_usage_end_minus_start": 0.005666174682015579, + "kv_usage_max": 0.01138222251747023, + "kv_usage_mean": 0.0003053257977473668, + "preemptions": 0.0, + "prefill_token_fraction": -1.837494489209668e-05, + "queue_running_mean": 0.1581681952266667, + "queue_waiting_mean": -2.0748791364133337, + "scheduler_steps_per_s": 7.079999999999998 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.23777777777777778, + "min": 0.23, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 11, + "max": 0.23555555555555557, + "min": 0.2152941176470588, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.014901960784313726, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.000980392156862745, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.000980392156862745, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.009212391585879554, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.00802203951970637, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.011595405949487936, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.06678383128295257, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004332755632582341, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.004332755632582341, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.009212391585879507, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00802203951970636, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.011595405949487903, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0694669171659951, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.07931047613736183, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.15410076637521355, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.006252244292828252, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.008098228875692393, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0027942363662112613, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.1038799885039528, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.10693511100544129, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.16338899500260595, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.019194189767720304, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.02098880197358749, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.03640721533141776, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.4510464857421108, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.4510464857421108, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5405943805567954, + "train_balanced_accuracy": 1.0 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0025621174693928428, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.002835030550300963, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003586973882340435, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 5.5072124142006764e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 8.069496195816078e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0004407744068921091, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": -3.984050365091196e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.2815404062570543e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.00015203565276306064, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.05829522533999976, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.014071006980000078, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.2987926121666664, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.0775293509333335, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.1423175943466668, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.7739302298, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 6.333333333333329, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 5.63333333333334, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 5.953333333333333, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.25, + "end_s": 75.0, + "qualifying_response_features": [], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.45379738064250086, + "action_delta": { + "distinct_n": 6, + "max": -0.2638104929979219, + "min": -1.413753104794269, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.9705757741391393, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.467554819246339, + "repeat_abs_p95": 2.634518964255186, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7001454483303338, + "min": -2.9313976513752067, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.185745779867744, + "action_delta": { + "distinct_n": 6, + "max": -5.845648825359461, + "min": -30.311859850911844, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.298783186954296, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 30.743850996116663, + "repeat_abs_p95": 93.908100721683, + "repeat_delta": { + "distinct_n": 8, + "max": 27.153830102554593, + "min": -100.52302983743428, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4510464857421108, + "action_delta": { + "distinct_n": 6, + "max": -0.26164909366742517, + "min": -1.4061952013981784, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.9732742893789799, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.4634320362350386, + "repeat_abs_p95": 2.613462716433853, + "repeat_delta": { + "distinct_n": 8, + "max": 0.6993897150480155, + "min": -2.908747043508294, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.005686813793577217, + "action_delta": { + "distinct_n": 6, + "max": 0.021360307900858633, + "min": 0.002180359902912321, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.4307684434738428, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0039746569890581895, + "repeat_abs_p95": 0.021315647296768075, + "repeat_delta": { + "distinct_n": 8, + "max": 0.024883252660879762, + "min": -0.01469009447770353, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.012588156742608592, + "action_delta": { + "distinct_n": 6, + "max": -0.005728618501492012, + "min": -0.03011422871811248, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 2.4566361209710017, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.005124143797752612, + "repeat_abs_p95": 0.018876919314381714, + "repeat_delta": { + "distinct_n": 8, + "max": 0.011349421052443798, + "min": -0.022930187609271374, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.002835030550300963, + "action_delta": { + "distinct_n": 6, + "max": 0.007512227776197788, + "min": 0.0005459147599019987, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.9399840028075197, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0014613680041681495, + "repeat_abs_p95": 0.004511807887601192, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005553318782013202, + "min": -0.0010848582943063204, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.002861916241176876, + "action_delta": { + "distinct_n": 6, + "max": 0.005666174682015579, + "min": -0.004438767356134465, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.547685400888532, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.00522547476440649, + "repeat_abs_p95": 0.007084887036630881, + "repeat_delta": { + "distinct_n": 8, + "max": 0.007595504930191788, + "min": -0.001265067308722334, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.007920628661389983, + "action_delta": { + "distinct_n": 6, + "max": 0.027975123151637415, + "min": 3.276805657914661e-05, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.977407604410328, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0013250942859486248, + "repeat_abs_p95": 0.0092433410331364, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0011681468294250097, + "min": -0.011994317136490906, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0003016315165324597, + "action_delta": { + "distinct_n": 6, + "max": 0.002385274257794787, + "min": -0.0002979372353175526, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.1832373452641962, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.001646124681066296, + "repeat_abs_p95": 0.004755246907733842, + "repeat_delta": { + "distinct_n": 8, + "max": 0.001880248703786131, + "min": -0.005007302519187606, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00010307207930221329, + "action_delta": { + "distinct_n": 6, + "max": 0.0003313271687591657, + "min": -0.00025035500653580733, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.0193424286695313, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.005328807517567591, + "repeat_abs_p95": 0.008606069879593642, + "repeat_delta": { + "distinct_n": 8, + "max": 0.004217023010198662, + "min": -0.008635306091418382, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.339365533853333, + "action_delta": { + "distinct_n": 6, + "max": 2.133982110813333, + "min": -0.4039724217999998, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.4388333493005985, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.7733357877066664, + "repeat_abs_p95": 3.1493800733973325, + "repeat_delta": { + "distinct_n": 8, + "max": 1.0551951064133327, + "min": -3.3683846067999994, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 1.1423175943466668, + "action_delta": { + "distinct_n": 6, + "max": -0.030365675946666667, + "min": -9.07047515872, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 42.90527668627242, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.02662417498666667, + "repeat_abs_p95": 6.55139452659, + "repeat_delta": { + "distinct_n": 8, + "max": 5.732538843706667, + "min": -6.992316817373333, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.700000000000003, + "action_delta": { + "distinct_n": 6, + "max": 8.959999999999994, + "min": 4.9466666666666725, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.0959651035986926, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 6.11333333333333, + "repeat_abs_p95": 14.715333333333335, + "repeat_delta": { + "distinct_n": 8, + "max": 15.373333333333335, + "min": -4.8799999999999955, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 0.23777777777777778, + "cell": "tp4_mns16", + "completed_fraction": 0.23555555555555557, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.001928339, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.2372549019607843, + "cell": "tp4_mns16", + "completed_fraction": 0.22627450980392155, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.022443998, + "max_internal_gap_s": 0.186336261, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23777777777777778, + "cell": "tp4_mns64", + "completed_fraction": 0.23555555555555557, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004213731, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.2372549019607843, + "cell": "tp4_mns64", + "completed_fraction": 0.23098039215686275, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003542116, + "max_internal_gap_s": 0.184687818, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.23166666666666666, + "cell": "tp4_mns64", + "completed_fraction": 0.22611111111111112, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.007493238, + "max_internal_gap_s": 0.185686919, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.23450980392156862, + "cell": "tp4_mns64", + "completed_fraction": 0.22745098039215686, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.106828533, + "max_internal_gap_s": 0.177241403, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.23166666666666666, + "cell": "tp4_mns16", + "completed_fraction": 0.22499999999999998, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.01771671, + "max_internal_gap_s": 0.174851445, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.23450980392156862, + "cell": "tp4_mns16", + "completed_fraction": 0.2231372549019608, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.049603839, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23, + "cell": "tp4_mns16", + "completed_fraction": 0.2205555555555556, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001863681, + "max_internal_gap_s": 0.6478406, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.23019607843137255, + "cell": "tp4_mns16", + "completed_fraction": 0.2152941176470588, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.17047515, + "max_internal_gap_s": 0.182092338, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23, + "cell": "tp4_mns64", + "completed_fraction": 0.22111111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.000487667, + "max_internal_gap_s": 0.652764433, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.23019607843137255, + "cell": "tp4_mns64", + "completed_fraction": 0.2184313725490196, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.006237065, + "max_internal_gap_s": 0.32072392, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.07082632707989239, + "tpot_mean_over_slo": -0.002512563182247657, + "ttft_max_over_slo_max": -0.0679067323411194, + "ttft_mean_over_slo_max": -0.007600825142914866 + }, + "delta_state": { + "batch_size.mean": -0.20807041709602903, + "batch_tokens.mean": -4.353835378889755, + "decode_batch_size.mean": -0.20629266327798934, + "graph_full_share": 0.0016140185178266098, + "graph_none_share": -0.005313603464127993, + "graph_padding_fraction": 0.001957578860938683, + "kv_usage_end_minus_start": -1.4782427070203497e-06, + "kv_usage_max": 0.004459034805309736, + "kv_usage_mean": -3.034391627598776e-05, + "preemptions": 0.0, + "prefill_token_fraction": 8.331880587686413e-07, + "queue_running_mean": -0.09870830548666643, + "queue_waiting_mean": -0.24030856458, + "scheduler_steps_per_s": 4.060000000000002 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.0027450980392156863, + "completed_over_admitted": 0.0008250825082508628, + "completed_pass_rate": 0.005833333333333357, + "outstanding_over_admitted": -0.0008250825082508247, + "tpot_max_over_slo": 0.2753294892902449, + "tpot_mean_over_slo": 0.018526148273117438, + "ttft_max_over_slo_max": -0.2765812376649895, + "ttft_mean_over_slo_max": -0.08009172416899431 + }, + "delta_state": { + "batch_size.mean": -1.5321059258942888, + "batch_tokens.mean": -32.485419814050886, + "decode_batch_size.mean": -1.5198458327871904, + "graph_full_share": 0.015038490257038628, + "graph_none_share": -0.024038730314888165, + "graph_padding_fraction": 0.006587055652816478, + "kv_usage_end_minus_start": -0.00022317841283692186, + "kv_usage_max": 0.025909186619248126, + "kv_usage_mean": 0.0012400736021554327, + "preemptions": 0.0, + "prefill_token_fraction": -2.8364032274130757e-05, + "queue_running_mean": 0.8917809416333338, + "queue_waiting_mean": -3.7602262873733334, + "scheduler_steps_per_s": 9.026666666666657 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.051139209137158914, + "tpot_mean_over_slo": -0.009262480213267976, + "ttft_max_over_slo_max": -0.03542824399967988, + "ttft_mean_over_slo_max": -0.003943418327769104 + }, + "delta_state": { + "batch_size.mean": -0.37986757946740823, + "batch_tokens.mean": -8.534190946122862, + "decode_batch_size.mean": -0.3768320500033102, + "graph_full_share": 0.0035430578279135316, + "graph_none_share": -0.008383916364329017, + "graph_padding_fraction": 0.001202567354812955, + "kv_usage_end_minus_start": 8.450631670098296e-06, + "kv_usage_max": 0.003782455175254884, + "kv_usage_mean": -0.0002369648399766014, + "preemptions": 0.0, + "prefill_token_fraction": 1.1258245187217852e-06, + "queue_running_mean": -0.3704274981866664, + "queue_waiting_mean": -0.11654787507333333, + "scheduler_steps_per_s": 6.200000000000003 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.14941176470588236, + "completed_over_admitted": 0.044764188649080716, + "completed_pass_rate": 0.32206255283178364, + "outstanding_over_admitted": -0.04476418864908073, + "tpot_max_over_slo": 0.2416662007889263, + "tpot_mean_over_slo": 0.010014700010439026, + "ttft_max_over_slo_max": -0.9367796959995758, + "ttft_mean_over_slo_max": -0.3197808858475722 + }, + "delta_state": { + "batch_size.mean": -0.8883519907951474, + "batch_tokens.mean": -20.195851559300422, + "decode_batch_size.mean": -0.8851481700211252, + "graph_full_share": 0.021219737900431856, + "graph_none_share": -0.03165408628562745, + "graph_padding_fraction": 0.008308104332878177, + "kv_usage_end_minus_start": -0.0007503119935704516, + "kv_usage_max": 0.022672590909591506, + "kv_usage_mean": 0.0015858745872947538, + "preemptions": 0.0, + "prefill_token_fraction": -0.00012884333456097963, + "queue_running_mean": 1.2845878308199996, + "queue_waiting_mean": -16.806316893553333, + "scheduler_steps_per_s": 7.573333333333338 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.001179245283018937, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0011792452830188694, + "tpot_max_over_slo": 0.04993343196889949, + "tpot_mean_over_slo": -0.008105257443938751, + "ttft_max_over_slo_max": -0.026339164838039622, + "ttft_mean_over_slo_max": -0.0029500323901768717 + }, + "delta_state": { + "batch_size.mean": -0.28903355041794665, + "batch_tokens.mean": -5.944731148030087, + "decode_batch_size.mean": -0.2869327863683875, + "graph_full_share": 0.0018351555779636541, + "graph_none_share": -0.005482404566122215, + "graph_padding_fraction": 0.000721905257347253, + "kv_usage_end_minus_start": -0.0025560168080752677, + "kv_usage_max": 0.005265908554136223, + "kv_usage_mean": -0.00021185230120904664, + "preemptions": 0.0, + "prefill_token_fraction": -6.655514854192468e-05, + "queue_running_mean": -0.30910058838666643, + "queue_waiting_mean": -0.06934922306666667, + "scheduler_steps_per_s": 6.219999999999999 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.19137254901960785, + "completed_over_admitted": 0.034146341463414664, + "completed_pass_rate": 0.42068965517241375, + "outstanding_over_admitted": -0.03414634146341463, + "tpot_max_over_slo": 0.1493312743232093, + "tpot_mean_over_slo": 0.00256767021710691, + "ttft_max_over_slo_max": -0.9516305106711418, + "ttft_mean_over_slo_max": -0.3356434934483016 + }, + "delta_state": { + "batch_size.mean": -0.7507094405947292, + "batch_tokens.mean": -14.968838436325854, + "decode_batch_size.mean": -0.7471714629717603, + "graph_full_share": 0.016020663089875264, + "graph_none_share": -0.02573950202606222, + "graph_padding_fraction": 0.006257606960140573, + "kv_usage_end_minus_start": 0.005366359584320524, + "kv_usage_max": 0.012606264387448385, + "kv_usage_mean": 0.0010847467256849628, + "preemptions": 0.0, + "prefill_token_fraction": 7.198152666310076e-05, + "queue_running_mean": 0.8715743724000014, + "queue_waiting_mean": -17.05635223198667, + "scheduler_steps_per_s": 7.959999999999994 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.49444444444444446, + "min": 0.4711111111111111, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 10, + "max": 0.49111111111111116, + "min": 0.4549019607843137, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "ttft_mean_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.07470588235294118, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.0013725490196078432, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.0013725490196078432, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0176627933732168, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0004125412541254314, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0004125412541254314, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.16103127641589182, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0029166666666666785, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0029166666666666785, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.01766279337321675, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00041254125412541233, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.00041254125412541233, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10023524173018411, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.11007880070155085, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.15624626393440935, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.002768793613415921, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 2.755351742962653e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0037510684140956846, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.48610396999962785, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.17224398500305443, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.17224398500305443, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.16186215208767066, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.04384627465595459, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.04384627465595459, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "scheduler_steps_per_s", + "features": { + "decode_batch_size.mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5620017564875353, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5170521246700739, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.6309901100122177, + "train_balanced_accuracy": 1.0 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003730087157476764, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004107592910539628, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.00427231725687758, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0004364472122379581, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0005272014047044875, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0006048648429397225, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.655367559091127e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.64073573609347e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": -1.3765422107681058e-05, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.28123689200666746, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.38643303345666746, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.39653631807333367, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -8.461432384313333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.000267425976667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.000267425976667, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 6.896666666666668, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 7.089999999999996, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 6.88666666666667, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [ + "decode_batch_size.mean", + "graph_padding_fraction", + "kv_usage_mean", + "queue_running_mean", + "scheduler_steps_per_s" + ], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.5, + "end_s": 150.0, + "qualifying_response_features": [ + "graph_padding_fraction" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.5652885100310687, + "action_delta": { + "distinct_n": 6, + "max": -0.20807041709602903, + "min": -1.5321059258942888, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.600949368258532, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.9406591301846219, + "repeat_abs_p95": 1.7016197794945327, + "repeat_delta": { + "distinct_n": 8, + "max": 1.5391576560173217, + "min": -1.749794672064679, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 11.751514691224358, + "action_delta": { + "distinct_n": 6, + "max": -4.353835378889755, + "min": -32.485419814050886, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.3300117500902243, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 35.60938265989476, + "repeat_abs_p95": 53.31315307889063, + "repeat_delta": { + "distinct_n": 8, + "max": 39.85703951344101, + "min": -55.14260767193173, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.5620017564875353, + "action_delta": { + "distinct_n": 6, + "max": -0.20629266327798934, + "min": -1.5198458327871904, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.6013807777033149, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.9345189891732675, + "repeat_abs_p95": 1.6881894328027403, + "repeat_delta": { + "distinct_n": 8, + "max": 1.5259244681432733, + "min": -1.7364812802700182, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.00929077404247608, + "action_delta": { + "distinct_n": 6, + "max": 0.021219737900431856, + "min": 0.0016140185178266098, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.067941220289058, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.008699705438808103, + "repeat_abs_p95": 0.016177686020265352, + "repeat_delta": { + "distinct_n": 8, + "max": 0.014828021569913696, + "min": -0.016904428416608552, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.01621132333960859, + "action_delta": { + "distinct_n": 6, + "max": -0.005313603464127993, + "min": -0.03165408628562745, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 2.171920932023271, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.007464048575887519, + "repeat_abs_p95": 0.014440194586440296, + "repeat_delta": { + "distinct_n": 8, + "max": 0.014932938342374844, + "min": -0.01352509903970471, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.004107592910539628, + "action_delta": { + "distinct_n": 6, + "max": 0.008308104332878177, + "min": 0.000721905257347253, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.563724343406081, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.0007382811686937355, + "repeat_abs_p95": 0.0022186303657579878, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0025539048662290243, + "min": -0.0015959777220260637, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00048674520320368675, + "action_delta": { + "distinct_n": 6, + "max": 0.005366359584320524, + "min": -0.0025560168080752677, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 2, + "zero": 0 + }, + "effect_to_repeat_median": 0.19573347569321622, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0024867754556537336, + "repeat_abs_p95": 0.0054480365502156845, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0058004047507104195, + "min": -0.004793638463582606, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.008936086470792304, + "action_delta": { + "distinct_n": 6, + "max": 0.025909186619248126, + "min": 0.003782455175254884, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 6.294267207393665, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0014197183208706776, + "repeat_abs_p95": 0.007519317647539844, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0022486150575440345, + "min": -0.009341973466342357, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0006608557828307821, + "action_delta": { + "distinct_n": 6, + "max": 0.0015858745872947538, + "min": -0.0002369648399766014, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.449805118176904, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0014692046758144578, + "repeat_abs_p95": 0.002327639544032804, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0016405734067875804, + "min": -0.002503034295596231, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 4.745959040802772e-05, + "action_delta": { + "distinct_n": 6, + "max": 7.198152666310076e-05, + "min": -0.00012884333456097963, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.015433261304563437, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.003075149799608101, + "repeat_abs_p95": 0.005455117882940929, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0032221068728307545, + "min": -0.005478806223512156, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.6210009352933339, + "action_delta": { + "distinct_n": 6, + "max": 1.2845878308199996, + "min": -0.3704274981866664, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.620669066679828, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.0005346949466665, + "repeat_abs_p95": 1.639787432226332, + "repeat_delta": { + "distinct_n": 8, + "max": 1.3395429762466655, + "min": -1.7843421426733315, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 2.000267425976667, + "action_delta": { + "distinct_n": 6, + "max": -0.06934922306666667, + "min": -17.05635223198667, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 82.33375700160043, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.02429462105, + "repeat_abs_p95": 8.566731735707661, + "repeat_delta": { + "distinct_n": 8, + "max": 13.045614636299998, + "min": -0.12448899892666668, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.896666666666668, + "action_delta": { + "distinct_n": 6, + "max": 9.026666666666657, + "min": 4.060000000000002, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.8987836663770638, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 7.673333333333332, + "repeat_abs_p95": 11.486333333333327, + "repeat_delta": { + "distinct_n": 8, + "max": 11.493333333333325, + "min": -9.206666666666663, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 0.47555555555555556, + "cell": "tp4_mns16", + "completed_fraction": 0.4733333333333334, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.001282354, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.4752941176470588, + "cell": "tp4_mns16", + "completed_fraction": 0.47058823529411764, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004049563, + "max_internal_gap_s": 0.186336261, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.47555555555555556, + "cell": "tp4_mns64", + "completed_fraction": 0.4733333333333334, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003246004, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.4752941176470588, + "cell": "tp4_mns64", + "completed_fraction": 0.47098039215686277, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00327497, + "max_internal_gap_s": 0.184687818, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.49444444444444446, + "cell": "tp4_mns64", + "completed_fraction": 0.49111111111111116, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.003493587, + "max_internal_gap_s": 0.444552932, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.49058823529411766, + "cell": "tp4_mns64", + "completed_fraction": 0.48588235294117643, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.0063142, + "max_internal_gap_s": 0.183290846, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.49444444444444446, + "cell": "tp4_mns16", + "completed_fraction": 0.49111111111111116, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.012013597, + "max_internal_gap_s": 0.445602102, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.49058823529411766, + "cell": "tp4_mns16", + "completed_fraction": 0.463921568627451, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.02999801, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.4711111111111111, + "cell": "tp4_mns16", + "completed_fraction": 0.4655555555555555, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.026096497, + "max_internal_gap_s": 0.6478406, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.4823529411764706, + "cell": "tp4_mns16", + "completed_fraction": 0.4549019607843137, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001098192, + "max_internal_gap_s": 0.182092338, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.4711111111111111, + "cell": "tp4_mns64", + "completed_fraction": 0.4661111111111111, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001270441, + "max_internal_gap_s": 0.652764433, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.4823529411764706, + "cell": "tp4_mns64", + "completed_fraction": 0.4713725490196079, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.048916774, + "max_internal_gap_s": 0.32072392, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.13955613873224326, + "tpot_mean_over_slo": 0.00021561369396999597, + "ttft_max_over_slo_max": -0.162763388332678, + "ttft_mean_over_slo_max": -0.0113370356308976 + }, + "delta_state": { + "batch_size.mean": -0.2315625922329474, + "batch_tokens.mean": -4.866425400698262, + "decode_batch_size.mean": -0.2298593364511925, + "graph_full_share": 0.0022019142911884337, + "graph_none_share": -0.0061464601290289606, + "graph_padding_fraction": 0.002459307837583828, + "kv_usage_end_minus_start": 8.281600267778799e-07, + "kv_usage_max": 0.012148644893025695, + "kv_usage_mean": 0.00010369316025509931, + "preemptions": 0.0, + "prefill_token_fraction": -3.204953873736649e-06, + "queue_running_mean": -0.004656362764444921, + "queue_waiting_mean": -0.38515557383555554, + "scheduler_steps_per_s": 4.035555555555547 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.14980392156862746, + "completed_over_admitted": 0.03292618162506633, + "completed_pass_rate": 0.21093318608503586, + "outstanding_over_admitted": -0.032926181625066386, + "tpot_max_over_slo": 0.3362407738628878, + "tpot_mean_over_slo": 0.035823037105198796, + "ttft_max_over_slo_max": -1.0155032691739811, + "ttft_mean_over_slo_max": -0.23418190946398737 + }, + "delta_state": { + "batch_size.mean": -0.8757080766407306, + "batch_tokens.mean": -21.903077459843473, + "decode_batch_size.mean": -0.8701416261315948, + "graph_full_share": 0.015083297248042693, + "graph_none_share": -0.024665889172553218, + "graph_padding_fraction": 0.006691320837900951, + "kv_usage_end_minus_start": -0.007731540433626116, + "kv_usage_max": 0.031275425686576486, + "kv_usage_mean": 0.0024751040544861297, + "preemptions": 0.0, + "prefill_token_fraction": -0.0005203792320278566, + "queue_running_mean": 2.356088796968889, + "queue_waiting_mean": -12.268314980844444, + "scheduler_steps_per_s": 7.142222222222216 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0007401924500369805, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.00074019245003701, + "tpot_max_over_slo": 0.17641319103142994, + "tpot_mean_over_slo": -0.008167534216044714, + "ttft_max_over_slo_max": -0.055667730843803526, + "ttft_mean_over_slo_max": -0.004452474842343956 + }, + "delta_state": { + "batch_size.mean": -0.391971250222225, + "batch_tokens.mean": -8.51800794109036, + "decode_batch_size.mean": -0.3888263577646134, + "graph_full_share": 0.0034883439917251513, + "graph_none_share": -0.00892717433146234, + "graph_padding_fraction": 0.0013420770820979835, + "kv_usage_end_minus_start": -0.0014785130164154392, + "kv_usage_max": 0.012914898602144231, + "kv_usage_mean": -0.00017462402054690008, + "preemptions": 0.0, + "prefill_token_fraction": -1.381071014561197e-05, + "queue_running_mean": -0.33874452614666684, + "queue_waiting_mean": -0.12769319111555558, + "scheduler_steps_per_s": 6.342222222222233 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.3898039215686275, + "completed_over_admitted": 0.047995780590717296, + "completed_pass_rate": 0.5534521158129175, + "outstanding_over_admitted": -0.0479957805907173, + "tpot_max_over_slo": 0.485827943940801, + "tpot_mean_over_slo": 0.015614815582012831, + "ttft_max_over_slo_max": -0.9946015174985708, + "ttft_mean_over_slo_max": -0.5387282368223152 + }, + "delta_state": { + "batch_size.mean": -1.1276789336012651, + "batch_tokens.mean": -25.023441704969116, + "decode_batch_size.mean": -1.1232562014682532, + "graph_full_share": 0.024054684043115837, + "graph_none_share": -0.03557238862010634, + "graph_padding_fraction": 0.008753688501931332, + "kv_usage_end_minus_start": -0.01018813309553912, + "kv_usage_max": 0.04505142122092243, + "kv_usage_mean": 0.0021208568963759205, + "preemptions": 0.0, + "prefill_token_fraction": -0.0001673196203758609, + "queue_running_mean": 1.6553335670222218, + "queue_waiting_mean": -26.90367780516, + "scheduler_steps_per_s": 8.879999999999995 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.06666873637623494, + "tpot_mean_over_slo": -0.009318131681115394, + "ttft_max_over_slo_max": -0.07750733549861857, + "ttft_mean_over_slo_max": -0.004705775197540927 + }, + "delta_state": { + "batch_size.mean": -0.370851789404929, + "batch_tokens.mean": -7.256413232949271, + "decode_batch_size.mean": -0.3681028118073941, + "graph_full_share": 0.0028280490692854654, + "graph_none_share": -0.00702385128031266, + "graph_padding_fraction": 0.0009512228076606274, + "kv_usage_end_minus_start": 3.1056000993068267e-07, + "kv_usage_max": 0.005265908554136223, + "kv_usage_mean": -0.00023540705993194654, + "preemptions": 0.0, + "prefill_token_fraction": 5.342859896728669e-07, + "queue_running_mean": -0.37729419491111127, + "queue_waiting_mean": -0.12941095815111112, + "scheduler_steps_per_s": 7.137777777777785 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.43333333333333335, + "completed_over_admitted": 0.05807814149947199, + "completed_pass_rate": 0.6218345526167699, + "outstanding_over_admitted": -0.05807814149947202, + "tpot_max_over_slo": 0.30513300693329926, + "tpot_mean_over_slo": 0.006471198446666382, + "ttft_max_over_slo_max": -0.9754032485070638, + "ttft_mean_over_slo_max": -0.5503063221993506 + }, + "delta_state": { + "batch_size.mean": -0.6183604933300941, + "batch_tokens.mean": -14.376542002465783, + "decode_batch_size.mean": -0.6159981002912982, + "graph_full_share": 0.01869117884155347, + "graph_none_share": -0.028652094681839838, + "graph_padding_fraction": 0.007072597788281806, + "kv_usage_end_minus_start": 0.0004982293932893622, + "kv_usage_max": 0.02549546289093474, + "kv_usage_mean": 0.0013450068711313955, + "preemptions": 0.0, + "prefill_token_fraction": -0.00032277984328055176, + "queue_running_mean": 1.3239998894977774, + "queue_waiting_mean": -27.13096374471111, + "scheduler_steps_per_s": 7.786666666666676 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.7505555555555555, + "min": 0.7316666666666667, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 9, + "max": 0.7472222222222221, + "min": 0.6968627450980392, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.19490196078431374, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.07490196078431373, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.07490196078431373, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.02436798652037714, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.016463090812533165, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.016833187037551656, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.27672605790645877, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10546659304251793, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10546659304251793, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.024367986520377156, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.016463090812533193, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.016833187037551697, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2407730989823646, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.22234457283277126, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2563269824471589, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.000848167884689166, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003343406070318189, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.007915214637991413, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5264552920028412, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5690833184198709, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5786824529156245, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.27171700600992804, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.12275947254744249, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.12275947254744249, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_running_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5024122290279558, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.49205045604934616, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.6294839919481041, + "train_balanced_accuracy": 1.0 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004207337435189895, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004575314337742389, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004575314337742389, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0005851914252922477, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0007243500156932474, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.00111227502831551, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -9.056516526073644e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.0001629923985771442, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -9.056516526073644e-05, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.49262768167555526, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.6596717633666662, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.8253386021288884, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -13.516544381655557, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -6.32673527734, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -6.32673527734, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 7.4622222222222305, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 7.140000000000001, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 6.742222222222225, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.75, + "end_s": 225.0, + "qualifying_response_features": [ + "graph_padding_fraction", + "queue_waiting_mean" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.5051658717761596, + "action_delta": { + "distinct_n": 6, + "max": -0.2315625922329474, + "min": -1.1276789336012651, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.7593203575357039, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.6652868802512071, + "repeat_abs_p95": 1.054027495945265, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7189335327348694, + "min": -1.232288950040175, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 11.447274971778072, + "action_delta": { + "distinct_n": 6, + "max": -4.866425400698262, + "min": -25.023441704969116, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.575768911929723, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 19.88171770756407, + "repeat_abs_p95": 34.7043135482259, + "repeat_delta": { + "distinct_n": 8, + "max": 13.173463929732975, + "min": -38.394254431520324, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.5024122290279558, + "action_delta": { + "distinct_n": 6, + "max": -0.2298593364511925, + "min": -1.1232562014682532, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.7617660157398797, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.6595361549963323, + "repeat_abs_p95": 1.045411693203953, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7159576534104133, + "min": -1.2228100223235518, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.009285820619883922, + "action_delta": { + "distinct_n": 6, + "max": 0.024054684043115837, + "min": 0.0022019142911884337, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.846168814253066, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0050297787223975154, + "repeat_abs_p95": 0.012718990273054656, + "repeat_delta": { + "distinct_n": 8, + "max": 0.010581275778524746, + "min": -0.013870067308570766, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.01679653175200778, + "action_delta": { + "distinct_n": 6, + "max": -0.0061464601290289606, + "min": -0.03557238862010634, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 4.06293763524991, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.004134085545956116, + "repeat_abs_p95": 0.011398711005122675, + "repeat_delta": { + "distinct_n": 8, + "max": 0.01240808302992423, + "min": -0.009524162959062651, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 0.004575314337742389, + "action_delta": { + "distinct_n": 6, + "max": 0.008753688501931332, + "min": 0.0009512228076606274, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.003367170865857, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.0009144470476570299, + "repeat_abs_p95": 0.0019536589959716906, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0021029671629594227, + "min": -0.0012192099761899004, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.0009883712048524007, + "action_delta": { + "distinct_n": 6, + "max": 0.0004982293932893622, + "min": -0.01018813309553912, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.2973800031975333, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.003323596725486211, + "repeat_abs_p95": 0.006865285634854577, + "repeat_delta": { + "distinct_n": 8, + "max": 0.004233894621933687, + "min": -0.008131118105255575, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.019205180746539485, + "action_delta": { + "distinct_n": 6, + "max": 0.04505142122092243, + "min": 0.005265908554136223, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 7.501452126191231, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0025601950693632802, + "repeat_abs_p95": 0.016512755779962775, + "repeat_delta": { + "distinct_n": 8, + "max": 0.012755779962795644, + "min": -0.01853574275843739, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.000790206965531671, + "action_delta": { + "distinct_n": 6, + "max": 0.0024751040544861297, + "min": -0.00023540705993194654, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 1.0014106568857974, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.000789093824894044, + "repeat_abs_p95": 0.0017615365202587913, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0004699285473880651, + "min": -0.0020330840290943755, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.056516526073644e-05, + "action_delta": { + "distinct_n": 6, + "max": 5.342859896728669e-07, + "min": -0.0005203792320278566, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.05219845322135223, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0017350162633495425, + "repeat_abs_p95": 0.004831444067139573, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0014404459696625294, + "min": -0.004836464815786923, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.8506470422044443, + "action_delta": { + "distinct_n": 6, + "max": 2.356088796968889, + "min": -0.37729419491111127, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 1.3604877117547791, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.6252515438800001, + "repeat_abs_p95": 1.1295860132264428, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7170174645377774, + "min": -1.2455528003599987, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 6.32673527734, + "action_delta": { + "distinct_n": 6, + "max": -0.12769319111555558, + "min": -27.13096374471111, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 1346.2237705715072, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.004699616375555556, + "repeat_abs_p95": 9.605207488385105, + "repeat_delta": { + "distinct_n": 8, + "max": 14.638355337377778, + "min": -0.25793291168444443, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 7.140000000000001, + "action_delta": { + "distinct_n": 6, + "max": 8.879999999999995, + "min": 4.035555555555547, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.6227272727272708, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 4.400000000000006, + "repeat_abs_p95": 9.05933333333333, + "repeat_delta": { + "distinct_n": 8, + "max": 9.337777777777774, + "min": -4.684444444444452, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 0.7388888888888889, + "cell": "tp4_mns16", + "completed_fraction": 0.7377777777777779, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.002047043, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7384313725490196, + "cell": "tp4_mns16", + "completed_fraction": 0.7101960784313726, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.000359161, + "max_internal_gap_s": 0.188948337, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7388888888888889, + "cell": "tp4_mns64", + "completed_fraction": 0.7377777777777779, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003650633, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7384313725490196, + "cell": "tp4_mns64", + "completed_fraction": 0.7345098039215686, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004944353, + "max_internal_gap_s": 0.219527251, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.7505555555555555, + "cell": "tp4_mns64", + "completed_fraction": 0.7472222222222221, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002633243, + "max_internal_gap_s": 0.646724237, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7435294117647059, + "cell": "tp4_mns64", + "completed_fraction": 0.74, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.000558705, + "max_internal_gap_s": 0.183301269, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.7505555555555555, + "cell": "tp4_mns16", + "completed_fraction": 0.7466666666666666, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.001410558, + "max_internal_gap_s": 0.583053185, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7435294117647059, + "cell": "tp4_mns16", + "completed_fraction": 0.7043137254901961, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.042708099, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7316666666666667, + "cell": "tp4_mns16", + "completed_fraction": 0.7311111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.000142513, + "max_internal_gap_s": 0.6478406, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7427450980392157, + "cell": "tp4_mns16", + "completed_fraction": 0.6968627450980392, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.049578949, + "max_internal_gap_s": 0.182904479, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7316666666666667, + "cell": "tp4_mns64", + "completed_fraction": 0.7311111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.00141287, + "max_internal_gap_s": 0.652764433, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7427450980392157, + "cell": "tp4_mns64", + "completed_fraction": 0.74, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.009657729, + "max_internal_gap_s": 0.32072392, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0011111111111111738, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0011111111111111113, + "tpot_max_over_slo": 0.13955613873224326, + "tpot_mean_over_slo": -0.0010090653746841205, + "ttft_max_over_slo_max": -0.162763388332678, + "ttft_mean_over_slo_max": -0.00961120104762887 + }, + "delta_state": { + "batch_size.mean": -0.26379559738229474, + "batch_tokens.mean": -5.348391834808808, + "decode_batch_size.mean": -0.2617666450994136, + "graph_full_share": 0.0022388856413185776, + "graph_none_share": -0.006502634438630757, + "graph_padding_fraction": 0.0021671821218428883, + "kv_usage_end_minus_start": 0.0006863000045611578, + "kv_usage_max": 0.012148644893025695, + "kv_usage_mean": 4.226955534929189e-05, + "preemptions": 0.0, + "prefill_token_fraction": 5.3820087398737115e-05, + "queue_running_mean": -0.06008688729666822, + "queue_waiting_mean": -0.31836611265666664, + "scheduler_steps_per_s": 4.469999999999999 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.3949019607843137, + "completed_over_admitted": 0.03529411764705881, + "completed_pass_rate": 0.41338259441707714, + "outstanding_over_admitted": -0.03529411764705882, + "tpot_max_over_slo": 0.3362407738628878, + "tpot_mean_over_slo": 0.031421107704212525, + "ttft_max_over_slo_max": -1.0149181675127088, + "ttft_mean_over_slo_max": -0.4177688425802161 + }, + "delta_state": { + "batch_size.mean": -0.6909306729068057, + "batch_tokens.mean": -12.641524933835228, + "decode_batch_size.mean": -0.6881317239210976, + "graph_full_share": 0.01650801205775032, + "graph_none_share": -0.026570069621619105, + "graph_padding_fraction": 0.007045814940879979, + "kv_usage_end_minus_start": 0.009741050899909953, + "kv_usage_max": 0.031275425686576486, + "kv_usage_mean": 0.0024717436230722416, + "preemptions": 0.0, + "prefill_token_fraction": 0.00029854606973478415, + "queue_running_mean": 2.260168520989998, + "queue_waiting_mean": -20.98940505023, + "scheduler_steps_per_s": 6.183333333333337 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.17641319103142994, + "tpot_mean_over_slo": -0.004863679616933292, + "ttft_max_over_slo_max": -0.16410399084755528, + "ttft_mean_over_slo_max": -0.005948038735831104 + }, + "delta_state": { + "batch_size.mean": -0.3452936706369112, + "batch_tokens.mean": -7.480013628048113, + "decode_batch_size.mean": -0.34254059469005593, + "graph_full_share": 0.002915066846803871, + "graph_none_share": -0.007828081051085262, + "graph_padding_fraction": 0.0013323004677419665, + "kv_usage_end_minus_start": 9.758748075849866e-06, + "kv_usage_max": 0.01567456253640287, + "kv_usage_mean": -7.388307997985809e-05, + "preemptions": 0.0, + "prefill_token_fraction": 2.1031949494210167e-06, + "queue_running_mean": -0.2080717158233334, + "queue_waiting_mean": -0.18922616308333332, + "scheduler_steps_per_s": 5.61666666666666 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.6321568627450981, + "completed_over_admitted": 0.04980392156862745, + "completed_pass_rate": 0.667771333885667, + "outstanding_over_admitted": -0.04980392156862745, + "tpot_max_over_slo": 0.485827943940801, + "tpot_mean_over_slo": 0.017498560395543927, + "ttft_max_over_slo_max": -0.9946015174985708, + "ttft_mean_over_slo_max": -0.6463312081620042 + }, + "delta_state": { + "batch_size.mean": -0.9812671529958905, + "batch_tokens.mean": -20.775938819398107, + "decode_batch_size.mean": -0.9782503281970616, + "graph_full_share": 0.024736354046330344, + "graph_none_share": -0.03589813374533629, + "graph_padding_fraction": 0.00872979044141211, + "kv_usage_end_minus_start": -0.0018192978173303676, + "kv_usage_max": 0.04505142122092243, + "kv_usage_mean": 0.002220219555318084, + "preemptions": 0.0, + "prefill_token_fraction": 1.1463365605646203e-06, + "queue_running_mean": 1.8554246239900003, + "queue_waiting_mean": -31.984563914093336, + "scheduler_steps_per_s": 8.263333333333335 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0005555555555556424, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0005555555555555552, + "tpot_max_over_slo": 0.18505735732578327, + "tpot_mean_over_slo": -0.007144351590753079, + "ttft_max_over_slo_max": -0.12056507783321042, + "ttft_mean_over_slo_max": -0.006789902099118429 + }, + "delta_state": { + "batch_size.mean": -0.3723911494360035, + "batch_tokens.mean": -7.297855407175092, + "decode_batch_size.mean": -0.3696269360813078, + "graph_full_share": 0.002705515086113608, + "graph_none_share": -0.007253399469024875, + "graph_padding_fraction": 0.001586911205060191, + "kv_usage_end_minus_start": -4.616416023950354e-05, + "kv_usage_max": 0.015583985375003384, + "kv_usage_mean": -0.0001345075402517243, + "preemptions": 0.0, + "prefill_token_fraction": 1.2342954064381573e-06, + "queue_running_mean": -0.3007637260933329, + "queue_waiting_mean": -0.20904505781333332, + "scheduler_steps_per_s": 6.739999999999995 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.6756862745098039, + "completed_over_admitted": 0.05372549019607842, + "completed_pass_rate": 0.7194154488517746, + "outstanding_over_admitted": -0.05372549019607843, + "tpot_max_over_slo": 0.37831041669495047, + "tpot_mean_over_slo": 0.007493808043833983, + "ttft_max_over_slo_max": -0.9795807876701778, + "ttft_mean_over_slo_max": -0.6540993991122069 + }, + "delta_state": { + "batch_size.mean": -0.8431446284631257, + "batch_tokens.mean": -19.08621547820553, + "decode_batch_size.mean": -0.839780585287075, + "graph_full_share": 0.021245526407298043, + "graph_none_share": -0.03191300655092645, + "graph_padding_fraction": 0.00783703914940752, + "kv_usage_end_minus_start": 0.000710563518775853, + "kv_usage_max": 0.033983984733163775, + "kv_usage_mean": 0.0015253876505999867, + "preemptions": 0.0, + "prefill_token_fraction": -0.00035310442232838746, + "queue_running_mean": 1.397843253986668, + "queue_waiting_mean": -32.16364194347333, + "scheduler_steps_per_s": 8.763333333333335 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 1, + "max": 1.0, + "min": 1.0, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 10, + "max": 0.9966666666666667, + "min": 0.9392156862745098, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_mean_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.31607843137254904, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.19745098039215686, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.19745098039215686, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.025179738562091547, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.01820261437908499, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.01820261437908499, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.3338856669428335, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.20669129720853857, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.20669129720853857, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.025179738562091502, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.018202614379084967, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.018202614379084967, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.28168388701036684, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.26064906559433554, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2563269824471589, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0013150642134503454, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003242371334574931, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.008244747510429903, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5718423892588665, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5711720880014279, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5793527541730631, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.3265605551305613, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.2136900218139225, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.2136900218139225, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.6047037606841914, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5288793300012027, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5153361593055767, + "train_balanced_accuracy": 1.0 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004711975177233855, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004606498531361434, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004606498531361434, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0007257522853100643, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0007838286029746393, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0011312445553336879, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 1.1903159835013888e-06, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.00017593506346097465, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 1.6247657549928185e-06, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.5948857690816673, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.6688781833449999, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.897668868346666, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -16.096804485953335, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -10.653885581443333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -10.653885581443333, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 7.501666666666665, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 5.326666666666668, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 5.899999999999999, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 1.0, + "end_s": 300.0, + "qualifying_response_features": [ + "graph_padding_fraction", + "queue_waiting_mean" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.5316609111714046, + "action_delta": { + "distinct_n": 6, + "max": -0.26379559738229474, + "min": -0.9812671529958905, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 1.4818088309881625, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.358791836067585, + "repeat_abs_p95": 0.9062470454242428, + "repeat_delta": { + "distinct_n": 8, + "max": 0.5513066192162643, + "min": -0.9545899290107105, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 10.06076928094167, + "action_delta": { + "distinct_n": 6, + "max": -5.348391834808808, + "min": -20.775938819398107, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.573954119692525, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 17.528873712643374, + "repeat_abs_p95": 31.853379856357762, + "repeat_delta": { + "distinct_n": 8, + "max": 13.785312549619732, + "min": -32.444783025775166, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.5288793300012027, + "action_delta": { + "distinct_n": 6, + "max": -0.2617666450994136, + "min": -0.9782503281970616, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 1.4866497181990304, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.3557524839421502, + "repeat_abs_p95": 0.8989199315256944, + "repeat_delta": { + "distinct_n": 8, + "max": 0.5489012995649425, + "min": -0.9473843415441898, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.009711539452277096, + "action_delta": { + "distinct_n": 6, + "max": 0.024736354046330344, + "min": 0.0022388856413185776, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 3.945566969995893, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0024613799553089843, + "repeat_abs_p95": 0.009783132290188977, + "repeat_delta": { + "distinct_n": 8, + "max": 0.008076086279964478, + "min": -0.010702310911079094, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.017199075336352183, + "action_delta": { + "distinct_n": 6, + "max": -0.006502634438630757, + "min": -0.03589813374533629, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 6.047036959442635, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.002844215349055424, + "repeat_abs_p95": 0.008425467057080636, + "repeat_delta": { + "distinct_n": 8, + "max": 0.00925018771871168, + "min": -0.006893842971194414, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.8333333333333334, + "action_abs_median": 0.004606498531361434, + "action_delta": { + "distinct_n": 6, + "max": 0.00872979044141211, + "min": 0.0013323004677419665, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.74258050189957, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.0008021652512903675, + "repeat_abs_p95": 0.0013850876756400163, + "repeat_delta": { + "distinct_n": 8, + "max": 0.001451018727788989, + "min": -0.0009098361529306493, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0006984317616685054, + "action_delta": { + "distinct_n": 6, + "max": 0.009741050899909953, + "min": -0.0018192978173303676, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.1291097203463754, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.005409598594085352, + "repeat_abs_p95": 0.012010926224984144, + "repeat_delta": { + "distinct_n": 8, + "max": 0.002621680737545762, + "min": -0.01455978249759804, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 1.0, + "action_abs_median": 0.023474994111489678, + "action_delta": { + "distinct_n": 6, + "max": 0.04505142122092243, + "min": 0.012148644893025695, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 23.009837103169843, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.001020215571550298, + "repeat_abs_p95": 0.011807784296490091, + "repeat_delta": { + "distinct_n": 8, + "max": 0.012755779962795644, + "min": -0.010047220916208355, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.0008299475954258555, + "action_delta": { + "distinct_n": 6, + "max": 0.0024717436230722416, + "min": -0.0001345075402517243, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 1.2357530752548687, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.000671612810071043, + "repeat_abs_p95": 0.0016041738838027961, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0005327474947367915, + "min": -0.0018473650504541306, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 2.7961641174079066e-05, + "action_delta": { + "distinct_n": 6, + "max": 0.00029854606973478415, + "min": -0.00035310442232838746, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.014748067700673472, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0018959528625436262, + "repeat_abs_p95": 0.0048750185359022645, + "repeat_delta": { + "distinct_n": 8, + "max": 0.001808064432466061, + "min": -0.004875322650742309, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.8493034900400005, + "action_delta": { + "distinct_n": 6, + "max": 2.260168520989998, + "min": -0.3007637260933329, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 2.784700761391589, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.30498913987999954, + "repeat_abs_p95": 0.9830922326288332, + "repeat_delta": { + "distinct_n": 8, + "max": 0.5377630984033335, + "min": -1.1432457121299997, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 10.653885581443333, + "action_delta": { + "distinct_n": 6, + "max": -0.18922616308333332, + "min": -32.16364194347333, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 898.7004216601898, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.01185476864666667, + "repeat_abs_p95": 7.209323996711327, + "repeat_delta": { + "distinct_n": 8, + "max": 10.997039040103335, + "min": -0.12967447423333334, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 6.461666666666666, + "action_delta": { + "distinct_n": 6, + "max": 8.763333333333335, + "min": 4.469999999999999, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.8488316642823093, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.4949999999999974, + "repeat_abs_p95": 6.136833333333333, + "repeat_delta": { + "distinct_n": 8, + "max": 6.530000000000001, + "min": -3.1566666666666663, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.99, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00637513, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9552941176470588, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00415122, + "max_internal_gap_s": 0.188948337, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9911111111111112, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.03110111, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9905882352941177, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.033263413, + "max_internal_gap_s": 0.325952494, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002571552, + "max_internal_gap_s": 0.646724237, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9964705882352941, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002910387, + "max_internal_gap_s": 0.26729158, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.005037876, + "max_internal_gap_s": 0.583053185, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9466666666666667, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.04911377, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9961111111111111, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001459145, + "max_internal_gap_s": 0.6478406, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9392156862745098, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.024192492, + "max_internal_gap_s": 0.182904479, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.004830812, + "max_internal_gap_s": 0.652764433, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9929411764705882, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.021026794, + "max_internal_gap_s": 0.32072392, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + } + ], + "decision": "STOP_NO_INCREMENTAL_TUNING_SIGNAL", + "load_consistency": { + "0.50->0.75:graph_padding_fraction": { + "consistent_load_regimes": [ + "low", + "mid" + ], + "direction": 1, + "passes_two_regimes": true + }, + "0.75->1.00:graph_padding_fraction": { + "consistent_load_regimes": [ + "low", + "mid" + ], + "direction": 1, + "passes_two_regimes": true + }, + "0.75->1.00:queue_waiting_mean": { + "consistent_load_regimes": [ + "low", + "mid" + ], + "direction": -1, + "passes_two_regimes": true + } + }, + "mechanism_features": [ + "graph_padding_fraction", + "queue_waiting_mean" + ], + "mechanism_gate": { + "by_transition": { + "0.10->0.25": [], + "0.25->0.50": [], + "0.50->0.75": [ + "graph_padding_fraction" + ], + "0.75->1.00": [ + "graph_padding_fraction", + "queue_waiting_mean" + ] + }, + "minimum_features": 2, + "passes": true, + "passing_transitions": [ + "0.75->1.00" + ] + }, + "provenance": { + "analysis_script": "/home/admin/cpfs/wjh/aituner/aituner-phase-aware-20260714/runs/intervention-response-v2/analyze_pilot.py", + "analysis_script_sha256": "4467aa8e3d1f3c3308847a9bc5b8846daa0b4381966fd8ec3643a8a56e017b08", + "manifest": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/pilot-manifest.json", + "manifest_sha256": "273db1181dcc9d6b64439650d0642ebe553b12e6aa9adebfbe3758a7977e5611", + "run_root": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot", + "streams": [ + { + "bytes": 54326667, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep1-mns16/opprof/opprof-v1-dp0-pid458427-1784022917661752826.jsonl", + "session": "rep1-mns16", + "sha256": "1c1418223dbc4e21585a9e642e95ff60be53a5762483a8fdf56ea2cf300cc3fb" + }, + { + "bytes": 56460016, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep1-mns64/opprof/opprof-v1-dp0-pid471212-1784023695315485142.jsonl", + "session": "rep1-mns64", + "sha256": "db4c76070c5d005922fa164545766c84e25ace254793bc64ffeeae8edc8e7a11" + }, + { + "bytes": 55778303, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep2-mns64/opprof/opprof-v1-dp0-pid483758-1784024456752184540.jsonl", + "session": "rep2-mns64", + "sha256": "7c24a7059ff6ce171fc38237c0a7ccd5dc7dcc4ccefde5e07d9c8a276969e004" + }, + { + "bytes": 52931532, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep2-mns16/opprof/opprof-v1-dp0-pid496100-1784025206790752505.jsonl", + "session": "rep2-mns16", + "sha256": "db3a43cecf82adfe71acec9e47e45445944277076a54ee5e14f91381c26aa6ca" + }, + { + "bytes": 55604208, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep3-mns16/opprof/opprof-v1-dp0-pid508913-1784025983046041167.jsonl", + "session": "rep3-mns16", + "sha256": "5465eb5ec08568152bf854597953ada99956b74a3cf8377c4a5c5a544eb8a5b7" + }, + { + "bytes": 58550249, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep3-mns64/opprof/opprof-v1-dp0-pid521797-1784026767359409666.jsonl", + "session": "rep3-mns64", + "sha256": "7ef67bcd4f9e1ab6b597a7eae03da466677eeba756d1206d02575c497dfd788e" + } + ] + }, + "quarter_blocks": [ + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.07082632707989239, + "tpot_mean_over_slo": -0.00490180963817316, + "ttft_max_over_slo_max": -0.04153848900265682, + "ttft_mean_over_slo_max": -0.006891098180069163 + }, + "delta_state": { + "batch_size.mean": -0.2638104929979219, + "batch_tokens.mean": -5.845648825359461, + "decode_batch_size.mean": -0.26164909366742517, + "graph_full_share": 0.002180359902912321, + "graph_none_share": -0.006323059607193614, + "graph_padding_fraction": 0.0018335461795120685, + "kv_usage_end_minus_start": 1.026766711520466e-05, + "kv_usage_max": 0.004459034805309736, + "kv_usage_mean": -0.00014393587383104527, + "preemptions": 0.0, + "prefill_token_fraction": -2.725586323304441e-05, + "queue_running_mean": -0.18631020918666685, + "queue_waiting_mean": -0.20975605228, + "scheduler_steps_per_s": 4.9466666666666725 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.00196078431372549, + "completed_over_admitted": 0.01983471074380172, + "completed_pass_rate": 0.008665511265164683, + "outstanding_over_admitted": -0.01983471074380165, + "tpot_max_over_slo": 0.2753294892902449, + "tpot_mean_over_slo": 0.04011748202807386, + "ttft_max_over_slo_max": -0.2852395010025551, + "ttft_mean_over_slo_max": -0.06592333248276636 + }, + "delta_state": { + "batch_size.mean": -0.8231037548470876, + "batch_tokens.mean": -19.74330932592642, + "decode_batch_size.mean": -0.8152070406366647, + "graph_full_share": 0.00839719842479858, + "graph_none_share": -0.018146448159773024, + "graph_padding_fraction": 0.005340401585168801, + "kv_usage_end_minus_start": -0.004438767356134465, + "kv_usage_max": 0.027975123151637415, + "kv_usage_mean": 0.002385274257794787, + "preemptions": 0.0, + "prefill_token_fraction": -0.00025035500653580733, + "queue_running_mean": 2.133982110813333, + "queue_waiting_mean": -3.33810440732, + "scheduler_steps_per_s": 6.320000000000007 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.004796163069544446, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.004796163069544366, + "tpot_max_over_slo": 0.051139209137158914, + "tpot_mean_over_slo": -0.0074211636478002385, + "ttft_max_over_slo_max": -0.03542824399967988, + "ttft_mean_over_slo_max": -0.0033018737683347885 + }, + "delta_state": { + "batch_size.mean": -0.2685691864197386, + "batch_tokens.mean": -6.959429182398708, + "decode_batch_size.mean": -0.26598172047692614, + "graph_full_share": 0.0028314394428595735, + "graph_none_share": -0.007209322700149463, + "graph_padding_fraction": 0.0012877200176958283, + "kv_usage_end_minus_start": -0.0011333814702583211, + "kv_usage_max": 0.0022799714777140823, + "kv_usage_mean": -0.0001951815494633533, + "preemptions": 0.0, + "prefill_token_fraction": -0.00014483809619469934, + "queue_running_mean": -0.2747586459066662, + "queue_waiting_mean": -0.08017956545333334, + "scheduler_steps_per_s": 5.586666666666659 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.02980392156862745, + "completed_over_admitted": 0.018394648829431426, + "completed_pass_rate": 0.13356766256590513, + "outstanding_over_admitted": -0.01839464882943144, + "tpot_max_over_slo": 0.23737520567053472, + "tpot_mean_over_slo": 0.010490282370595683, + "ttft_max_over_slo_max": -0.38453158099825185, + "ttft_mean_over_slo_max": -0.19098879755079737 + }, + "delta_state": { + "batch_size.mean": -1.413753104794269, + "batch_tokens.mean": -30.311859850911844, + "decode_batch_size.mean": -1.4061952013981784, + "graph_full_share": 0.021360307900858633, + "graph_none_share": -0.03011422871811248, + "graph_padding_fraction": 0.007512227776197788, + "kv_usage_end_minus_start": 0.0021360977816522464, + "kv_usage_max": 0.022672590909591506, + "kv_usage_mean": 0.0010254846876152635, + "preemptions": 0.0, + "prefill_token_fraction": 0.0003313271687591657, + "queue_running_mean": 0.7838954335199997, + "queue_waiting_mean": -9.07047515872, + "scheduler_steps_per_s": 8.959999999999994 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0024154589371980784, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0024154589371980714, + "tpot_max_over_slo": -0.006984168036628269, + "tpot_mean_over_slo": -0.01111313281352852, + "ttft_max_over_slo_max": -0.017940445327743276, + "ttft_mean_over_slo_max": -0.0018007220652844075 + }, + "delta_state": { + "batch_size.mean": -0.3420664768336481, + "batch_tokens.mean": -6.950334498514508, + "decode_batch_size.mean": -0.3395672760130166, + "graph_full_share": 0.0029764291623558536, + "graph_none_share": -0.005728618501492012, + "graph_padding_fraction": 0.0005459147599019987, + "kv_usage_end_minus_start": -0.003587734700701506, + "kv_usage_max": 3.276805657914661e-05, + "kv_usage_mean": -0.0002979372353175526, + "preemptions": 0.0, + "prefill_token_fraction": -6.130606240972725e-05, + "queue_running_mean": -0.4039724217999998, + "queue_waiting_mean": -0.030365675946666667, + "scheduler_steps_per_s": 7.826666666666668 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.013628620102214661, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.013628620102214647, + "tpot_max_over_slo": 0.08779462519483128, + "tpot_mean_over_slo": -0.005083324937856265, + "ttft_max_over_slo_max": -0.17233173300822574, + "ttft_mean_over_slo_max": -0.03508650576710582 + }, + "delta_state": { + "batch_size.mean": -0.5655282844513536, + "batch_tokens.mean": -11.412062377336781, + "decode_batch_size.mean": -0.562525695471205, + "graph_full_share": 0.009956297228293054, + "graph_none_share": -0.01796699078506772, + "graph_padding_fraction": 0.0038365149210898572, + "kv_usage_end_minus_start": 0.005666174682015579, + "kv_usage_max": 0.01138222251747023, + "kv_usage_mean": 0.0003053257977473668, + "preemptions": 0.0, + "prefill_token_fraction": -1.837494489209668e-05, + "queue_running_mean": 0.1581681952266667, + "queue_waiting_mean": -2.0748791364133337, + "scheduler_steps_per_s": 7.079999999999998 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.23777777777777778, + "min": 0.23, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 11, + "max": 0.23555555555555557, + "min": 0.2152941176470588, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.014901960784313726, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.000980392156862745, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.000980392156862745, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.009212391585879554, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.00802203951970637, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.011595405949487936, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.06678383128295257, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004332755632582341, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.004332755632582341, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.009212391585879507, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00802203951970636, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.011595405949487903, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0694669171659951, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.07931047613736183, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.15410076637521355, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.006252244292828252, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.008098228875692393, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0027942363662112613, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.1038799885039528, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.10693511100544129, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.16338899500260595, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.019194189767720304, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.02098880197358749, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.03640721533141776, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.4510464857421108, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.4510464857421108, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5405943805567954, + "train_balanced_accuracy": 1.0 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0025621174693928428, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.002835030550300963, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003586973882340435, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 5.5072124142006764e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 8.069496195816078e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0004407744068921091, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": -3.984050365091196e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.2815404062570543e-05, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.00015203565276306064, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.05829522533999976, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.014071006980000078, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.2987926121666664, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.0775293509333335, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.1423175943466668, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -1.7739302298, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 6.333333333333329, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 5.63333333333334, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 5.953333333333333, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.25, + "end_s": 75.0, + "qualifying_response_features": [], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.45379738064250086, + "action_delta": { + "distinct_n": 6, + "max": -0.2638104929979219, + "min": -1.413753104794269, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.9705757741391393, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.467554819246339, + "repeat_abs_p95": 2.634518964255186, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7001454483303338, + "min": -2.9313976513752067, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.185745779867744, + "action_delta": { + "distinct_n": 6, + "max": -5.845648825359461, + "min": -30.311859850911844, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.298783186954296, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 30.743850996116663, + "repeat_abs_p95": 93.908100721683, + "repeat_delta": { + "distinct_n": 8, + "max": 27.153830102554593, + "min": -100.52302983743428, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4510464857421108, + "action_delta": { + "distinct_n": 6, + "max": -0.26164909366742517, + "min": -1.4061952013981784, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.9732742893789799, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.4634320362350386, + "repeat_abs_p95": 2.613462716433853, + "repeat_delta": { + "distinct_n": 8, + "max": 0.6993897150480155, + "min": -2.908747043508294, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.005686813793577217, + "action_delta": { + "distinct_n": 6, + "max": 0.021360307900858633, + "min": 0.002180359902912321, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.4307684434738428, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0039746569890581895, + "repeat_abs_p95": 0.021315647296768075, + "repeat_delta": { + "distinct_n": 8, + "max": 0.024883252660879762, + "min": -0.01469009447770353, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.012588156742608592, + "action_delta": { + "distinct_n": 6, + "max": -0.005728618501492012, + "min": -0.03011422871811248, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 2.4566361209710017, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.005124143797752612, + "repeat_abs_p95": 0.018876919314381714, + "repeat_delta": { + "distinct_n": 8, + "max": 0.011349421052443798, + "min": -0.022930187609271374, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.002835030550300963, + "action_delta": { + "distinct_n": 6, + "max": 0.007512227776197788, + "min": 0.0005459147599019987, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.9399840028075197, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0014613680041681495, + "repeat_abs_p95": 0.004511807887601192, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005553318782013202, + "min": -0.0010848582943063204, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.002861916241176876, + "action_delta": { + "distinct_n": 6, + "max": 0.005666174682015579, + "min": -0.004438767356134465, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.547685400888532, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.00522547476440649, + "repeat_abs_p95": 0.007084887036630881, + "repeat_delta": { + "distinct_n": 8, + "max": 0.007595504930191788, + "min": -0.001265067308722334, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.007920628661389983, + "action_delta": { + "distinct_n": 6, + "max": 0.027975123151637415, + "min": 3.276805657914661e-05, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.977407604410328, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0013250942859486248, + "repeat_abs_p95": 0.0092433410331364, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0011681468294250097, + "min": -0.011994317136490906, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0003016315165324597, + "action_delta": { + "distinct_n": 6, + "max": 0.002385274257794787, + "min": -0.0002979372353175526, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.1832373452641962, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.001646124681066296, + "repeat_abs_p95": 0.004755246907733842, + "repeat_delta": { + "distinct_n": 8, + "max": 0.001880248703786131, + "min": -0.005007302519187606, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00010307207930221329, + "action_delta": { + "distinct_n": 6, + "max": 0.0003313271687591657, + "min": -0.00025035500653580733, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.0193424286695313, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.005328807517567591, + "repeat_abs_p95": 0.008606069879593642, + "repeat_delta": { + "distinct_n": 8, + "max": 0.004217023010198662, + "min": -0.008635306091418382, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.339365533853333, + "action_delta": { + "distinct_n": 6, + "max": 2.133982110813333, + "min": -0.4039724217999998, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.4388333493005985, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.7733357877066664, + "repeat_abs_p95": 3.1493800733973325, + "repeat_delta": { + "distinct_n": 8, + "max": 1.0551951064133327, + "min": -3.3683846067999994, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 1.1423175943466668, + "action_delta": { + "distinct_n": 6, + "max": -0.030365675946666667, + "min": -9.07047515872, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 42.90527668627242, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.02662417498666667, + "repeat_abs_p95": 6.55139452659, + "repeat_delta": { + "distinct_n": 8, + "max": 5.732538843706667, + "min": -6.992316817373333, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.700000000000003, + "action_delta": { + "distinct_n": 6, + "max": 8.959999999999994, + "min": 4.9466666666666725, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.0959651035986926, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 6.11333333333333, + "repeat_abs_p95": 14.715333333333335, + "repeat_delta": { + "distinct_n": 8, + "max": 15.373333333333335, + "min": -4.8799999999999955, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 0.0, + "trial_sanity": [ + { + "admitted_fraction": 0.23777777777777778, + "cell": "tp4_mns16", + "completed_fraction": 0.23555555555555557, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.001928339, + "max_internal_gap_s": 0.648316718, + "start_gap_s": 0.042673865 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.2372549019607843, + "cell": "tp4_mns16", + "completed_fraction": 0.22627450980392155, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.022443998, + "max_internal_gap_s": 0.186336261, + "start_gap_s": 0.042930986 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23777777777777778, + "cell": "tp4_mns64", + "completed_fraction": 0.23555555555555557, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004213731, + "max_internal_gap_s": 0.637319097, + "start_gap_s": 0.043186851 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.2372549019607843, + "cell": "tp4_mns64", + "completed_fraction": 0.23098039215686275, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003542116, + "max_internal_gap_s": 0.184687818, + "start_gap_s": 0.043187905 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.23166666666666666, + "cell": "tp4_mns64", + "completed_fraction": 0.22611111111111112, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.007493238, + "max_internal_gap_s": 0.185686919, + "start_gap_s": 0.107508086 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.23450980392156862, + "cell": "tp4_mns64", + "completed_fraction": 0.22745098039215686, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.106828533, + "max_internal_gap_s": 0.177241403, + "start_gap_s": 0.107570476 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.23166666666666666, + "cell": "tp4_mns16", + "completed_fraction": 0.22499999999999998, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.01771671, + "max_internal_gap_s": 0.174851445, + "start_gap_s": 0.107276851 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.23450980392156862, + "cell": "tp4_mns16", + "completed_fraction": 0.2231372549019608, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.049603839, + "max_internal_gap_s": 0.184443946, + "start_gap_s": 0.10804117 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23, + "cell": "tp4_mns16", + "completed_fraction": 0.2205555555555556, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001863681, + "max_internal_gap_s": 0.6478406, + "start_gap_s": 0.120533656 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.23019607843137255, + "cell": "tp4_mns16", + "completed_fraction": 0.2152941176470588, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.17047515, + "max_internal_gap_s": 0.182092338, + "start_gap_s": 0.124699964 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.23, + "cell": "tp4_mns64", + "completed_fraction": 0.22111111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.000487667, + "max_internal_gap_s": 0.652764433, + "start_gap_s": 0.123493588 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.23019607843137255, + "cell": "tp4_mns64", + "completed_fraction": 0.2184313725490196, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.006237065, + "max_internal_gap_s": 0.32072392, + "start_gap_s": 0.122478295 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.07082632707989239, + "tpot_mean_over_slo": -0.002512563182247657, + "ttft_max_over_slo_max": -0.0679067323411194, + "ttft_mean_over_slo_max": -0.007600825142914866 + }, + "delta_state": { + "batch_size.mean": -0.15804783275966194, + "batch_tokens.mean": -3.0894547775551757, + "decode_batch_size.mean": -0.15660557052023627, + "graph_full_share": 0.001103652377702713, + "graph_none_share": -0.0043728510978748095, + "graph_padding_fraction": 0.0020921453623694904, + "kv_usage_end_minus_start": -1.1745909822225009e-05, + "kv_usage_max": 0.003994488881572833, + "kv_usage_mean": 8.324804127910184e-05, + "preemptions": 0.0, + "prefill_token_fraction": 3.1223127688440044e-05, + "queue_running_mean": -0.011106401786666886, + "queue_waiting_mean": -0.27086107688, + "scheduler_steps_per_s": 3.173333333333332 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.0027450980392156863, + "completed_over_admitted": 0.0008250825082508628, + "completed_pass_rate": 0.005833333333333357, + "outstanding_over_admitted": -0.0008250825082508247, + "tpot_max_over_slo": 0.2753294892902449, + "tpot_mean_over_slo": 0.018526148273117438, + "ttft_max_over_slo_max": -0.2765812376649895, + "ttft_mean_over_slo_max": -0.08009172416899431 + }, + "delta_state": { + "batch_size.mean": -2.2691156081486135, + "batch_tokens.mean": -44.89764761415685, + "decode_batch_size.mean": -2.2524677264815143, + "graph_full_share": 0.021894194879364726, + "graph_none_share": -0.030171302893482743, + "graph_padding_fraction": 0.007904176391026585, + "kv_usage_end_minus_start": 0.002935218401001971, + "kv_usage_max": 0.01698116274304451, + "kv_usage_mean": 9.548656867379608e-05, + "preemptions": 0.0, + "prefill_token_fraction": 0.00015029680675249235, + "queue_running_mean": -0.35042022754666746, + "queue_waiting_mean": -4.181868913853333, + "scheduler_steps_per_s": 11.733333333333334 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.051139209137158914, + "tpot_mean_over_slo": -0.009262480213267976, + "ttft_max_over_slo_max": -0.03542824399967988, + "ttft_mean_over_slo_max": -0.003943418327769104 + }, + "delta_state": { + "batch_size.mean": -0.5154516769784294, + "batch_tokens.mean": -10.397397526512549, + "decode_batch_size.mean": -0.511830034908046, + "graph_full_share": 0.004410718283113013, + "graph_none_share": -0.009755038550069609, + "graph_padding_fraction": 0.0011193731937274784, + "kv_usage_end_minus_start": 0.0011418132801096492, + "kv_usage_max": 0.003853870265263404, + "kv_usage_mean": -0.0002787530429213137, + "preemptions": 0.0, + "prefill_token_fraction": 0.00014412819519771247, + "queue_running_mean": -0.46609635046666753, + "queue_waiting_mean": -0.15291618469333335, + "scheduler_steps_per_s": 6.813333333333333 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.14941176470588236, + "completed_over_admitted": 0.044764188649080716, + "completed_pass_rate": 0.32206255283178364, + "outstanding_over_admitted": -0.04476418864908073, + "tpot_max_over_slo": 0.2416662007889263, + "tpot_mean_over_slo": 0.010014700010439026, + "ttft_max_over_slo_max": -0.9367796959995758, + "ttft_mean_over_slo_max": -0.3197808858475722 + }, + "delta_state": { + "batch_size.mean": -0.2846481455922074, + "batch_tokens.mean": -9.105088784407371, + "decode_batch_size.mean": -0.28638908954680886, + "graph_full_share": 0.02080155419765417, + "graph_none_share": -0.03302550516502446, + "graph_padding_fraction": 0.009148827020083142, + "kv_usage_end_minus_start": -0.0029587377234424617, + "kv_usage_max": 0.021920351504425173, + "kv_usage_mean": 0.0021466714075248396, + "preemptions": 0.0, + "prefill_token_fraction": -0.0004908425008621764, + "queue_running_mean": 1.7869539801733332, + "queue_waiting_mean": -24.542158628386666, + "scheduler_steps_per_s": 6.186666666666667 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.001179245283018937, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0011792452830188694, + "tpot_max_over_slo": 0.04993343196889949, + "tpot_mean_over_slo": -0.008105257443938751, + "ttft_max_over_slo_max": -0.026339164838039622, + "ttft_mean_over_slo_max": -0.0029500323901768717 + }, + "delta_state": { + "batch_size.mean": -0.22671705086071015, + "batch_tokens.mean": -4.740617496917622, + "decode_batch_size.mean": -0.22507516605851396, + "graph_full_share": 0.0006042607986376236, + "graph_none_share": -0.005169444156433126, + "graph_padding_fraction": 0.0008846012466042618, + "kv_usage_end_minus_start": 0.003668730096473505, + "kv_usage_max": 0.008719338263833865, + "kv_usage_mean": -0.00012598990958833517, + "preemptions": 0.0, + "prefill_token_fraction": -7.139411620438185e-05, + "queue_running_mean": -0.21431314688000036, + "queue_waiting_mean": -0.10833277018666666, + "scheduler_steps_per_s": 4.613333333333344 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.19137254901960785, + "completed_over_admitted": 0.034146341463414664, + "completed_pass_rate": 0.42068965517241375, + "outstanding_over_admitted": -0.03414634146341463, + "tpot_max_over_slo": 0.1493312743232093, + "tpot_mean_over_slo": 0.00256767021710691, + "ttft_max_over_slo_max": -0.9516305106711418, + "ttft_mean_over_slo_max": -0.3356434934483016 + }, + "delta_state": { + "batch_size.mean": -1.1031807575648234, + "batch_tokens.mean": -22.138099877318325, + "decode_batch_size.mean": -1.098093622958313, + "graph_full_share": 0.024621184785722505, + "graph_none_share": -0.036507530380663095, + "graph_padding_fraction": 0.008444758995033443, + "kv_usage_end_minus_start": -0.0009272853178361462, + "kv_usage_max": 0.012606264387448385, + "kv_usage_mean": 0.0018642037638938545, + "preemptions": 0.0, + "prefill_token_fraction": 0.0001249562235060564, + "queue_running_mean": 1.5849805495733342, + "queue_waiting_mean": -32.03782532756, + "scheduler_steps_per_s": 8.840000000000003 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.49444444444444446, + "min": 0.4711111111111111, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 10, + "max": 0.49111111111111116, + "min": 0.4549019607843137, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 0.8333333333333334, + "best_balanced_accuracy": 0.8333333333333333, + "best_feature": "ttft_mean_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.07470588235294118, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.0013725490196078432, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.0013725490196078432, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0176627933732168, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0004125412541254314, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0004125412541254314, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.16103127641589182, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0029166666666666785, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0029166666666666785, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.01766279337321675, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00041254125412541233, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.00041254125412541233, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10023524173018411, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.11007880070155085, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.15624626393440935, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.002768793613415921, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 2.755351742962653e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0037510684140956846, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.48610396999962785, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.17224398500305443, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.17224398500305443, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.16186215208767066, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.04384627465595459, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.04384627465595459, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "graph_padding_fraction", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.2557321278026614, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.6615843945084134, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.22149733003352257, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004782066094380461, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004998160876698038, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004998160876698038, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0008691069271527596, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 8.936730497644896e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 8.936730497644896e-05, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.16666666666666666, + "balanced_accuracy": 0.16666666666666666, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": 0.00013454220935188443, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": 7.808967559724822e-05, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0001472125009751024, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_running_mean": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.6853337013466669, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": -0.2823666872133339, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.4082582890066675, + "train_balanced_accuracy": 0.75 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -12.347537406539999, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.2263649953666667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -2.2263649953666667, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 5.400000000000006, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": 6.726666666666674, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 4.68, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [ + "graph_padding_fraction" + ], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.5, + "end_s": 150.0, + "qualifying_response_features": [ + "graph_padding_fraction" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4000499112853184, + "action_delta": { + "distinct_n": 6, + "max": -0.15804783275966194, + "min": -2.2691156081486135, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.2802630070966653, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 1.4274089021935654, + "repeat_abs_p95": 2.672248195457046, + "repeat_delta": { + "distinct_n": 8, + "max": 3.099204835224729, + "min": -1.621627553120839, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.75124315545996, + "action_delta": { + "distinct_n": 6, + "max": -3.0894547775551757, + "min": -44.89764761415685, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.2665496795389639, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 36.58321095086717, + "repeat_abs_p95": 57.539092487805476, + "repeat_delta": { + "distinct_n": 8, + "max": 63.80774773212434, + "min": -40.23384050335905, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.39910956222742744, + "action_delta": { + "distinct_n": 6, + "max": -0.15660557052023627, + "min": -2.2524677264815143, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.2818240506833578, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.4161657291479544, + "repeat_abs_p95": 2.6502790089519457, + "repeat_delta": { + "distinct_n": 8, + "max": 3.0729743073386615, + "min": -1.6090373367280613, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.012606136240383592, + "action_delta": { + "distinct_n": 6, + "max": 0.024621184785722505, + "min": 0.0006042607986376236, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.1046254041741963, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.011412136813753415, + "repeat_abs_p95": 0.020065382031526628, + "repeat_delta": { + "distinct_n": 8, + "max": 0.012164469199423045, + "min": -0.020447806270125324, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.019963170721776176, + "action_delta": { + "distinct_n": 6, + "max": -0.0043728510978748095, + "min": -0.036507530380663095, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 1.8698607257091873, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.010676287515587392, + "repeat_abs_p95": 0.017837185208830555, + "repeat_delta": { + "distinct_n": 8, + "max": 0.01883615600387016, + "min": -0.012662463335994696, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.004998160876698038, + "action_delta": { + "distinct_n": 6, + "max": 0.009148827020083142, + "min": 0.0008846012466042618, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.841347922729203, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.0008556519732799599, + "repeat_abs_p95": 0.002114872464387871, + "repeat_delta": { + "distinct_n": 8, + "max": 0.002067321151258763, + "min": -0.0021404770176112367, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00203851584055581, + "action_delta": { + "distinct_n": 6, + "max": 0.003668730096473505, + "min": -0.0029587377234424617, + "n": 6 + }, + "action_signs": { + "consistency": 0.5, + "negative": 3, + "positive": 3, + "zero": 0 + }, + "effect_to_repeat_median": 0.689091144297968, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0029582673604557908, + "repeat_abs_p95": 0.008930052912586221, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0019231063523722502, + "min": -0.009814473798313572, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 0.010662801325641125, + "action_delta": { + "distinct_n": 6, + "max": 0.021920351504425173, + "min": 0.003853870265263404, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.783811852453453, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0018435595067149424, + "repeat_abs_p95": 0.006457102557288577, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0035006848054948136, + "min": -0.008049019808254454, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00020237147625482444, + "action_delta": { + "distinct_n": 6, + "max": 0.0021466714075248396, + "min": -0.0002787530429213137, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.1687685271313482, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0011991067273895441, + "repeat_abs_p95": 0.0023351857861895597, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0027611025339853053, + "min": -0.001368780132029305, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00013454220935188443, + "action_delta": { + "distinct_n": 6, + "max": 0.00015029680675249235, + "min": -0.0004908425008621764, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.0774352591631262, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0017374799388022444, + "repeat_abs_p95": 0.0025224968780584887, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0025533385639567863, + "min": -0.0024652194613902223, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.4082582890066675, + "action_delta": { + "distinct_n": 6, + "max": 1.7869539801733332, + "min": -0.46609635046666753, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 4, + "positive": 2, + "zero": 0 + }, + "effect_to_repeat_median": 0.38701557093710687, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 1.0548885359266662, + "repeat_abs_p95": 2.4461823174926662, + "repeat_delta": { + "distinct_n": 8, + "max": 2.9756512754266673, + "min": -1.3536673999333333, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 2.2263649953666667, + "action_delta": { + "distinct_n": 6, + "max": -0.10833277018666666, + "min": -32.03782532756, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 88.19287762590257, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.025244272046666672, + "repeat_abs_p95": 15.854935623909324, + "repeat_delta": { + "distinct_n": 8, + "max": 20.359169682466664, + "min": -0.11883915477333334, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 6.5, + "action_delta": { + "distinct_n": 6, + "max": 11.733333333333334, + "min": 3.173333333333332, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 0.5247578040904196, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 12.38666666666667, + "repeat_abs_p95": 17.40266666666666, + "repeat_delta": { + "distinct_n": 8, + "max": 15.599999999999994, + "min": -18.373333333333335, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 75.0, + "trial_sanity": [ + { + "admitted_fraction": 0.47555555555555556, + "cell": "tp4_mns16", + "completed_fraction": 0.4733333333333334, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.001282354, + "max_internal_gap_s": 0.471713815, + "start_gap_s": 0.002667743 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.4752941176470588, + "cell": "tp4_mns16", + "completed_fraction": 0.47058823529411764, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004049563, + "max_internal_gap_s": 0.179027071, + "start_gap_s": 0.035944018 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.47555555555555556, + "cell": "tp4_mns64", + "completed_fraction": 0.4733333333333334, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003246004, + "max_internal_gap_s": 0.48099535, + "start_gap_s": 0.000371266 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.4752941176470588, + "cell": "tp4_mns64", + "completed_fraction": 0.47098039215686277, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00327497, + "max_internal_gap_s": 0.178353557, + "start_gap_s": 0.002613787 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.49444444444444446, + "cell": "tp4_mns64", + "completed_fraction": 0.49111111111111116, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.003493587, + "max_internal_gap_s": 0.444552932, + "start_gap_s": 0.00122792 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.49058823529411766, + "cell": "tp4_mns64", + "completed_fraction": 0.48588235294117643, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.0063142, + "max_internal_gap_s": 0.183290846, + "start_gap_s": 0.062765702 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.49444444444444446, + "cell": "tp4_mns16", + "completed_fraction": 0.49111111111111116, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.012013597, + "max_internal_gap_s": 0.445602102, + "start_gap_s": 0.037343372 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.49058823529411766, + "cell": "tp4_mns16", + "completed_fraction": 0.463921568627451, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.02999801, + "max_internal_gap_s": 0.176673609, + "start_gap_s": 0.001405349 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.4711111111111111, + "cell": "tp4_mns16", + "completed_fraction": 0.4655555555555555, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.026096497, + "max_internal_gap_s": 0.24359672, + "start_gap_s": 0.004691362 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.4823529411764706, + "cell": "tp4_mns16", + "completed_fraction": 0.4549019607843137, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001098192, + "max_internal_gap_s": 0.177840927, + "start_gap_s": 0.00439926 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.4711111111111111, + "cell": "tp4_mns64", + "completed_fraction": 0.4661111111111111, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001270441, + "max_internal_gap_s": 0.272099734, + "start_gap_s": 0.006329393 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.4823529411764706, + "cell": "tp4_mns64", + "completed_fraction": 0.4713725490196079, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.048916774, + "max_internal_gap_s": 0.184986457, + "start_gap_s": 0.001848903 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.13955613873224326, + "tpot_mean_over_slo": 0.00021561369396999597, + "ttft_max_over_slo_max": -0.162763388332678, + "ttft_mean_over_slo_max": -0.0113370356308976 + }, + "delta_state": { + "batch_size.mean": -0.2888291061573778, + "batch_tokens.mean": -6.113142522726264, + "decode_batch_size.mean": -0.28726153067008475, + "graph_full_share": 0.003556405337845958, + "graph_none_share": -0.008054518886335793, + "graph_padding_fraction": 0.003354803792462677, + "kv_usage_end_minus_start": -7.914574800538254e-06, + "kv_usage_max": 0.012148644893025695, + "kv_usage_mean": 0.0003717674926776663, + "preemptions": 0.0, + "prefill_token_fraction": -1.0418331766737232e-05, + "queue_running_mean": 0.18344752267999986, + "queue_waiting_mean": -0.6748495923466666, + "scheduler_steps_per_s": 3.9866666666666646 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.14980392156862746, + "completed_over_admitted": 0.03292618162506633, + "completed_pass_rate": 0.21093318608503586, + "outstanding_over_admitted": -0.032926181625066386, + "tpot_max_over_slo": 0.3362407738628878, + "tpot_mean_over_slo": 0.035823037105198796, + "ttft_max_over_slo_max": -1.0155032691739811, + "ttft_mean_over_slo_max": -0.23418190946398737 + }, + "delta_state": { + "batch_size.mean": 0.7340835489721993, + "batch_tokens.mean": 4.870249914593785, + "decode_batch_size.mean": 0.7235214534720207, + "graph_full_share": 0.01432646120104708, + "graph_none_share": -0.025264419492621093, + "graph_padding_fraction": 0.006957260165806216, + "kv_usage_end_minus_start": -0.007513472509556363, + "kv_usage_max": 0.031275425686576486, + "kv_usage_mean": 0.0049451651951918, + "preemptions": 0.0, + "prefill_token_fraction": -0.0014595808679910283, + "queue_running_mean": 5.284704507639999, + "queue_waiting_mean": -29.284492367786665, + "scheduler_steps_per_s": 3.373333333333335 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0007401924500369805, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.00074019245003701, + "tpot_max_over_slo": 0.17641319103142994, + "tpot_mean_over_slo": -0.008167534216044714, + "ttft_max_over_slo_max": -0.055667730843803526, + "ttft_mean_over_slo_max": -0.004452474842343956 + }, + "delta_state": { + "batch_size.mean": -0.4169380613868432, + "batch_tokens.mean": -8.458043870097356, + "decode_batch_size.mean": -0.4135692006454068, + "graph_full_share": 0.003382014563382052, + "graph_none_share": -0.01001403055152162, + "graph_padding_fraction": 0.0016407136927011953, + "kv_usage_end_minus_start": -0.0014920741368527057, + "kv_usage_max": 0.013006718003583773, + "kv_usage_mean": -4.9942287421177005e-05, + "preemptions": 0.0, + "prefill_token_fraction": -4.5771229986280026e-05, + "queue_running_mean": -0.27537858206666677, + "queue_waiting_mean": -0.14998382319999998, + "scheduler_steps_per_s": 6.626666666666665 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.3898039215686275, + "completed_over_admitted": 0.047995780590717296, + "completed_pass_rate": 0.5534521158129175, + "outstanding_over_admitted": -0.0479957805907173, + "tpot_max_over_slo": 0.485827943940801, + "tpot_mean_over_slo": 0.015614815582012831, + "ttft_max_over_slo_max": -0.9946015174985708, + "ttft_mean_over_slo_max": -0.5387282368223152 + }, + "delta_state": { + "batch_size.mean": -1.6273974153257296, + "batch_tokens.mean": -34.29456683018134, + "decode_batch_size.mean": -1.6204010215371003, + "graph_full_share": 0.02986827684967852, + "graph_none_share": -0.04355833918802347, + "graph_padding_fraction": 0.009679698181700893, + "kv_usage_end_minus_start": -0.0096678119183089, + "kv_usage_max": 0.04533198050309872, + "kv_usage_mean": 0.003191336087705077, + "preemptions": 0.0, + "prefill_token_fraction": -0.00023339635256569569, + "queue_running_mean": 2.3989840409999985, + "queue_waiting_mean": -47.09879207850667, + "scheduler_steps_per_s": 11.493333333333325 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.06666873637623494, + "tpot_mean_over_slo": -0.009318131681115394, + "ttft_max_over_slo_max": -0.07750733549861857, + "ttft_mean_over_slo_max": -0.004705775197540927 + }, + "delta_state": { + "batch_size.mean": -0.5672759530887257, + "batch_tokens.mean": -10.329959059848179, + "decode_batch_size.mean": -0.562993756337625, + "graph_full_share": 0.005089834536355586, + "graph_none_share": -0.01044980089531864, + "graph_padding_fraction": 0.0013844048532868167, + "kv_usage_end_minus_start": 0.002908292329360629, + "kv_usage_max": 0.00626947513206999, + "kv_usage_mean": -0.0002826822094093009, + "preemptions": 0.0, + "prefill_token_fraction": 0.00012710629028711562, + "queue_running_mean": -0.5141460273999989, + "queue_waiting_mean": -0.24953442832, + "scheduler_steps_per_s": 8.97333333333333 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.43333333333333335, + "completed_over_admitted": 0.05807814149947199, + "completed_pass_rate": 0.6218345526167699, + "outstanding_over_admitted": -0.05807814149947202, + "tpot_max_over_slo": 0.30513300693329926, + "tpot_mean_over_slo": 0.006471198446666382, + "ttft_max_over_slo_max": -0.9754032485070638, + "ttft_mean_over_slo_max": -0.5503063221993506 + }, + "delta_state": { + "batch_size.mean": -0.33070896537783234, + "batch_tokens.mean": -13.234057192010994, + "decode_batch_size.mean": -0.33092189282726103, + "graph_full_share": 0.024782576956253966, + "graph_none_share": -0.03529189804944581, + "graph_padding_fraction": 0.00864361242258734, + "kv_usage_end_minus_start": -0.004863029113173378, + "kv_usage_max": 0.02549546289093474, + "kv_usage_mean": 0.0018655214057756511, + "preemptions": 0.0, + "prefill_token_fraction": -0.0010506631885456308, + "queue_running_mean": 2.228850923693333, + "queue_waiting_mean": -47.28131520793333, + "scheduler_steps_per_s": 7.439999999999998 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 6, + "max": 0.7505555555555555, + "min": 0.7316666666666667, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 9, + "max": 0.7472222222222221, + "min": 0.6968627450980392, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_max_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.19490196078431374, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.07490196078431373, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.07490196078431373, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.02436798652037714, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.016463090812533165, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.016833187037551656, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.27672605790645877, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10546659304251793, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.10546659304251793, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.024367986520377156, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.016463090812533193, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.016833187037551697, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2407730989823646, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.22234457283277126, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2563269824471589, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.6666666666666666, + "balanced_accuracy": 0.6666666666666666, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.000848167884689166, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003343406070318189, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.007915214637991413, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5264552920028412, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5690833184198709, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5786824529156245, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.27171700600992804, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.12275947254744249, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.12275947254744249, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": -0.3722455467363339, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.21812996140096796, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.21812996140096796, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.005142163057644268, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.005156031979134446, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.005156031979134446, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0009077895591772371, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0011186444492266587, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0017815517901913717, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00013958379127598786, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.000530540760156184, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.00013958379127598786, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.9767361708133331, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 1.2061492231866664, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 1.2912157818399992, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -23.674163253413333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -14.979670980066667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -14.979670980066667, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 7.033333333333331, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.6799999999999997, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.6799999999999997, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 0.75, + "end_s": 225.0, + "qualifying_response_features": [ + "graph_padding_fraction", + "queue_waiting_mean" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.49210700723778444, + "action_delta": { + "distinct_n": 6, + "max": 0.7340835489721993, + "min": -1.6273974153257296, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 1.187549092220796, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.41438876966131266, + "repeat_abs_p95": 1.7988231433741386, + "repeat_delta": { + "distinct_n": 8, + "max": 1.2969259412104588, + "min": -2.069075483000736, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 9.394001464972767, + "action_delta": { + "distinct_n": 6, + "max": 4.870249914593785, + "min": -34.29456683018134, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.5103708522227296, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 18.40622642155347, + "repeat_abs_p95": 44.616864312796594, + "repeat_delta": { + "distinct_n": 8, + "max": 20.29593930708012, + "min": -57.34701495152672, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.48828147849151593, + "action_delta": { + "distinct_n": 6, + "max": 0.7235214534720207, + "min": -1.6204010215371003, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 1.1843674420861072, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.41227195306168873, + "repeat_abs_p95": 1.7851097930098754, + "repeat_delta": { + "distinct_n": 8, + "max": 1.2901539913069193, + "min": -2.051624455465314, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.009708147868701333, + "action_delta": { + "distinct_n": 6, + "max": 0.02986827684967852, + "min": 0.003382014563382052, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 2.9605949673769345, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0032791205739644558, + "repeat_abs_p95": 0.008038878324209003, + "repeat_delta": { + "distinct_n": 8, + "max": 0.008664142823959997, + "min": -0.006877672824671444, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.017857110193969867, + "action_delta": { + "distinct_n": 6, + "max": -0.008054518886335793, + "min": -0.04355833918802347, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 4.464300302534616, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.003999979612444856, + "repeat_abs_p95": 0.010649544454368921, + "repeat_delta": { + "distinct_n": 8, + "max": 0.008444852161820365, + "min": -0.011836686458048917, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 0.005156031979134446, + "action_delta": { + "distinct_n": 6, + "max": 0.009679698181700893, + "min": 0.0013844048532868167, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 5.791648307818219, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.0008902529478826008, + "repeat_abs_p95": 0.0032904967132956443, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0038875531917088053, + "min": -0.000986602698250582, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.0038856607212670036, + "action_delta": { + "distinct_n": 6, + "max": 0.002908292329360629, + "min": -0.0096678119183089, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 2.2991048875374145, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0016900754473315738, + "repeat_abs_p95": 0.006128076218419801, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0017580081358981037, + "min": -0.007809750200217236, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.019251090447259256, + "action_delta": { + "distinct_n": 6, + "max": 0.04533198050309872, + "min": 0.00626947513206999, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 7.386921150773906, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.002606104770083051, + "repeat_abs_p95": 0.016512755779962775, + "repeat_delta": { + "distinct_n": 8, + "max": 0.012755779962795644, + "min": -0.01853574275843739, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 0.0011186444492266587, + "action_delta": { + "distinct_n": 6, + "max": 0.0049451651951918, + "min": -0.0002826822094093009, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.9840910262940343, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.001136728635194791, + "repeat_abs_p95": 0.0026446247991943482, + "repeat_delta": { + "distinct_n": 8, + "max": 0.00023211260883475478, + "min": -0.0029335843047816335, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.00018025132142640565, + "action_delta": { + "distinct_n": 6, + "max": 0.00012710629028711562, + "min": -0.0014595808679910283, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 0.07514942281617999, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0023985722667133613, + "repeat_abs_p95": 0.003432972436972548, + "repeat_delta": { + "distinct_n": 8, + "max": -0.00011986074750158782, + "min": -0.0035071769437554634, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.3333333333333333, + "action_abs_median": 1.371498475546666, + "action_delta": { + "distinct_n": 6, + "max": 5.284704507639999, + "min": -0.5141460273999989, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 5.458580798787922, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.2512555050666663, + "repeat_abs_p95": 2.2674077304893334, + "repeat_delta": { + "distinct_n": 8, + "max": 0.25758021949333276, + "min": -2.6281402471466677, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 14.979670980066667, + "action_delta": { + "distinct_n": 6, + "max": -0.14998382319999998, + "min": -47.28131520793333, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 263.3579940744849, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.05687949983333334, + "repeat_abs_p95": 11.769436231303327, + "repeat_delta": { + "distinct_n": 8, + "max": 17.82422918966667, + "min": -0.5248207372, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 7.033333333333331, + "action_delta": { + "distinct_n": 6, + "max": 11.493333333333325, + "min": 3.373333333333335, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 1.8158347676419953, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 3.873333333333335, + "repeat_abs_p95": 7.130000000000002, + "repeat_delta": { + "distinct_n": 8, + "max": 7.200000000000003, + "min": -3.3866666666666703, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 150.0, + "trial_sanity": [ + { + "admitted_fraction": 0.7388888888888889, + "cell": "tp4_mns16", + "completed_fraction": 0.7377777777777779, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.002047043, + "max_internal_gap_s": 0.302592239, + "start_gap_s": 0.003297384 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7384313725490196, + "cell": "tp4_mns16", + "completed_fraction": 0.7101960784313726, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.000359161, + "max_internal_gap_s": 0.188948337, + "start_gap_s": 0.002535254 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7388888888888889, + "cell": "tp4_mns64", + "completed_fraction": 0.7377777777777779, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.003650633, + "max_internal_gap_s": 0.322890476, + "start_gap_s": 0.001316118 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7384313725490196, + "cell": "tp4_mns64", + "completed_fraction": 0.7345098039215686, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.004944353, + "max_internal_gap_s": 0.219527251, + "start_gap_s": 0.003464116 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.7505555555555555, + "cell": "tp4_mns64", + "completed_fraction": 0.7472222222222221, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002633243, + "max_internal_gap_s": 0.646724237, + "start_gap_s": 0.001383425 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7435294117647059, + "cell": "tp4_mns64", + "completed_fraction": 0.74, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.000558705, + "max_internal_gap_s": 0.183301269, + "start_gap_s": 0.161925118 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 0.7505555555555555, + "cell": "tp4_mns16", + "completed_fraction": 0.7466666666666666, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.001410558, + "max_internal_gap_s": 0.583053185, + "start_gap_s": 0.155434911 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7435294117647059, + "cell": "tp4_mns16", + "completed_fraction": 0.7043137254901961, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.042708099, + "max_internal_gap_s": 0.179974019, + "start_gap_s": 0.02943376 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7316666666666667, + "cell": "tp4_mns16", + "completed_fraction": 0.7311111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.000142513, + "max_internal_gap_s": 0.318659499, + "start_gap_s": 0.034846458 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 0.7427450980392157, + "cell": "tp4_mns16", + "completed_fraction": 0.6968627450980392, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.049578949, + "max_internal_gap_s": 0.182904479, + "start_gap_s": 0.084632833 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 0.7316666666666667, + "cell": "tp4_mns64", + "completed_fraction": 0.7311111111111112, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.00141287, + "max_internal_gap_s": 0.456938178, + "start_gap_s": 0.003994197 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 0.7427450980392157, + "cell": "tp4_mns64", + "completed_fraction": 0.74, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.009657729, + "max_internal_gap_s": 0.179710886, + "start_gap_s": 0.001797426 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + }, + { + "action_pairs": 6, + "actions": [ + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0011111111111111738, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0011111111111111113, + "tpot_max_over_slo": 0.13955613873224326, + "tpot_mean_over_slo": -0.0010090653746841205, + "ttft_max_over_slo_max": -0.162763388332678, + "ttft_mean_over_slo_max": -0.00961120104762887 + }, + "delta_state": { + "batch_size.mean": -0.36880112566985623, + "batch_tokens.mean": -6.920339794120963, + "decode_batch_size.mean": -0.3657101237877507, + "graph_full_share": 0.002401511671261014, + "graph_none_share": -0.007647132972108277, + "graph_padding_fraction": 0.0013086971950763866, + "kv_usage_end_minus_start": 0.0006854718445343799, + "kv_usage_max": 0.008271873870579816, + "kv_usage_mean": -0.00014200125936815727, + "preemptions": 0.0, + "prefill_token_fraction": 0.00022380184306991513, + "queue_running_mean": -0.2263784608933328, + "queue_waiting_mean": -0.11799772912, + "scheduler_steps_per_s": 5.7733333333333405 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 1, + "request_hash": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "5856cc9ff24c9a984a2abecc59ab92c1f87b02b2a9dc0a47eae7a08462169373", + "result_sha256": "4e788371d6f24cab1be2bd4aed4bfe0354b5a5c23973bac9bf69ad0aa4c1b7d9", + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "a91536c628053d3cf530bf0b8ebe51628cc41e71af31b90310d3307d1badc473", + "result_sha256": "2dd1d5ae610db6d738291032ab07a3ad73c46235b30e4ff38b87ce2715204bc6", + "trial_id": "sessions/rep1-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.3949019607843137, + "completed_over_admitted": 0.03529411764705881, + "completed_pass_rate": 0.41338259441707714, + "outstanding_over_admitted": -0.03529411764705882, + "tpot_max_over_slo": 0.3362407738628878, + "tpot_mean_over_slo": 0.031421107704212525, + "ttft_max_over_slo_max": -1.0149181675127088, + "ttft_mean_over_slo_max": -0.4177688425802161 + }, + "delta_state": { + "batch_size.mean": -0.025927986090335864, + "batch_tokens.mean": 18.396156394293996, + "decode_batch_size.mean": -0.03259251609001801, + "graph_full_share": 0.02053093742241252, + "graph_none_share": -0.032184416438265945, + "graph_padding_fraction": 0.008177832545205635, + "kv_usage_end_minus_start": 0.01748278407834214, + "kv_usage_max": 0.016581905042334366, + "kv_usage_mean": 0.002461661215058867, + "preemptions": 0.0, + "prefill_token_fraction": 0.0028536948076697266, + "queue_running_mean": 1.9724076930533343, + "queue_waiting_mean": -47.152675258386665, + "scheduler_steps_per_s": 3.3066666666666578 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 1, + "request_hash": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.5603921568627451, + "mns": 16, + "requests_sha256": "8dcab1a24d009309a6dd54676e762ee4e74717ed530d19b3965f30feba32f193", + "result_sha256": "ae7cbca9785fa548acf8bd82646dfe53694ffd0b58d2ef6953e3d531d4d7ec4f", + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e7bf606cb77817e67a3e64e60f035a5fb344977a7991be741a81db8b3a8d28fe", + "result_sha256": "9604c8a0fe449d3798c6de566e4c804665d66b859a3b43c716dd5ecbb60bee1c", + "trial_id": "sessions/rep1-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": 0.0, + "tpot_max_over_slo": 0.17641319103142994, + "tpot_mean_over_slo": -0.004863679616933292, + "ttft_max_over_slo_max": -0.16410399084755528, + "ttft_mean_over_slo_max": -0.005948038735831104 + }, + "delta_state": { + "batch_size.mean": -0.21169295708183622, + "batch_tokens.mean": -4.472135801297441, + "decode_batch_size.mean": -0.21006339951083586, + "graph_full_share": 0.001298043937749238, + "graph_none_share": -0.004668542387939491, + "graph_padding_fraction": 0.0013031829071826536, + "kv_usage_end_minus_start": 0.0014933728423490722, + "kv_usage_max": 0.01910758793466949, + "kv_usage_mean": 0.0002283394725866271, + "preemptions": 0.0, + "prefill_token_fraction": 4.923759477581946e-05, + "queue_running_mean": 0.18394671514666605, + "queue_waiting_mean": -0.37382507898666667, + "scheduler_steps_per_s": 3.4399999999999977 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 2, + "request_hash": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "a5ce6cd64890663ef30486701236af3de6169f86b72349928b07adafcb406bcb", + "result_sha256": "289f8f99f5056b26822bab5c1feb4252042c7852b80c5399271ddb1b6db25fbb", + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "1c6931ea4a0913e2460e18c59f5a07e9c21cb2650454f2dd2f2997c416f63209", + "result_sha256": "9988f2b6c9e8c45144eb7b53aa1bea62942ae8fd36aba10c9e6bf5c8cbd65a50", + "trial_id": "sessions/rep2-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.6321568627450981, + "completed_over_admitted": 0.04980392156862745, + "completed_pass_rate": 0.667771333885667, + "outstanding_over_admitted": -0.04980392156862745, + "tpot_max_over_slo": 0.485827943940801, + "tpot_mean_over_slo": 0.017498560395543927, + "ttft_max_over_slo_max": -0.9946015174985708, + "ttft_mean_over_slo_max": -0.6463312081620042 + }, + "delta_state": { + "batch_size.mean": -0.5058158071040495, + "batch_tokens.mean": -7.22996895580593, + "decode_batch_size.mean": -0.5073220598343564, + "graph_full_share": 0.026735082314620406, + "graph_none_share": -0.0367740255019158, + "graph_padding_fraction": 0.008672030283845366, + "kv_usage_end_minus_start": 0.008358614300674416, + "kv_usage_max": 0.02410580556943176, + "kv_usage_mean": 0.002518308420146869, + "preemptions": 0.0, + "prefill_token_fraction": 0.0005037242486639615, + "queue_running_mean": 2.4556977948933323, + "queue_waiting_mean": -47.22722224089333, + "scheduler_steps_per_s": 6.413333333333327 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 2, + "request_hash": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.31450980392156863, + "mns": 16, + "requests_sha256": "e00c0d9acfb0c22624fdde79f4e8a0300fc2aba09ad4851908d68c1f46be30d3", + "result_sha256": "d081e1ac0143b4e8a63f4c60b58905320ca6e6163fa858e78f5643221fd87c2b", + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "e1543f873745dbd1d3a1461e63549747b6e1517d63e281e6cdfa974d2e2798e4", + "result_sha256": "8556783dc3ba3274269500add337c75c493e3eece4f66260449549ee86506035", + "trial_id": "sessions/rep2-mns64/mid/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": 0.0, + "completed_over_admitted": 0.0005555555555556424, + "completed_pass_rate": 0.0, + "outstanding_over_admitted": -0.0005555555555555552, + "tpot_max_over_slo": 0.18505735732578327, + "tpot_mean_over_slo": -0.007144351590753079, + "ttft_max_over_slo_max": -0.12056507783321042, + "ttft_mean_over_slo_max": -0.006789902099118429 + }, + "delta_state": { + "batch_size.mean": -0.36923651780535494, + "batch_tokens.mean": -7.266086823190449, + "decode_batch_size.mean": -0.3664963060152342, + "graph_full_share": 0.0022555226793449235, + "graph_none_share": -0.007954072283154097, + "graph_padding_fraction": 0.003323634208537993, + "kv_usage_end_minus_start": -4.647472024943422e-05, + "kv_usage_max": 0.01645626968867886, + "kv_usage_mean": 0.00016819101878902745, + "preemptions": 0.0, + "prefill_token_fraction": 3.148363682092814e-06, + "queue_running_mean": -0.07117231964000048, + "queue_waiting_mean": -0.44794735680000003, + "scheduler_steps_per_s": 5.546666666666667 + }, + "full_action_efficacy": 0, + "full_feasibility_transition": "true->true", + "group": { + "level": "low", + "offered_rate_per_gpu": 1.5, + "replicate": 3, + "request_hash": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 16, + "requests_sha256": "de93d4364969393741a920ea72831b69003cbd9f10bcfd1809848fcf4e948549", + "result_sha256": "3b32b5ff3aea07cb7b7cf252ba23852542cb8429c83123fe772b2c8093c408a6", + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "3623a99013725d7b42845bce09678368b3dc8d82e0ab26fad549c7eab8a6b6fe", + "result_sha256": "77766d9ec36ff11893feb73626768fa593160394f2b4720ff9253d1dfd5fa950", + "trial_id": "sessions/rep3-mns64/low/result.json" + } + }, + { + "delta_outcome": { + "admitted_fraction": 0.0, + "admitted_input_tokens_mean_over_limit": 0.0, + "completed_fail_fraction_of_total": -0.6756862745098039, + "completed_over_admitted": 0.05372549019607842, + "completed_pass_rate": 0.7194154488517746, + "outstanding_over_admitted": -0.05372549019607843, + "tpot_max_over_slo": 0.37831041669495047, + "tpot_mean_over_slo": 0.007493808043833983, + "ttft_max_over_slo_max": -0.9795807876701778, + "ttft_mean_over_slo_max": -0.6540993991122069 + }, + "delta_state": { + "batch_size.mean": -1.6247957869409735, + "batch_tokens.mean": -35.238499877789366, + "decode_batch_size.mean": -1.6176751208671583, + "graph_full_share": 0.029945447264822622, + "graph_none_share": -0.04283615284473359, + "graph_padding_fraction": 0.010093615119271773, + "kv_usage_end_minus_start": -0.0001447507354654931, + "kv_usage_max": 0.03542758976690741, + "kv_usage_mean": 0.0020665348540741674, + "preemptions": 0.0, + "prefill_token_fraction": -0.0004446109784110863, + "queue_running_mean": 1.6193733474533332, + "queue_waiting_mean": -47.261676539759996, + "scheduler_steps_per_s": 11.693333333333328 + }, + "full_action_efficacy": 1, + "full_feasibility_transition": "false->true", + "group": { + "level": "mid", + "offered_rate_per_gpu": 2.125, + "replicate": 3, + "request_hash": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "tp": 4 + }, + "source": { + "cell": "tp4_mns16", + "early_stopped": false, + "full_feasible": false, + "full_pass_rate": 0.2635294117647059, + "mns": 16, + "requests_sha256": "c7fca9b2484c8ea68c6c11438b8e59db74fa60e6189dbb81d8834038156dbf68", + "result_sha256": "bb45674fa5e19ef18d8c13d08493f1a08a62fb682b5d5e8b7d8a1db2fa3b2ea6", + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + "target": { + "cell": "tp4_mns64", + "early_stopped": false, + "full_feasible": true, + "full_pass_rate": 1.0, + "mns": 64, + "requests_sha256": "d75e1711b8c0485cf1aceaa9e26ae079fa2adc5487d37f28006c01e83c5a782b", + "result_sha256": "1654b044d815beb3a5b8913fde4283b11507464a2de9f41b5ddc3cdb9f2d150b", + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + } + ], + "coverage_at_end": { + "admitted_fraction_of_total": { + "distinct_n": 1, + "max": 1.0, + "min": 1.0, + "n": 12 + }, + "completed_fraction_of_total": { + "distinct_n": 10, + "max": 0.9966666666666667, + "min": 0.9392156862745098, + "n": 12 + } + }, + "efficacy": { + "cross_validation_possible": true, + "label_balance_sufficient": true, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "outcome_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "ttft_mean_over_slo_max", + "features": { + "admitted_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "admitted_input_tokens_mean_over_limit": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 1 + ], + "threshold": 0.0, + "train_balanced_accuracy": 0.5 + } + ] + }, + "completed_fail_fraction_of_total": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.31607843137254904, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.19745098039215686, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.19745098039215686, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.025179738562091547, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.01820261437908499, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.01820261437908499, + "train_balanced_accuracy": 1.0 + } + ] + }, + "completed_pass_rate": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.3338856669428335, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.20669129720853857, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.20669129720853857, + "train_balanced_accuracy": 1.0 + } + ] + }, + "outstanding_over_admitted": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.025179738562091502, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.018202614379084967, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.018202614379084967, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_max_over_slo": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.28168388701036684, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.26064906559433554, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.2563269824471589, + "train_balanced_accuracy": 1.0 + } + ] + }, + "tpot_mean_over_slo": { + "accuracy": 0.8333333333333334, + "balanced_accuracy": 0.8333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0013150642134503454, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.003242371334574931, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.008244747510429903, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_max_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5718423892588665, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5711720880014279, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.5793527541730631, + "train_balanced_accuracy": 1.0 + } + ] + }, + "ttft_mean_over_slo_max": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.3265605551305613, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.2136900218139225, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -0.2136900218139225, + "train_balanced_accuracy": 1.0 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "positive": 3, + "telemetry_delta": { + "best_accuracy": 1.0, + "best_balanced_accuracy": 1.0, + "best_feature": "queue_waiting_mean", + "features": { + "decode_batch_size.mean": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.4369091829247953, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.19915131993888435, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.12132795780042693, + "train_balanced_accuracy": 0.75 + } + ] + }, + "graph_padding_fraction": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.00599783224619168, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.005750733376871814, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.004743264870141011, + "train_balanced_accuracy": 1.0 + } + ] + }, + "kv_usage_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0011474371633303972, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.0011173629364315974, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.001345000343822747, + "train_balanced_accuracy": 1.0 + } + ] + }, + "prefill_token_fraction": { + "accuracy": 0.5, + "balanced_accuracy": 0.5, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.00022073130736449675, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": -0.00022073130736449675, + "train_balanced_accuracy": 0.75 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 0.0003637630458669383, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_running_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.9016600312999996, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 0.7741005139066663, + "train_balanced_accuracy": 1.0 + }, + { + "direction": 1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": 1.0781772041000002, + "train_balanced_accuracy": 1.0 + } + ] + }, + "queue_waiting_mean": { + "accuracy": 1.0, + "balanced_accuracy": 1.0, + "folds": [ + { + "direction": -1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -23.837584798846667, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -23.800311307593333, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 1 + ], + "threshold": -23.763250168686668, + "train_balanced_accuracy": 1.0 + } + ] + }, + "scheduler_steps_per_s": { + "accuracy": 0.3333333333333333, + "balanced_accuracy": 0.3333333333333333, + "folds": [ + { + "direction": 1, + "held_out_replicate": 1, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 5.979999999999997, + "train_balanced_accuracy": 1.0 + }, + { + "direction": -1, + "held_out_replicate": 2, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 1, + 0 + ], + "threshold": 4.426666666666662, + "train_balanced_accuracy": 0.75 + }, + { + "direction": -1, + "held_out_replicate": 3, + "test_labels": [ + 0, + 1 + ], + "test_predictions": [ + 0, + 0 + ], + "threshold": 3.3733333333333277, + "train_balanced_accuracy": 0.75 + } + ] + } + }, + "labels": { + "distinct_n": 2, + "max": 1.0, + "min": 0.0, + "n": 6 + }, + "negative": 3, + "positive": 3 + }, + "telemetry_qualifying_features": [], + "transitions": { + "false->true": 3, + "true->true": 3 + } + }, + "end_fraction": 1.0, + "end_s": 300.0, + "qualifying_response_features": [ + "graph_padding_fraction", + "queue_waiting_mean", + "scheduler_steps_per_s" + ], + "repeat_pairs": 8, + "response_statistics": { + "batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.3690188217376056, + "action_delta": { + "distinct_n": 6, + "max": -0.025927986090335864, + "min": -1.6247957869409735, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.8618355767528564, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.4281777541929288, + "repeat_abs_p95": 0.9733377249226858, + "repeat_delta": { + "distinct_n": 8, + "max": 0.7022084961695088, + "min": -1.119330386559012, + "n": 8 + } + }, + "batch_tokens.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 7.248027889498189, + "action_delta": { + "distinct_n": 6, + "max": 18.396156394293996, + "min": -35.238499877789366, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 5, + "positive": 1, + "zero": 0 + }, + "effect_to_repeat_median": 1.0330903781421452, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 7.015870095056599, + "repeat_abs_p95": 31.643382433390233, + "repeat_delta": { + "distinct_n": 8, + "max": 15.47659343283351, + "min": -40.3485765106131, + "n": 8 + } + }, + "decode_batch_size.mean": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.36610321490149245, + "action_delta": { + "distinct_n": 6, + "max": -0.03259251609001801, + "min": -1.6176751208671583, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 0.8640718152580315, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.4236953554516365, + "repeat_abs_p95": 0.965912412061184, + "repeat_delta": { + "distinct_n": 8, + "max": 0.6959888014852451, + "min": -1.1112558946789974, + "n": 8 + } + }, + "graph_full_share": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.011466224546836767, + "action_delta": { + "distinct_n": 6, + "max": 0.029945447264822622, + "min": 0.001298043937749238, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 2.7129568200282717, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.004226467764686825, + "repeat_abs_p95": 0.0058786421963618, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005744029111679261, + "min": -0.005951126165037013, + "n": 8 + } + }, + "graph_none_share": { + "action_above_repeat_p95_fraction": 0.8333333333333334, + "action_abs_median": 0.02006924436071002, + "action_delta": { + "distinct_n": 6, + "max": -0.004668542387939491, + "min": -0.04283615284473359, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 6.0905462240848625, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0032951468755539302, + "repeat_abs_p95": 0.005637894418237397, + "repeat_delta": { + "distinct_n": 8, + "max": 0.005589923520491251, + "min": -0.005663724901639168, + "n": 8 + } + }, + "graph_padding_fraction": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 0.005750733376871814, + "action_delta": { + "distinct_n": 6, + "max": 0.010093615119271773, + "min": 0.0013031829071826536, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 10.921714369329514, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.000526541272038856, + "repeat_abs_p95": 0.002367923136830107, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0028531471739462067, + "min": -0.0004812194431302371, + "n": 8 + } + }, + "kv_usage_end_minus_start": { + "action_above_repeat_p95_fraction": 0.16666666666666666, + "action_abs_median": 0.001089422343441726, + "action_delta": { + "distinct_n": 6, + "max": 0.01748278407834214, + "min": -0.0001447507354654931, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 0.1436227894695951, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.007585302774476166, + "repeat_abs_p95": 0.014011837897733656, + "repeat_delta": { + "distinct_n": 8, + "max": 0.008574911878879932, + "min": -0.015719863447740123, + "n": 8 + } + }, + "kv_usage_max": { + "action_above_repeat_p95_fraction": 0.8333333333333334, + "action_abs_median": 0.017844746488501928, + "action_delta": { + "distinct_n": 6, + "max": 0.03542758976690741, + "min": 0.008271873870579816, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 6.740328646178248, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.002647459408172903, + "repeat_abs_p95": 0.011185326764651764, + "repeat_delta": { + "distinct_n": 8, + "max": 0.012433819170465488, + "min": -0.0019690160530919654, + "n": 8 + } + }, + "kv_usage_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.0011474371633303972, + "action_delta": { + "distinct_n": 6, + "max": 0.002518308420146869, + "min": -0.00014200125936815727, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 1.7625178532585835, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.0006510215832475053, + "repeat_abs_p95": 0.001132085438086033, + "repeat_delta": { + "distinct_n": 8, + "max": 0.0007778503404774276, + "min": -0.0012902061862114789, + "n": 8 + } + }, + "preemptions": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0, + "action_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 6 + }, + "action_signs": { + "consistency": 0.0, + "negative": 0, + "positive": 0, + "zero": 6 + }, + "effect_to_repeat_median": 0.0, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": false, + "qualifies": false, + "repeat_abs_median": 0.0, + "repeat_abs_p95": 0.0, + "repeat_delta": { + "distinct_n": 1, + "max": 0.0, + "min": 0.0, + "n": 8 + } + }, + "prefill_token_fraction": { + "action_above_repeat_p95_fraction": 0.0, + "action_abs_median": 0.0003342064107405007, + "action_delta": { + "distinct_n": 6, + "max": 0.0028536948076697266, + "min": -0.0004446109784110863, + "n": 6 + }, + "action_signs": { + "consistency": 0.8333333333333334, + "negative": 1, + "positive": 5, + "zero": 0 + }, + "effect_to_repeat_median": 0.121317147777762, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.002754815925550158, + "repeat_abs_p95": 0.005037760791822205, + "repeat_delta": { + "distinct_n": 8, + "max": 0.00289215301640644, + "min": -0.005053892022705009, + "n": 8 + } + }, + "queue_running_mean": { + "action_above_repeat_p95_fraction": 0.5, + "action_abs_median": 0.922875904173333, + "action_delta": { + "distinct_n": 6, + "max": 2.4556977948933323, + "min": -0.2263784608933328, + "n": 6 + }, + "action_signs": { + "consistency": 0.6666666666666666, + "negative": 2, + "positive": 4, + "zero": 0 + }, + "effect_to_repeat_median": 2.147073922684241, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": false, + "repeat_abs_median": 0.4298295901333322, + "repeat_abs_p95": 0.8487974595846658, + "repeat_delta": { + "distinct_n": 7, + "max": 0.8555136968933326, + "min": -0.8363244474399991, + "n": 8 + } + }, + "queue_waiting_mean": { + "action_above_repeat_p95_fraction": 0.8333333333333334, + "action_abs_median": 23.800311307593333, + "action_delta": { + "distinct_n": 6, + "max": -0.11799772912, + "min": -47.261676539759996, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 6, + "positive": 0, + "zero": 0 + }, + "effect_to_repeat_median": 1182.9517515641658, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 0.020119426913332024, + "repeat_abs_p95": 0.19249742833733324, + "repeat_delta": { + "distinct_n": 8, + "max": 0.25510083812, + "min": -0.001456834226666667, + "n": 8 + } + }, + "scheduler_steps_per_s": { + "action_above_repeat_p95_fraction": 0.6666666666666666, + "action_abs_median": 5.660000000000004, + "action_delta": { + "distinct_n": 6, + "max": 11.693333333333328, + "min": 3.3066666666666578, + "n": 6 + }, + "action_signs": { + "consistency": 1.0, + "negative": 0, + "positive": 6, + "zero": 0 + }, + "effect_to_repeat_median": 2.830000000000002, + "effect_to_repeat_median_is_infinite": false, + "gate_feature": true, + "qualifies": true, + "repeat_abs_median": 2.0, + "repeat_abs_p95": 4.641333333333332, + "repeat_delta": { + "distinct_n": 8, + "max": 4.986666666666665, + "min": -4.0, + "n": 8 + } + } + }, + "sanity": { + "invariants": { + "all_results_uncensored": true, + "expected_action_pair_count": true, + "expected_repeat_pair_count": true, + "finite_deltas": true, + "layer1_boundary_and_internal_gaps_bounded": true, + "nonnegative_counters": true, + "per_cell_state_vectors_not_all_identical": true, + "ratios_bounded": true, + "state_vectors_not_all_identical": true + }, + "red_flags": [], + "trials": 12 + }, + "start_s": 225.0, + "trial_sanity": [ + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.99, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00637513, + "max_internal_gap_s": 0.526093105, + "start_gap_s": 0.001941686 + }, + "trial_id": "sessions/rep1-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9552941176470588, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.00415122, + "max_internal_gap_s": 0.18536399, + "start_gap_s": 0.00588794 + }, + "trial_id": "sessions/rep1-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9911111111111112, + "level": "low", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.03110111, + "max_internal_gap_s": 0.52361076, + "start_gap_s": 0.000397498 + }, + "trial_id": "sessions/rep1-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9905882352941177, + "level": "mid", + "replicate": 1, + "telemetry_coverage": { + "end_gap_s": 0.033263413, + "max_internal_gap_s": 0.325952494, + "start_gap_s": 0.001285913 + }, + "trial_id": "sessions/rep1-mns64/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002571552, + "max_internal_gap_s": 0.533068266, + "start_gap_s": 0.002834243 + }, + "trial_id": "sessions/rep2-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9964705882352941, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.002910387, + "max_internal_gap_s": 0.26729158, + "start_gap_s": 0.006516028 + }, + "trial_id": "sessions/rep2-mns64/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.005037876, + "max_internal_gap_s": 0.524592261, + "start_gap_s": 0.003957027 + }, + "trial_id": "sessions/rep2-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9466666666666667, + "level": "mid", + "replicate": 2, + "telemetry_coverage": { + "end_gap_s": 0.04911377, + "max_internal_gap_s": 0.184204303, + "start_gap_s": 0.00971298 + }, + "trial_id": "sessions/rep2-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9961111111111111, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.001459145, + "max_internal_gap_s": 0.184019771, + "start_gap_s": 0.003328883 + }, + "trial_id": "sessions/rep3-mns16/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns16", + "completed_fraction": 0.9392156862745098, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.024192492, + "max_internal_gap_s": 0.174342785, + "start_gap_s": 0.000436554 + }, + "trial_id": "sessions/rep3-mns16/mid/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9966666666666667, + "level": "low", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.004830812, + "max_internal_gap_s": 0.197897948, + "start_gap_s": 0.002055499 + }, + "trial_id": "sessions/rep3-mns64/low/result.json" + }, + { + "admitted_fraction": 1.0, + "cell": "tp4_mns64", + "completed_fraction": 0.9929411764705882, + "level": "mid", + "replicate": 3, + "telemetry_coverage": { + "end_gap_s": 0.021026794, + "max_internal_gap_s": 0.186718691, + "start_gap_s": 0.041331534 + }, + "trial_id": "sessions/rep3-mns64/mid/result.json" + } + ] + } + ], + "sanity": { + "controller": { + "gpu_hours_total": 5.092428478929731, + "invariants": { + "all_sessions_complete": true, + "all_stream_validation_invariants_pass": true, + "completed_session_count": true, + "controller_complete": true, + "exact_session_set": true, + "no_controller_failures": true, + "under_h20_hour_cap": true + }, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/controller-state.json", + "red_flags": [], + "sha256": "f928a7bc8b4f6877a1f3ec71715177711428820c6a60669b29092e7371bdd734" + }, + "cumulative_coverage": { + "invariants": { + "admitted_fraction_monotonic": true, + "completed_fraction_monotonic": true, + "same_trials_at_every_checkpoint": true + }, + "red_flags": [] + }, + "red_flags": [], + "stream_bytes": { + "distinct_n": 6, + "max": 58550249.0, + "min": 52931532.0, + "n": 6 + }, + "streams": 6 + }, + "schema": "intervention-response-phase-aware-pilot-analysis-v3", + "stable_adjacent_efficacy_features": { + "0.25->0.50": [], + "0.50->0.75": [], + "0.75->1.00": [] + }, + "stable_adjacent_features": { + "0.10->0.25": [], + "0.25->0.50": [], + "0.50->0.75": [ + "graph_padding_fraction" + ], + "0.75->1.00": [ + "graph_padding_fraction", + "queue_waiting_mean" + ] + }, + "stable_incremental_efficacy_candidates": [], + "stable_incremental_efficacy_features": [], + "status": "COMPLETE" +} diff --git a/runs/intervention-response-v2/pilot-controller-state-v3.json b/runs/intervention-response-v2/pilot-controller-state-v3.json new file mode 100644 index 0000000..8b1562c --- /dev/null +++ b/runs/intervention-response-v2/pilot-controller-state-v3.json @@ -0,0 +1,342 @@ +{ + "completed_at": 1784027453.4005148, + "completed_sessions": 6, + "failures": [], + "gpu_hours_total": 5.092428478929731, + "hard_cap_h20_hours": 6.641945318645901, + "schema": "intervention-response-phase-aware-pilot-state-v3", + "sessions": { + "rep1-mns16": { + "burnin": { + "elapsed_s": 64.194648239, + "feasible": false, + "pass_rate": 0.5235294117647059 + }, + "completed_at": 1784023620.6911042, + "gpu_hours": 0.856446467505561, + "mns": 16, + "replicate": 1, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 301.068878512, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + }, + { + "early_stopped": false, + "elapsed_s": 313.268231477, + "feasible": false, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 0.5603921568627451, + "selected_count": 2550 + } + ], + "started_at": 1784022846.257515, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns16", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 60017, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep1-mns16/opprof/opprof-v1-dp0-pid458427-1784022917661752826.jsonl" + } + }, + "rep1-mns64": { + "burnin": { + "elapsed_s": 60.79926316, + "feasible": true, + "pass_rate": 1.0 + }, + "completed_at": 1784024382.2164423, + "gpu_hours": 0.8415956518385146, + "mns": 64, + "replicate": 1, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 301.022700013, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + }, + { + "early_stopped": false, + "elapsed_s": 301.100993458, + "feasible": true, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 1.0, + "selected_count": 2550 + } + ], + "started_at": 1784023620.9865165, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns64", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 62456, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep1-mns64/opprof/opprof-v1-dp0-pid471212-1784023695315485142.jsonl" + } + }, + "rep2-mns16": { + "burnin": { + "elapsed_s": 63.518994307, + "feasible": false, + "pass_rate": 0.6647058823529411 + }, + "completed_at": 1784025910.710227, + "gpu_hours": 0.8481734045346578, + "mns": 16, + "replicate": 2, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 315.780805484, + "feasible": false, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 0.31450980392156863, + "selected_count": 2550 + }, + { + "early_stopped": false, + "elapsed_s": 300.346368314, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + } + ], + "started_at": 1784025143.731658, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns16", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 58465, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep2-mns16/opprof/opprof-v1-dp0-pid496100-1784025206790752505.jsonl" + } + }, + "rep2-mns64": { + "burnin": { + "elapsed_s": 60.803771161, + "feasible": true, + "pass_rate": 1.0 + }, + "completed_at": 1784025143.4362338, + "gpu_hours": 0.8412917502721151, + "mns": 64, + "replicate": 2, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 300.605515019, + "feasible": true, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 1.0, + "selected_count": 2550 + }, + { + "early_stopped": false, + "elapsed_s": 300.354084826, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + } + ], + "started_at": 1784024382.5080929, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns64", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 61696, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep2-mns64/opprof/opprof-v1-dp0-pid483758-1784024456752184540.jsonl" + } + }, + "rep3-mns16": { + "burnin": { + "elapsed_s": 64.324514503, + "feasible": false, + "pass_rate": 0.5588235294117647 + }, + "completed_at": 1784026691.6293187, + "gpu_hours": 0.8631463612450494, + "mns": 16, + "replicate": 3, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 300.711109876, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + }, + { + "early_stopped": false, + "elapsed_s": 317.011840655, + "feasible": false, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 0.2635294117647059, + "selected_count": 2550 + } + ], + "started_at": 1784025911.00421, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns16", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 61438, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep3-mns16/opprof/opprof-v1-dp0-pid508913-1784025983046041167.jsonl" + } + }, + "rep3-mns64": { + "burnin": { + "elapsed_s": 60.850091109, + "feasible": true, + "pass_rate": 1.0 + }, + "completed_at": 1784027453.3790896, + "gpu_hours": 0.8417748435338338, + "mns": 64, + "replicate": 3, + "runs": [ + { + "early_stopped": false, + "elapsed_s": 300.710505069, + "feasible": true, + "level": "low", + "offered_req_s_per_gpu": 1.5, + "pass_rate": 1.0, + "selected_count": 1800 + }, + { + "early_stopped": false, + "elapsed_s": 300.869326418, + "feasible": true, + "level": "mid", + "offered_req_s_per_gpu": 2.125, + "pass_rate": 1.0, + "selected_count": 2550 + } + ], + "started_at": 1784026691.96421, + "status": "complete", + "validation": { + "accounting_mode": "graceful-footer", + "cell": "tp4_mns64", + "invariants": { + "anchor_intervals_present": true, + "compile_capture_pre_ready": true, + "encoded_balanced": true, + "footer_sidecar_agrees": true, + "last_step_matches": true, + "layer1_contiguous": true, + "layer1_zero_drops": true, + "one_footer_last": true, + "one_ready_marker": true, + "sidecar_final": true, + "warmup_exact_16": true, + "warmup_long": true, + "written_matches_records": true + }, + "layer1_records": 64776, + "post_ready_capture_events": [], + "stream": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/runs/pilot/sessions/rep3-mns64/opprof/opprof-v1-dp0-pid521797-1784026767359409666.jsonl" + } + } + }, + "started_at": 1784022845.964628, + "status": "complete" +} diff --git a/runs/intervention-response-v2/pilot-manifest-v3.json b/runs/intervention-response-v2/pilot-manifest-v3.json new file mode 100644 index 0000000..4590e51 --- /dev/null +++ b/runs/intervention-response-v2/pilot-manifest-v3.json @@ -0,0 +1,324 @@ +{ + "budget": { + "expected_h20_hours": [ + 4.8, + 6.641945318645901 + ], + "expected_wall_minutes": [ + 75, + 95 + ], + "global_hard_cap_h20_hours": 8.0, + "hard_cap_h20_hours": 6.641945318645901, + "prior_attempt_h20_hours": 1.3580546813540988, + "session_estimate_h20_hours": 1.0 + }, + "burnin": { + "anchor": 0.18919793755240089, + "arrival_order_sha256": "6c0ac4cb9a30ef501eeeacc8e6cc631c345e976db5ccf530ea5a1ec706d62a24", + "input_length_order_sha256": "7939cc20e1a00d1031d27d71508789f38decbbbb6ea59a1df18b2ec342fd2ef8", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "84f4809acbc8acd3b1d14dfa357134a1dc0b9287341624b33f598dafeef54dc7", + "selected_count": 510, + "study": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/studies/burnin-tp4.json", + "study_sha256": "5d6c2098042909a863efd3112818fbee9bafe96f22898ac98b66846dbe1fef0f" + }, + "checkpoints": { + "fractions": [ + 0.1, + 0.25, + 0.5, + 0.75, + 1.0 + ], + "seconds": [ + 30.0, + 75.0, + 150.0, + 225.0, + 300.0 + ] + }, + "engine": { + "disable_slo_early_stop": true, + "duration_s": 300.0, + "mns_endpoints": [ + 16, + 64 + ], + "replay_time_scale": 0.5, + "safety_deadline_s": 360.0, + "tp": 4 + }, + "private": { + "merged_traces": { + "1": { + "bytes": 337429767, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep1.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9420, + "sha256": "68983266aa0e66aa589562f7c08edbd966f9ba4405e20c105adb43777d2dfbf5", + "source_sha256": [ + "b242d1d9086df3accab57b4c92445d5edd581e12f47e12cea227aa63964c6930", + "d23b549f7b69af3647308677bbf76f818a3c226a1c98f9a9f93f09ceee46be87" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low1.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high1.jsonl" + ] + }, + "2": { + "bytes": 337509330, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep2.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9457, + "sha256": "f38e8938f6a481fc6725b71b21aa04ff7eaf79783cdfd6e41aa2f074156f00c2", + "source_sha256": [ + "4cbb0baac082bd54af562ce2f39104c5c23b4671672da365a67b1e8c146adf9f", + "bb0bcd2564a88000f435f12feb21c7c902eafc9ea5fe916adfe9d1eae47f3f9a" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low2.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high2.jsonl" + ] + }, + "3": { + "bytes": 337450256, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep3.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9431, + "sha256": "3094084b0bb20cc02eecf465091a5c919b4e5b112f704cdc36a563d1efdcee46", + "source_sha256": [ + "1f7ececb142f9a363d2d1ca25eb7b8488b2cc319a51b55faa384f2a3d51f2142", + "6f326234791e1cff4ff866bface0d097d0d6e3844eebb1c97653d8e9c35e9397" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low3.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high3.jsonl" + ] + } + } + }, + "repetitions": { + "1": { + "duration_s": 300.0, + "load_order": [ + "low", + "mid" + ], + "merged_trace": { + "bytes": 337429767, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep1.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9420, + "sha256": "68983266aa0e66aa589562f7c08edbd966f9ba4405e20c105adb43777d2dfbf5", + "source_sha256": [ + "b242d1d9086df3accab57b4c92445d5edd581e12f47e12cea227aa63964c6930", + "d23b549f7b69af3647308677bbf76f818a3c226a1c98f9a9f93f09ceee46be87" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low1.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high1.jsonl" + ] + }, + "selections": { + "low": { + "anchor": 0.34815061842962747, + "arrival_order_sha256": "0b9473052a28b2eb98bb4bda59c726d6c74cd42656c00046f0559cefd38e0c42", + "input_length_order_sha256": "0172aa1fd7d994e9d95857c32293e25f6c537616b40e01034308a0d57cd94139", + "offered_req_s": 6.0, + "offered_req_s_per_gpu": 1.5, + "request_id_order_sha256": "af50f0c3edfc6f9fb74734b2501378fc8ba6fdd3dd3fe84f2bb1d68540eee69b", + "selected_count": 1800, + "target_count": 1800, + "target_req_s_per_gpu": 1.5 + }, + "mid": { + "anchor": 0.48686986110831465, + "arrival_order_sha256": "c2ad99986ce558da5901a9c5ec0a00bd69f198c981d8779235f2773a5c87f1c0", + "input_length_order_sha256": "9442bfebdc3fab5062dc1f4d688dc28c02afe3fd806c56dd8159f0ac7e6d0b94", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "0bb61dbc9c26875e991d0d4f984134910d37463e5063f86ee960cf4f8aafb771", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + } + }, + "source_roles": [ + "low1", + "high1" + ], + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep1-tp4.json", + "study_sha256": "ecfff96e33d458eb1e3b9a6d24386f00cc6f1b19ff926e2ec6320b3f671a7ae3" + }, + "2": { + "duration_s": 300.0, + "load_order": [ + "mid", + "low" + ], + "merged_trace": { + "bytes": 337509330, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep2.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9457, + "sha256": "f38e8938f6a481fc6725b71b21aa04ff7eaf79783cdfd6e41aa2f074156f00c2", + "source_sha256": [ + "4cbb0baac082bd54af562ce2f39104c5c23b4671672da365a67b1e8c146adf9f", + "bb0bcd2564a88000f435f12feb21c7c902eafc9ea5fe916adfe9d1eae47f3f9a" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low2.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high2.jsonl" + ] + }, + "selections": { + "low": { + "anchor": 0.3411190868318572, + "arrival_order_sha256": "f4eafbdf7420e8a97e38d46fff1aec5fdbc2a049133862a793b4635b51061b7f", + "input_length_order_sha256": "cdf82cfc546651f1859c53348eace72348572f09e6e1879ad77d51b489a998fe", + "offered_req_s": 6.0, + "offered_req_s_per_gpu": 1.5, + "request_id_order_sha256": "adc728f189a1ce0301609cdbae7d1dc2318609b13eafb94bcb90903bb7998eb1", + "selected_count": 1800, + "target_count": 1800, + "target_req_s_per_gpu": 1.5 + }, + "mid": { + "anchor": 0.4825698948735577, + "arrival_order_sha256": "b9fc12cf3f86bc8a79bee65296e65aa2b8bf2aeca46b2887094c669adcbb9a00", + "input_length_order_sha256": "d8d4bd6fc8ba852a45605b673b6b3e4f33b58f459e69f2a032d226ee175b074e", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "56a0616b6b54abafd37875c7cb25f8639afef2706ccc55dfbe568f45859ea382", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + } + }, + "source_roles": [ + "low2", + "high2" + ], + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep2-tp4.json", + "study_sha256": "d92a576db031db24bb58f354ea725d7f7567cb76699d387117ac5a6c9317bbb9" + }, + "3": { + "duration_s": 300.0, + "load_order": [ + "low", + "mid" + ], + "merged_trace": { + "bytes": 337450256, + "path": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/traces/rep3.jsonl", + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + "rows": 9431, + "sha256": "3094084b0bb20cc02eecf465091a5c919b4e5b112f704cdc36a563d1efdcee46", + "source_sha256": [ + "1f7ececb142f9a363d2d1ca25eb7b8488b2cc319a51b55faa384f2a3d51f2142", + "6f326234791e1cff4ff866bface0d097d0d6e3844eebb1c97653d8e9c35e9397" + ], + "sources": [ + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/low3.jsonl", + "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/private/traces/high3.jsonl" + ] + }, + "selections": { + "low": { + "anchor": 0.3459572828521402, + "arrival_order_sha256": "68de977e2ea7396ffe9406675a3d2c5ebecd1b9ea5afc8116c836cc53cfca188", + "input_length_order_sha256": "dffa79545e738b899bec2e9d282764a3388fe38b162982fb2d26928e17ef6016", + "offered_req_s": 6.0, + "offered_req_s_per_gpu": 1.5, + "request_id_order_sha256": "4993c26442fc6b5a4c09b6f576fd8b3cb1be7ce83b7a771e4a57ee7600de688d", + "selected_count": 1800, + "target_count": 1800, + "target_req_s_per_gpu": 1.5 + }, + "mid": { + "anchor": 0.48664343020532463, + "arrival_order_sha256": "efce7339e22d3618cb4d55e6b55bfddb2c563c18faba2a992d5829c13e3f55e9", + "input_length_order_sha256": "0792b05fff6729fbd92ab2bb4cb6d31bea7799e232ad42772936bc06efbafb54", + "offered_req_s": 8.5, + "offered_req_s_per_gpu": 2.125, + "request_id_order_sha256": "2a2fabe2c4cf176aeb7e0d32fb8e7dbb1f27429a2e7a0cd18d7d186f23096f19", + "selected_count": 2550, + "target_count": 2550, + "target_req_s_per_gpu": 2.125 + } + }, + "source_roles": [ + "low3", + "high3" + ], + "study": "/home/admin/cpfs/wjh/intervention-response-v3-20260714/private/studies/rep3-tp4.json", + "study_sha256": "fb8ffe256dace32f4ca8a8d49b662d98c3b69b94ecc8fa826e43068b238884ab" + } + }, + "sanity": { + "distinct_selected_sets": 6, + "invariants": { + "all_counts_positive": true, + "load_levels_two": true, + "selection_hashes_unique": true, + "selection_sets_disjoint_across_repetitions": true, + "six_sessions": true, + "three_repetitions": true + }, + "red_flags": [], + "selected_sets": 6 + }, + "schema": "intervention-response-phase-aware-pilot-manifest-v3", + "sessions": [ + { + "mns": 16, + "replicate": 1, + "session": "rep1-mns16" + }, + { + "mns": 64, + "replicate": 1, + "session": "rep1-mns64" + }, + { + "mns": 64, + "replicate": 2, + "session": "rep2-mns64" + }, + { + "mns": 16, + "replicate": 2, + "session": "rep2-mns16" + }, + { + "mns": 16, + "replicate": 3, + "session": "rep3-mns16" + }, + { + "mns": 64, + "replicate": 3, + "session": "rep3-mns64" + } + ], + "source": { + "base_manifest": "/home/admin/cpfs/wjh/fidelity-prefix-pilot-20260714/runs/p1b/pilot-manifest.json", + "base_manifest_sha256": "86b249a087ab1bcd51cd8de3d1da6bf34afd41af977eb58c93229b941dbf5c8a", + "prior_attempt_failure": [ + { + "failure": "TimeoutError('client timeout: tp4_mns16 high')", + "session": "rep1-mns16" + } + ], + "prior_attempt_h20_hours": 1.3580546813540988, + "prior_attempt_state": "/home/admin/cpfs/wjh/intervention-response-v2-20260714/runs/pilot/controller-state.json", + "prior_attempt_state_sha256": "c88030ae36862abfc643a855eb3630fb18cd74a58b388514aa31ec1ab485a431", + "source_trace": "/home/admin/cpfs/wjh/aituner/aituner/trace_windows/traces/chat_w20260312_1000.jsonl", + "source_trace_sha256": "875ba869775deb78086477919f03b322da14e2673c7d070e26528c4190912757", + "window_id": "chat_w20260312_1000" + }, + "status": "PASS" +} diff --git a/runs/intervention-response-v2/pilot_controller.py b/runs/intervention-response-v2/pilot_controller.py new file mode 100644 index 0000000..fe41848 --- /dev/null +++ b/runs/intervention-response-v2/pilot_controller.py @@ -0,0 +1,578 @@ +#!/usr/bin/env python3 +"""Serialized, resumable controller for the 300-second phase-aware pilot.""" + +from __future__ import annotations + +import argparse +import json +import os +import shlex +import signal +import subprocess +import sys +import time +from pathlib import Path +from typing import Any, Mapping + + +HERE = Path(__file__).resolve().parent +PHASE6 = HERE.parent / "opprof-phase6" +sys.path.insert(0, str(PHASE6)) + +import opprof_phase6_controller as base # noqa: E402 + + +SCHEMA = "intervention-response-phase-aware-pilot-state-v3" +SESSION_ESTIMATE_H20_HOURS = 1.0 +SAFETY_H20_HOURS = 0.20 +CLIENT_TIMEOUT_S = 450.0 + + +def atomic_json(path: Path, payload: Any) -> None: + base.atomic_json(path, payload) + + +def wait_all_idle(timeout_s: float = 30.0) -> None: + deadline = time.monotonic() + timeout_s + last_error: Exception | None = None + while time.monotonic() < deadline: + try: + base.assert_all_idle() + return + except RuntimeError as error: + last_error = error + time.sleep(1.0) + raise last_error or RuntimeError("GPU idle timeout") + + +def configure(args: argparse.Namespace, manifest: dict[str, Any]) -> None: + base.WORKDIR = args.run_root.parent + base.RUN_ROOT = args.run_root + base.STATE = args.run_root / "controller-state.json" + base.SOURCE = args.vllm_source + base.VENV = args.venv + base.AITUNER = args.aituner_root + base.MODEL = args.model + base.CLIENT = args.client + base.GPU_LIMIT = float(manifest["budget"]["hard_cap_h20_hours"]) + base.MARKER = "intervention-response-phase-aware-v3" + base.CELLS = { + f"tp4_mns{mns}": {"tp": 4, "mns": int(mns)} + for mns in manifest["engine"]["mns_endpoints"] + } + + +def validate_inputs(args: argparse.Namespace, manifest: dict[str, Any]) -> None: + if manifest.get("schema") != "intervention-response-phase-aware-pilot-manifest-v3": + raise RuntimeError("unexpected phase-aware pilot manifest schema") + if manifest.get("status") != "PASS": + raise RuntimeError("phase-aware pilot manifest did not pass preflight") + if abs( + float(manifest["budget"]["session_estimate_h20_hours"]) + - SESSION_ESTIMATE_H20_HOURS + ) > 1e-12: + raise RuntimeError("controller and manifest session-cost estimates disagree") + failed_invariants = [ + name + for name, passed in manifest.get("sanity", {}).get("invariants", {}).items() + if not passed + ] + if failed_invariants: + raise RuntimeError(f"phase-aware pilot invariants failed: {failed_invariants}") + + required = { + "manifest": args.manifest, + "aituner_root": args.aituner_root, + "vllm_source": args.vllm_source, + "venv_python": args.venv / "bin/python", + "venv_vllm": args.venv / "bin/vllm", + "model": args.model, + "client": args.client, + "burnin_study": Path(manifest["burnin"]["study"]), + } + for replicate, repetition in manifest["repetitions"].items(): + required[f"rep{replicate}_study"] = Path(repetition["study"]) + required[f"rep{replicate}_trace"] = Path( + repetition["merged_trace"]["path"] + ) + missing = {name: str(path) for name, path in required.items() if not path.exists()} + if missing: + raise RuntimeError(f"phase-aware pilot input paths missing: {missing}") + + +def warmup_selection(repetition: Mapping[str, Any]) -> Mapping[str, Any]: + return max( + repetition["selections"].values(), + key=lambda selection: float(selection["offered_req_s_per_gpu"]), + ) + + +def dry_run_plan(args: argparse.Namespace, manifest: dict[str, Any]) -> dict[str, Any]: + sessions = [] + for index, session in enumerate(manifest["sessions"]): + cell = f"tp4_mns{int(session['mns'])}" + entry = {"cell": cell, "gpus": (0, 1, 2, 3), "port": 8950 + index} + repetition = manifest["repetitions"][str(session["replicate"])] + session_root = args.run_root / "sessions" / str(session["session"]) + warmup = warmup_selection(repetition) + commands = { + "server": base.server_command(cell, entry["gpus"], entry["port"]), + "warmup": client_command( + entry, + study=repetition["study"], + anchor=float(warmup["anchor"]), + output=session_root / "warmup", + warmup=True, + ), + "burnin": client_command( + entry, + study=manifest["burnin"]["study"], + anchor=float(manifest["burnin"]["anchor"]), + output=session_root / "burnin", + warmup=False, + ), + } + for level in repetition["load_order"]: + selection = repetition["selections"][level] + commands[level] = client_command( + entry, + study=repetition["study"], + anchor=float(selection["anchor"]), + output=session_root / level, + warmup=False, + ) + sessions.append( + { + "session": session["session"], + "replicate": int(session["replicate"]), + "mns": int(session["mns"]), + "port": entry["port"], + "load_order": repetition["load_order"], + "remaining_projection_h20_hours": remaining_projection( + len(manifest["sessions"]), index + ), + "commands": { + role: shlex.join(command) for role, command in commands.items() + }, + } + ) + return { + "schema": "intervention-response-phase-aware-pilot-dry-run-v3", + "status": "PASS", + "manifest": str(args.manifest), + "run_root": str(args.run_root), + "session_count": len(sessions), + "projected_h20_hours": remaining_projection(len(sessions), 0), + "hard_cap_h20_hours": float(manifest["budget"]["hard_cap_h20_hours"]), + "sessions": sessions, + } + + +def load_state(path: Path, hard_cap: float) -> dict[str, Any]: + if path.exists(): + return json.loads(path.read_text(encoding="utf-8")) + return { + "schema": SCHEMA, + "status": "initialized", + "hard_cap_h20_hours": hard_cap, + "gpu_hours_total": 0.0, + "completed_sessions": 0, + "sessions": {}, + "failures": [], + "started_at": time.time(), + } + + +def save_state(path: Path, state: dict[str, Any]) -> None: + atomic_json(path, state) + + +def append_echo(run_root: Path, line: str) -> None: + run_root.mkdir(parents=True, exist_ok=True) + with (run_root / "launch-echo.log").open("a", encoding="utf-8") as target: + target.write(line + "\n") + print(line, flush=True) + + +def remaining_projection(session_count: int, index: int) -> float: + return (session_count - index) * SESSION_ESTIMATE_H20_HOURS + SAFETY_H20_HOURS + + +def start_server( + *, session: dict[str, Any], index: int, run_root: Path +) -> dict[str, Any]: + cell = f"tp4_mns{int(session['mns'])}" + gpus = (0, 1, 2, 3) + session_root = run_root / "sessions" / str(session["session"]) + session_root.mkdir(parents=True, exist_ok=True) + port = 8950 + index + command = base.server_command(cell, gpus, port) + with (session_root / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"SERVER {shlex.join(command)}\n") + server_log = (session_root / "server.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update( + { + "CUDA_VISIBLE_DEVICES": "0,1,2,3", + "VLLM_OPPROF_DIR": str(session_root / "opprof"), + "OPPROF_PHASE6_MARKER": base.MARKER, + "AITUNER_ROOT": str(base.AITUNER), + "HF_HUB_OFFLINE": "1", + "TRANSFORMERS_OFFLINE": "1", + "PYTHONUNBUFFERED": "1", + } + ) + server = subprocess.Popen( + command, + cwd=base.SOURCE, + env=environment, + stdout=server_log, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + base.OWNED_PGIDS.add(server.pid) + return { + "cell": cell, + "gpus": gpus, + "port": port, + "dir": session_root, + "server": server, + "server_handle": server_log, + "spawned_at": time.time(), + "results": [], + } + + +def client_command( + entry: dict[str, Any], + *, + study: str, + anchor: float, + output: Path, + warmup: bool, +) -> list[str]: + config = base.CELLS[entry["cell"]] + command = [ + "taskset", + "-c", + base.cpu_mask(entry["gpus"]), + str(base.VENV / "bin/python"), + str(base.CLIENT), + "warmup" if warmup else "run-anchor", + "--study", + study, + "--cell", + entry["cell"], + "--anchor", + str(anchor), + "--tp", + str(config["tp"]), + "--mns", + str(config["mns"]), + "--base-url", + f"http://127.0.0.1:{entry['port']}", + "--result-dir", + str(output), + "--disable-slo-early-stop", + ] + return command + + +def run_client( + *, + entry: dict[str, Any], + role: str, + study: str, + selection: dict[str, Any], + output: Path, + state: dict[str, Any], + warmup: bool = False, +) -> dict[str, Any]: + command = client_command( + entry, + study=study, + anchor=float(selection["anchor"]), + output=output, + warmup=warmup, + ) + with (entry["dir"] / "commands.log").open("a", encoding="utf-8") as log: + log.write(f"CLIENT role={role} {shlex.join(command)}\n") + handle = (output.parent / f"{output.name}.log").open("ab", buffering=0) + environment = os.environ.copy() + environment.update({"AITUNER_ROOT": str(base.AITUNER), "PYTHONUNBUFFERED": "1"}) + process = subprocess.Popen( + command, + cwd=base.WORKDIR, + env=environment, + stdout=handle, + stderr=subprocess.STDOUT, + start_new_session=True, + ) + deadline = time.monotonic() + (180.0 if warmup else CLIENT_TIMEOUT_S) + try: + while process.poll() is None: + if time.monotonic() > deadline: + raise TimeoutError(f"client timeout: {entry['cell']} {role}") + if entry["server"].poll() is not None: + raise RuntimeError(f"server exited during {entry['cell']} {role}") + base.assert_no_other_compute() + if state["gpu_hours_total"] + base.live_gpu_hours([entry]) >= base.GPU_LIMIT: + raise RuntimeError("phase-aware pilot H20-hour hard cap reached") + time.sleep(1.0) + except Exception: + try: + os.killpg(process.pid, signal.SIGTERM) + except ProcessLookupError: + pass + try: + process.wait(timeout=10.0) + except subprocess.TimeoutExpired: + try: + os.killpg(process.pid, signal.SIGKILL) + except ProcessLookupError: + pass + process.wait(timeout=10.0) + raise + finally: + handle.close() + if process.returncode: + raise RuntimeError( + f"client failed: cell={entry['cell']} role={role} rc={process.returncode}" + ) + result = json.loads((output / "result.json").read_text(encoding="utf-8")) + validate_result( + result=result, + selection=selection, + role=role, + warmup=warmup, + ) + entry["results"].append( + { + "anchor": float(selection["anchor"]), + "dir": str(output), + "kind": result["kind"], + } + ) + return result + + +def validate_result( + *, result: dict[str, Any], selection: dict[str, Any], role: str, warmup: bool +) -> None: + if result.get("slo_early_stop_disabled") is not True: + raise RuntimeError(f"SLO early stop was not disabled: {role}") + if warmup: + if result["kind"] != "warmup" or int(result["selection"]["count"]) != 16: + raise RuntimeError(f"invalid warmup result: {role}") + if not all( + result["invariants"].get(key, False) + for key in ("warmup_16", "warmup_exact_16", "warmup_long") + ): + raise RuntimeError(f"warmup invariant failed: {role}") + return + if bool(result["early_stopped"]): + raise RuntimeError(f"uncensored run early-stopped: {role}") + if int(result["selection"]["count"]) != int(selection["selected_count"]): + raise RuntimeError(f"selection count mismatch: {role}") + for key, manifest_key in ( + ("request_id_order_sha256", "request_id_order_sha256"), + ("arrival_order_sha256", "arrival_order_sha256"), + ("raw_length_order_sha256", "input_length_order_sha256"), + ): + if result["selection"][key] != selection[manifest_key]: + raise RuntimeError(f"selection hash mismatch {key}: {role}") + if int(result["observed_count"]) != int(selection["selected_count"]): + raise RuntimeError(f"request accounting mismatch: {role}") + + +def execute_session( + *, + index: int, + session: dict[str, Any], + manifest: dict[str, Any], + run_root: Path, + state_path: Path, + state: dict[str, Any], +) -> None: + name = str(session["session"]) + if state["sessions"].get(name, {}).get("status") == "complete": + return + projection = remaining_projection(len(manifest["sessions"]), index) + if state["gpu_hours_total"] + projection > base.GPU_LIMIT: + state["status"] = "budget_projection_stop" + state["budget_stop"] = { + "before_session": name, + "spent_h20_hours": state["gpu_hours_total"], + "remaining_projection_h20_hours": projection, + "hard_cap_h20_hours": base.GPU_LIMIT, + } + save_state(state_path, state) + raise RuntimeError(f"projected pilot cost exceeds hard cap before {name}") + + replicate = str(session["replicate"]) + repetition = manifest["repetitions"][replicate] + echo = ( + f"PHASE_PILOT_SESSION_ECHO session={name} tp=4 mns={session['mns']} " + f"gpus=0-3 workload={manifest['source']['window_id']} duration_s=300 " + f"loads={','.join(repetition['load_order'])} disable_slo_early_stop=true " + f"spent_h20h={state['gpu_hours_total']:.6f} " + f"remaining_projection_h20h={projection:.3f} cap_h20h={base.GPU_LIMIT:.1f} " + f"manifest={run_root / 'pilot-manifest.json'}" + ) + append_echo(run_root, echo) + wait_all_idle() + session_state = { + "status": "starting", + "replicate": int(replicate), + "mns": int(session["mns"]), + "started_at": time.time(), + "runs": [], + } + state["status"] = "running" + state["sessions"][name] = session_state + save_state(state_path, state) + entry = start_server(session=session, index=index, run_root=run_root) + failure: Exception | None = None + try: + base.wait_ready(entry) + warmup = warmup_selection(repetition) + session_state["status"] = "warmup" + save_state(state_path, state) + run_client( + entry=entry, + role="warmup", + study=repetition["study"], + selection=warmup, + output=entry["dir"] / "warmup", + state=state, + warmup=True, + ) + session_state["status"] = "burnin" + save_state(state_path, state) + burnin = manifest["burnin"] + burnin_result = run_client( + entry=entry, + role="burnin", + study=burnin["study"], + selection=burnin, + output=entry["dir"] / "burnin", + state=state, + ) + session_state["burnin"] = { + "pass_rate": burnin_result["pass_rate"], + "feasible": burnin_result["feasible"], + "elapsed_s": burnin_result["interval"]["elapsed_s"], + } + session_state["status"] = "measured" + save_state(state_path, state) + for level in repetition["load_order"]: + selection = repetition["selections"][level] + result = run_client( + entry=entry, + role=level, + study=repetition["study"], + selection=selection, + output=entry["dir"] / level, + state=state, + ) + session_state["runs"].append( + { + "level": level, + "selected_count": selection["selected_count"], + "offered_req_s_per_gpu": selection["offered_req_s_per_gpu"], + "pass_rate": result["pass_rate"], + "feasible": result["feasible"], + "elapsed_s": result["interval"]["elapsed_s"], + "early_stopped": result["early_stopped"], + } + ) + save_state(state_path, state) + session_state["status"] = "stopping" + save_state(state_path, state) + except Exception as error: # noqa: BLE001 + failure = error + finally: + try: + base.stop_entry(entry) + except Exception as error: # noqa: BLE001 + failure = failure or error + time.sleep(2.0) + try: + wait_all_idle() + except Exception as error: # noqa: BLE001 + failure = failure or error + + session_hours = base.live_gpu_hours([entry]) + state["gpu_hours_total"] += session_hours + session_state["gpu_hours"] = session_hours + if failure is not None: + session_state["status"] = "failed" + session_state["failure"] = repr(failure) + state["status"] = "failed" + state["failures"].append({"session": name, "failure": repr(failure)}) + save_state(state_path, state) + raise failure + validation = base.validate_cell(entry) + session_state["validation"] = validation + session_state["status"] = "complete" + session_state["completed_at"] = time.time() + state["completed_sessions"] += 1 + save_state(state_path, state) + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--manifest", type=Path, required=True) + result.add_argument("--run-root", type=Path, required=True) + result.add_argument("--aituner-root", type=Path, required=True) + result.add_argument("--vllm-source", type=Path, required=True) + result.add_argument("--venv", type=Path, required=True) + result.add_argument("--model", type=Path, required=True) + result.add_argument("--client", type=Path, required=True) + result.add_argument("--dry-run", action="store_true") + return result + + +def main() -> None: + args = parser().parse_args() + manifest = json.loads(args.manifest.read_text(encoding="utf-8")) + validate_inputs(args, manifest) + configure(args, manifest) + if args.dry_run: + print(json.dumps(dry_run_plan(args, manifest), indent=2, sort_keys=True)) + return + args.run_root.mkdir(parents=True, exist_ok=True) + copied_manifest = args.run_root / "pilot-manifest.json" + if not copied_manifest.exists(): + atomic_json(copied_manifest, manifest) + state_path = args.run_root / "controller-state.json" + state = load_state(state_path, base.GPU_LIMIT) + state["status"] = "running" + save_state(state_path, state) + for index, session in enumerate(manifest["sessions"]): + execute_session( + index=index, + session=session, + manifest=manifest, + run_root=args.run_root, + state_path=state_path, + state=state, + ) + state["status"] = "complete" + state["completed_at"] = time.time() + save_state(state_path, state) + wait_all_idle() + print( + json.dumps( + { + "status": state["status"], + "completed_sessions": state["completed_sessions"], + "gpu_hours_total": state["gpu_hours_total"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v2/platform-v3.json b/runs/intervention-response-v2/platform-v3.json new file mode 100644 index 0000000..5a45de1 --- /dev/null +++ b/runs/intervention-response-v2/platform-v3.json @@ -0,0 +1,46 @@ +{ + "schema": "intervention-response-platform-v3", + "host_alias": "dash0", + "hostname": "ds-07429c65-1-6c5fd97778-9vhkr", + "os": "Ubuntu 24.04.2 LTS", + "kernel": "Linux 5.10.134-013.8.2.kangaroo.al8.x86_64 x86_64 GNU/Linux", + "cpu": { + "logical_cpus": 160, + "model": "Intel(R) Xeon(R) Processor" + }, + "gpu": { + "count": 8, + "experiment_ids": [ + 0, + 1, + 2, + 3 + ], + "memory_mib_each": 97871, + "model": "NVIDIA H20" + }, + "runtime": { + "driver": "580.95.05", + "python": "3.12.3", + "torch": "2.11.0+cu130", + "torch_cuda": "13.0", + "vllm_editable_source": "/home/admin/cpfs/wjh/opprof-phase2-dash0-20260711/vllm-v0.24.0", + "vllm_git_head": "4b253fd8619764b6971a7f2e3a3aa7545f6ace05", + "vllm_git_status_clean": true, + "vllm_import_path": "/home/admin/cpfs/wjh/opprof-phase2-dash0-20260711/vllm-v0.24.0/vllm/__init__.py", + "vllm_reported_version": "0.24.1.dev3+g668cfb7e2" + }, + "model": { + "bytes": 61084188040, + "path": "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + }, + "aituner_commit": "7fd956355072399b1b56946ecd8007da8f9d2fb0", + "experiment": { + "completed_at": "2026-07-14T19:10:53+08:00", + "incremental_h20_hours": 5.092428478929731, + "prior_invalid_attempt_h20_hours": 1.3580546813540988, + "started_at": "2026-07-14T17:54:05+08:00", + "total_h20_hours": 6.45048316028383, + "wall_minutes": 76.79059811433156 + } +} diff --git a/runs/intervention-response-v2/prepare_pilot.py b/runs/intervention-response-v2/prepare_pilot.py new file mode 100644 index 0000000..14c5a00 --- /dev/null +++ b/runs/intervention-response-v2/prepare_pilot.py @@ -0,0 +1,352 @@ +#!/usr/bin/env python3 +"""Prepare the uncensored 300-second TP4 two-load matched pilot on dash0.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import os +import sys +from pathlib import Path +from typing import Any + + +AITUNER_ROOT = Path(os.environ.get("AITUNER_ROOT", Path(__file__).resolve().parents[2])) +sys.path.insert(0, str(AITUNER_ROOT / "src")) + +from aituner.spec import load_study_spec # noqa: E402 +from aituner.trace import load_trace_requests, select_requests_for_threshold # noqa: E402 + + +SCHEMA = "intervention-response-phase-aware-pilot-manifest-v3" +TP = 4 +MNS_ENDPOINTS = (16, 64) +REPLAY_TIME_SCALE = 0.5 +EXPECTED_DURATION_S = 300.0 +SAFETY_DEADLINE_S = 360.0 +LOADS_REQ_S_GPU = {"low": 1.5, "mid": 2.125} +REPLICATE_ROLE_PAIRS = (("low1", "high1"), ("low2", "high2"), ("low3", "high3")) +LOAD_ORDERS = { + 1: ("low", "mid"), + 2: ("mid", "low"), + 3: ("low", "mid"), +} +SESSION_ORDER = ( + (1, 16), + (1, 64), + (2, 64), + (2, 16), + (3, 16), + (3, 64), +) + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def order_hash(values: list[str]) -> str: + return hashlib.sha256("\n".join(values).encode()).hexdigest() + + +def attainable_anchor(requests: list[Any], target_count: int) -> tuple[float, list[Any]]: + ordered = sorted(float(request.sampling_u) for request in requests) + if not ordered: + raise ValueError("no requests remain after study filtering") + if target_count <= 0 or target_count > len(ordered): + raise ValueError( + f"target count {target_count} is outside available range 1..{len(ordered)}" + ) + indices = sorted( + { + max(0, min(len(ordered) - 1, target_count - 1)), + max(0, min(len(ordered) - 1, target_count)), + } + ) + candidates = [] + for index in indices: + anchor = ordered[index] + selected = select_requests_for_threshold(requests, threshold=anchor) + candidates.append((abs(len(selected) - target_count), len(selected), anchor, selected)) + _error, _count, anchor, selected = min( + candidates, key=lambda item: (item[0], item[1], item[2]) + ) + return anchor, selected + + +def selection_record(selected: list[Any], *, duration_s: float) -> dict[str, Any]: + return { + "anchor": max(float(request.sampling_u) for request in selected), + "selected_count": len(selected), + "offered_req_s": len(selected) / duration_s, + "offered_req_s_per_gpu": len(selected) / duration_s / TP, + "request_id_order_sha256": order_hash([request.row_id for request in selected]), + "arrival_order_sha256": order_hash( + [f"{request.arrival_s:.12f}" for request in selected] + ), + "input_length_order_sha256": order_hash( + [str(request.prompt_tokens_hint) for request in selected] + ), + } + + +def resolve_role_trace(base: dict[str, Any], role: str) -> Path: + windows_path = Path(base["private"]["windows"]) + windows = json.loads(windows_path.read_text(encoding="utf-8"))["windows"] + window_id = f"fidelity_pilot_{role}" + record = next(item for item in windows if item["window_id"] == window_id) + trace = Path(record["trace_file"]) + if not trace.is_absolute(): + trace = (windows_path.parent / trace).resolve() + return trace + + +def private_request_id( + *, source_sha256: str, line_number: int, original_id: str +) -> str: + payload = f"{source_sha256}:{line_number}:{original_id}".encode() + return f"phase-v3-{hashlib.sha256(payload).hexdigest()}" + + +def merge_role_traces(sources: tuple[Path, Path], target: Path) -> dict[str, Any]: + target.parent.mkdir(parents=True, exist_ok=True) + temporary = target.with_suffix(target.suffix + ".tmp") + rows = 0 + with temporary.open("w", encoding="utf-8") as output: + for source in sources: + source_digest = sha256_file(source) + with source.open(encoding="utf-8") as input_file: + for line_number, line in enumerate(input_file, start=1): + if not line.strip(): + continue + row = json.loads(line) + original_id = str( + row.get("request_id") or row.get("id") or line_number + ) + row["request_id"] = private_request_id( + source_sha256=source_digest, + line_number=line_number, + original_id=original_id, + ) + output.write(json.dumps(row, ensure_ascii=False) + "\n") + rows += 1 + os.replace(temporary, target) + return { + "path": str(target), + "sha256": sha256_file(target), + "bytes": target.stat().st_size, + "rows": rows, + "sources": [str(source) for source in sources], + "source_sha256": [sha256_file(source) for source in sources], + "request_id_scheme": "sha256(source_sha256:line_number:original_id)", + } + + +def materialize_study( + source: Path, target: Path, *, replicate: int, trace_override: Path +) -> Path: + payload = json.loads(source.read_text(encoding="utf-8")) + payload["study_id"] = f"phase-aware-telemetry-v3-rep{replicate}" + payload["hardware"]["host_candidates"] = ["dash0"] + payload["engine"]["engine_version"] = "0.24.1.dev3+opprof" + trace = payload["trace"] + trace["trace_file_override"] = str(trace_override) + trace["replay_time_scale"] = REPLAY_TIME_SCALE + trace["early_stop_max_lag_s"] = None + trace["early_stop_max_elapsed_s"] = SAFETY_DEADLINE_S + trace["restart_engine_after_early_stop"] = False + trace["adaptive_stop"] = {"enabled": False} + atomic_json(target, payload) + return target + + +def build_manifest( + *, base_manifest_path: Path, private_root: Path, prior_attempt_state: Path +) -> dict[str, Any]: + base = json.loads(base_manifest_path.read_text(encoding="utf-8")) + if base.get("schema") != "fidelity-prefix-pilot-manifest-v1": + raise ValueError("unexpected base P1 manifest schema") + + repetitions = {} + selection_hashes = [] + selected_ids_by_replicate: dict[int, set[str]] = {} + merged_traces = {} + for replicate, roles in enumerate(REPLICATE_ROLE_PAIRS, start=1): + source_study = Path(base["private"]["studies"][roles[1]]["tp4"]) + source_traces = tuple(resolve_role_trace(base, role) for role in roles) + merged_trace = private_root / "traces" / f"rep{replicate}.jsonl" + merged_traces[str(replicate)] = merge_role_traces( + source_traces, merged_trace + ) + target_study = private_root / "studies" / f"rep{replicate}-tp4.json" + materialize_study( + source_study, + target_study, + replicate=replicate, + trace_override=merged_trace, + ) + study = load_study_spec(target_study) + window, requests = load_trace_requests(study, study_spec_path=target_study) + duration_s = float(window.window_end - window.window_start) + if not math.isclose( + duration_s, EXPECTED_DURATION_S, rel_tol=0.0, abs_tol=1e-9 + ): + raise ValueError( + f"rep{replicate}: replay duration {duration_s} != {EXPECTED_DURATION_S}" + ) + selections = {} + selected_ids_by_level: dict[str, set[str]] = {} + for level, target_rate in LOADS_REQ_S_GPU.items(): + target_count = round(target_rate * duration_s * TP) + anchor, selected = attainable_anchor(requests, target_count) + record = selection_record(selected, duration_s=duration_s) + record.update( + { + "anchor": anchor, + "target_count": target_count, + "target_req_s_per_gpu": target_rate, + } + ) + selections[level] = record + selected_ids_by_level[level] = {request.row_id for request in selected} + selection_hashes.append(record["request_id_order_sha256"]) + selected_ids_by_replicate[replicate] = selected_ids_by_level["mid"] + repetitions[str(replicate)] = { + "source_roles": list(roles), + "merged_trace": merged_traces[str(replicate)], + "study": str(target_study), + "study_sha256": sha256_file(target_study), + "duration_s": duration_s, + "load_order": list(LOAD_ORDERS[replicate]), + "selections": selections, + } + + burnin = base["cells"]["tp4_mns16"]["targets"]["low"]["selections"][ + "burnin" + ] + burnin = dict(burnin) + burnin["study_sha256"] = sha256_file(Path(burnin["study"])) + sessions = [ + { + "session": f"rep{replicate}-mns{mns}", + "replicate": replicate, + "mns": mns, + } + for replicate, mns in SESSION_ORDER + ] + invariants = { + "three_repetitions": len(repetitions) == 3, + "six_sessions": len(sessions) == 6, + "load_levels_two": all( + len(item["selections"]) == 2 for item in repetitions.values() + ), + "selection_hashes_unique": len(selection_hashes) == len(set(selection_hashes)), + "selection_sets_disjoint_across_repetitions": all( + not selected_ids_by_replicate[left] & selected_ids_by_replicate[right] + for left in selected_ids_by_replicate + for right in selected_ids_by_replicate + if left < right + ), + "all_counts_positive": all( + selection["selected_count"] > 0 + for item in repetitions.values() + for selection in item["selections"].values() + ), + } + red_flags = [name for name, passed in invariants.items() if not passed] + prior_state = json.loads(prior_attempt_state.read_text(encoding="utf-8")) + if prior_state.get("status") != "failed": + raise ValueError("prior three-load attempt was not recorded as failed") + prior_h20_hours = float(prior_state["gpu_hours_total"]) + incremental_cap_h20_hours = 8.0 - prior_h20_hours + if incremental_cap_h20_hours < 6.5: + raise ValueError("insufficient global H20-hour budget for the two-load rerun") + return { + "schema": SCHEMA, + "status": "PASS" if not red_flags else "STOP", + "source": { + "base_manifest": str(base_manifest_path), + "base_manifest_sha256": sha256_file(base_manifest_path), + "window_id": base["source"]["window_id"], + "source_trace": base["source"]["trace"], + "source_trace_sha256": base["source"]["trace_sha256"], + "prior_attempt_state": str(prior_attempt_state), + "prior_attempt_state_sha256": sha256_file(prior_attempt_state), + "prior_attempt_h20_hours": prior_h20_hours, + "prior_attempt_failure": prior_state["failures"], + }, + "engine": { + "tp": TP, + "mns_endpoints": list(MNS_ENDPOINTS), + "replay_time_scale": REPLAY_TIME_SCALE, + "duration_s": EXPECTED_DURATION_S, + "safety_deadline_s": SAFETY_DEADLINE_S, + "disable_slo_early_stop": True, + }, + "burnin": burnin, + "private": {"merged_traces": merged_traces}, + "repetitions": repetitions, + "sessions": sessions, + "checkpoints": { + "fractions": [0.1, 0.25, 0.5, 0.75, 1.0], + "seconds": [30.0, 75.0, 150.0, 225.0, 300.0], + }, + "budget": { + "global_hard_cap_h20_hours": 8.0, + "prior_attempt_h20_hours": prior_h20_hours, + "hard_cap_h20_hours": incremental_cap_h20_hours, + "session_estimate_h20_hours": 1.0, + "expected_wall_minutes": [75, 95], + "expected_h20_hours": [4.8, incremental_cap_h20_hours], + }, + "sanity": { + "red_flags": red_flags, + "invariants": invariants, + "selected_sets": len(selection_hashes), + "distinct_selected_sets": len(set(selection_hashes)), + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--base-manifest", type=Path, required=True) + parser.add_argument("--private-root", type=Path, required=True) + parser.add_argument("--public-manifest", type=Path, required=True) + parser.add_argument("--prior-attempt-state", type=Path, required=True) + args = parser.parse_args() + manifest = build_manifest( + base_manifest_path=args.base_manifest, + private_root=args.private_root, + prior_attempt_state=args.prior_attempt_state, + ) + atomic_json(args.public_manifest, manifest) + print( + json.dumps( + { + "status": manifest["status"], + "manifest": str(args.public_manifest), + "sanity": manifest["sanity"], + }, + sort_keys=True, + ) + ) + if manifest["status"] != "PASS": + raise RuntimeError(f"phase-aware pilot preflight failed: {manifest['sanity']}") + + +if __name__ == "__main__": + main() diff --git a/runs/intervention-response-v2/test_analysis.py b/runs/intervention-response-v2/test_analysis.py new file mode 100644 index 0000000..5a0c816 --- /dev/null +++ b/runs/intervention-response-v2/test_analysis.py @@ -0,0 +1,208 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import math +from pathlib import Path +from types import SimpleNamespace + + +HERE = Path(__file__).resolve().parent + + +def load_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_phase_aware_v2", HERE / "analyze_existing.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def load_prepare_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_phase_aware_prepare", HERE / "prepare_pilot.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def load_controller_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_phase_aware_controller", HERE / "pilot_controller.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def load_pilot_analysis_module(): + spec = importlib.util.spec_from_file_location( + "intervention_response_phase_aware_pilot_analysis", HERE / "analyze_pilot.py" + ) + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + spec.loader.exec_module(module) + return module + + +def main() -> None: + module = load_module() + assert module.common_decile_fractions( + trace_duration_s=60.0, minimum_elapsed_s=19.448 + ) == (0.1, 0.2, 0.3) + assert module.common_decile_fractions( + trace_duration_s=60.0, minimum_elapsed_s=60.0 + )[-1] == 1.0 + stats = module.numeric([0.0, 1.0, 2.0]) + assert stats == { + "n": 3, + "min": 0.0, + "max": 2.0, + "distinct_n": 3, + "median": 1.0, + } + assert math.isclose(module._pearson([1.0, 2.0], [2.0, 4.0]), 1.0) + assert module._pearson([1.0, 1.0], [2.0, 3.0]) is None + prepare = load_prepare_module() + requests = [ + SimpleNamespace( + sampling_u=index / 10.0, + row_id=f"r{index}", + arrival_s=float(index), + prompt_tokens_hint=100 + index, + ) + for index in range(1, 6) + ] + _anchor, selected = prepare.attainable_anchor(requests, 3) + assert len(selected) == 3 + record = prepare.selection_record(selected, duration_s=3.0) + assert record["selected_count"] == 3 + assert record["offered_req_s_per_gpu"] == 0.25 + assert len(prepare.SESSION_ORDER) == 6 + assert {mns for _replicate, mns in prepare.SESSION_ORDER} == {16, 64} + first_id = prepare.private_request_id( + source_sha256="a" * 64, line_number=1, original_id="1" + ) + assert first_id == prepare.private_request_id( + source_sha256="a" * 64, line_number=1, original_id="1" + ) + assert first_id != prepare.private_request_id( + source_sha256="b" * 64, line_number=1, original_id="1" + ) + controller = load_controller_module() + assert math.isclose(controller.remaining_projection(6, 0), 6.2) + assert math.isclose(controller.remaining_projection(6, 5), 1.2) + parsed = controller.parser().parse_args( + [ + "--manifest", + "/tmp/manifest.json", + "--run-root", + "/tmp/run", + "--aituner-root", + "/tmp/aituner", + "--vllm-source", + "/tmp/vllm", + "--venv", + "/tmp/venv", + "--model", + "/tmp/model", + "--client", + "/tmp/client.py", + "--dry-run", + ] + ) + assert parsed.dry_run is True + pilot_analysis = load_pilot_analysis_module() + stable = pilot_analysis.stable_adjacent_features( + [ + {"end_fraction": 0.1, "qualifying_response_features": ["queue"]}, + { + "end_fraction": 0.25, + "qualifying_response_features": ["kv", "queue"], + }, + {"end_fraction": 0.5, "qualifying_response_features": ["queue"]}, + ] + ) + assert stable == {"0.10->0.25": ["queue"], "0.25->0.50": ["queue"]} + load_consistency = { + "0.10->0.25:queue": {"passes_two_regimes": True}, + "0.25->0.50:queue": {"passes_two_regimes": True}, + } + mechanism = pilot_analysis.mechanism_gate(stable, load_consistency) + assert mechanism["passes"] is False + stable["0.25->0.50"].append("kv") + load_consistency["0.25->0.50:kv"] = {"passes_two_regimes": True} + mechanism = pilot_analysis.mechanism_gate(stable, load_consistency) + assert mechanism["passes"] is True + assert mechanism["passing_transitions"] == ["0.25->0.50"] + efficacy = pilot_analysis.stable_adjacent_efficacy_features( + [ + { + "end_fraction": 0.1, + "efficacy": {"telemetry_qualifying_features": ["early"]}, + }, + { + "end_fraction": 0.25, + "efficacy": {"telemetry_qualifying_features": ["queue"]}, + }, + { + "end_fraction": 0.5, + "efficacy": {"telemetry_qualifying_features": ["kv", "queue"]}, + }, + ] + ) + assert efficacy == {"0.25->0.50": ["queue"]} + coverage = pilot_analysis.telemetry_coverage( + [ + {"step_index": 1, "submit_mono_ns": 100_000_000}, + {"step_index": 2, "submit_mono_ns": 200_000_000}, + ], + start_ns=0, + end_ns=300_000_000, + ) + assert coverage == { + "start_gap_s": 0.1, + "end_gap_s": 0.1, + "max_internal_gap_s": 0.1, + } + coverage_gate = pilot_analysis.cumulative_coverage_gate( + [ + { + "trial_sanity": [ + { + "trial_id": "a", + "admitted_fraction": 0.25, + "completed_fraction": 0.2, + } + ] + }, + { + "trial_sanity": [ + { + "trial_id": "a", + "admitted_fraction": 0.5, + "completed_fraction": 0.4, + } + ] + }, + ] + ) + assert coverage_gate["red_flags"] == [] + assert controller.warmup_selection( + { + "selections": { + "low": {"offered_req_s_per_gpu": 1.5}, + "mid": {"offered_req_s_per_gpu": 2.125}, + } + } + )["offered_req_s_per_gpu"] == 2.125 + print("phase-aware intervention response v3 analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/opprof-phase6/opprof_phase6_client.py b/runs/opprof-phase6/opprof_phase6_client.py index 11c5e3c..600fb3a 100644 --- a/runs/opprof-phase6/opprof_phase6_client.py +++ b/runs/opprof-phase6/opprof_phase6_client.py @@ -95,7 +95,11 @@ def run_replay(args: argparse.Namespace, *, warmup: bool) -> dict[str, Any]: base_url=args.base_url, timeout_s=study.engine.request_timeout_s, max_concurrency=study.trace.max_concurrency, - target_pass_rate=(0.0 if warmup else study.slo.target_pass_rate), + target_pass_rate=( + 0.0 + if warmup or args.disable_slo_early_stop + else study.slo.target_pass_rate + ), max_lag_s=study.trace.early_stop_max_lag_s, max_elapsed_s=( 120.0 if warmup else _probe_drain_deadline( @@ -122,6 +126,11 @@ def run_replay(args: argparse.Namespace, *, warmup: bool) -> dict[str, Any]: "tpot_ms": outcome.tpot_ms, "completion_tokens": outcome.completion_tokens, "completion_tokens_source": outcome.completion_tokens_source, + "completed_mono_ns": outcome.completed_mono_ns, + "completed_elapsed_s": ( + (outcome.completed_mono_ns - interval_start_mono_ns) / 1e9 + if outcome.completed_mono_ns is not None else None + ), "slo_pass": evaluation.passed, "reasons": evaluation.reasons, "error": outcome.error, @@ -157,6 +166,7 @@ def run_replay(args: argparse.Namespace, *, warmup: bool) -> dict[str, Any]: "feasible": bool(slo_summary["feasible"]), "early_stopped": early_stopped, "early_stop_reason": early_stop_reason, + "slo_early_stop_disabled": bool(args.disable_slo_early_stop), "ttft_ms": numeric([item.ttft_ms for item in outcomes]), "tpot_ms": numeric([item.tpot_ms for item in outcomes]), "invariants": { @@ -235,6 +245,7 @@ def parser() -> argparse.ArgumentParser: q.add_argument("--mns", type=int, required=True) q.add_argument("--base-url", required=True) q.add_argument("--result-dir", required=True) + q.add_argument("--disable-slo-early-stop", action="store_true") return p diff --git a/runs/opprof-phase6/test_phase6_tools.py b/runs/opprof-phase6/test_phase6_tools.py index 0cc338d..6c71c47 100644 --- a/runs/opprof-phase6/test_phase6_tools.py +++ b/runs/opprof-phase6/test_phase6_tools.py @@ -20,6 +20,7 @@ def main() -> None: controller = load("p6controller", HERE / "opprof_phase6_controller.py") solo = load("p6solo", HERE / "opprof_phase6_solo_controller.py") analysis = load("p6analysis", HERE / "analyze_phase6.py") + client = load("p6client", HERE / "opprof_phase6_client.py") assert len(controller.CELLS) == 12 primary = sum(3 if item.get("trap") else 2 for item in controller.CELLS.values()) assert primary == 25 @@ -36,6 +37,18 @@ def main() -> None: assert sum(solo.CELL_ESTIMATE.values()) + solo.SAFETY_HOURS == 3.44 assert solo.next_below([.1, .2, .3], {.2, .3}) == .1 assert solo.next_above([.1, .2, .3], {.1, .2}) == .3 + parsed = client.parser().parse_args([ + "run-anchor", + "--study", "study.json", + "--cell", "tp4_mns16", + "--anchor", "0.5", + "--tp", "4", + "--mns", "16", + "--base-url", "http://127.0.0.1:8000", + "--result-dir", "out", + "--disable-slo-early-stop", + ]) + assert parsed.disable_slo_early_stop is True print("phase6 tools: PASS") diff --git a/runs/simulator-tuning-latency-matrix-v0/experiment-card.md b/runs/simulator-tuning-latency-matrix-v0/experiment-card.md new file mode 100644 index 0000000..72edd6e --- /dev/null +++ b/runs/simulator-tuning-latency-matrix-v0/experiment-card.md @@ -0,0 +1,228 @@ +# EXP-SIM-TUNING-LATENCY-MATRIX-V0：现有 simulator 能否选择真机低延迟 serving config？ + +> **状态：** protocol 已冻结；Qwen3.6 release-runtime topology smoke 已通过（2026-07-17）。 +> 没有 Frontier Qwen3.6 contract/profile、trace replay 或 latency-selection 结果；本 card +> 不授权 600-s matrix launch。 + +## Claim 与决策 + +- **研究问题：** 对冻结的 `(H20 hardware, model, serving engine, request trace, config + candidate surface)`，现有 simulator 能否为一个明确的 latency objective 选出真机最优、或 + 统计上等价的 serving-engine config？ +- **Parent claim：** “simulator 已经解决 config tuning”只有在完整 candidate surface 上产生 + 正确的真机 latency selection 时才成立；绝对 latency error 或单个吞吐/SLO 结果均不充分。 +- **Hypothesis：** simulator 在 fixed-shape / prefill-only 条件下可能足够，但在 decode、真实 + arrival 与 prefix/KV state 同时存在时会因 state-conditioned residual 或 coverage failure 而 + 失去 selection fidelity。 +- **Competing hypothesis：** 经冻结、非 evaluation 的 operator/collective calibration 后， + simulator 能跨 non-MoE hybrid/MoE 与四种 workload condition 稳定选出真机低延迟 config。 +- **判定：** 每个 `model × workload × metric` 是独立 tuning task，且 config 越低 latency 越好。 + 只有所有预注册且 simulator 声称支持的 task 均通过下面的 coverage 与 selection gates，才能 + 写“该 simulator 已解决其声明范围内的 tuning”；任一失败 cell 是明确 boundary，不被平均值 + 掩盖。 + +## 固定系统与模型覆盖 + +| 角色 | 状态 | 运行时契约 | 当前可复用 artifact / 含义 | +|---|---|---|---| +| MoE | **可开始准备** | Qwen3-30B-A3B，BF16，community vLLM 0.20.0，dash0 H20；Frontier/Vidur-class fork | `frontier-phase-factorial-v0` 有 `ISL=2048, OSL=1, r=4` 的 request-level real/sim pilot；`frontier-fidelity-envelope-v1` 有 fixed-PD 与 exact Trace-PD materializers。历史 pilot 不是本 card 的最终 600-s evaluation。 | +| Non-MoE hybrid (not all-attention dense) | **TP1/2/4 smoke passed; profile pending** | `Qwen/Qwen3.6-27B@cea40373b9214dd387123e68841890af30dcd469`，community release `vLLM 0.20.2` / Torch `2.11.0` / Transformers `5.14.1`，dash0 H20；Frontier profile 必须在此 exact stack 采集 | Exact release serves text-only 2048→1 at TP1/2/4. vLLM resolves a 784-token attention page for its GDN/Mamba state, so requested block=16 is not the physical engine state. Qwen3.6 trace, Frontier hybrid model contract, and profile remain missing. | + +所有公开结果固定在单机 `dash0`（8×H20）；每个 model 使用相同 engine build 的 real +server 与 simulator contract。Qwen3.6 bring-up 已确认真实 runtime/page legality；其余最低成本 +gate 是冻结 trace/tokenizer 与完成 Frontier hybrid profile-schema audit。若不能做到，论文结论 +必须限于 MoE，不得写跨模型 generalization。 + +### Qwen3.6 prerequisite audit (runtime + smoke, 2026-07-17) + +| Prerequisite | Evidence | Verdict | Decision | +|---|---|---|---| +| Exact model snapshot | Immutable ModelScope snapshot `Qwen/Qwen3.6-27B@cea40373b9214dd387123e68841890af30dcd469` is staged at `/home/admin/cpfs/wjh/models/Qwen/Qwen3.6-27B`; its sorted-file inventory digest is `031cea103726cbf27732b6971e898e210aee77f0c32e4bc703c947e0372349c4`. | **PASS** | Bind every Qwen3.6 trace, real server, and profile manifest to this revision and inventory. | +| Community vLLM runtime | Clean environment `/home/admin/cpfs/wjh/venvs/qwen36-vllm-0.20.2` contains official wheel `vLLM 0.20.2`, Torch `2.11.0`, and Transformers `5.14.1`; `pip check` passes and CPU-only `get_config(..., trust_remote_code=false)` resolves `Qwen3_5ForConditionalGeneration`. | **PASS (CPU)** | Use exactly this release environment for real serving and profile collection. | +| Topology compatibility | Fresh TP1/2/4 servers each completed a text-only 2048-prompt/1-completion request with exact usage; all GPUs were released after cleanup. Every topology resolved the attention page from requested 16 to 784 tokens for Mamba/GDN state. | **PASS (GPU smoke)** | This is an engine-configuration legality result, not a latency rank or simulator result. Profile/simulator input must expose the 784-token state/page contract. | +| Qwen3.6 Frontier model/profile | No local or remote Qwen3.6-named Frontier profile/model directory was found | **MISSING** | New dense profile root is required; Qwen30/MoE profile CSVs are forbidden inputs. | +| Old dense artifacts | Existing Qwen3.5-27B trace is internal-vLLM/hybrid/speculative | **N/A as baseline** | Keep only as historical context; never feed it to this matrix or profile calibration. | + +This audit is not a failed simulator result. The topology gate is closed, but the model contract, +profile root, and frozen trace are still reproducibility blockers: the Qwen3.6 half is `NOT RUN`, +not `FAIL`. For a paper claim about a conventional all-attention dense transformer, Qwen3.6 is not +admissible evidence; it is only a non-MoE hybrid case. + +## Workload matrix + +每个 model 都运行四个固定 trace。`P` 是 PD disaggregation 的 P-node workload，不是缺失 +decode 的坏数据；它的 `OSL=1` 是设计变量。固定 cases 与 trace cases都使用 open-loop replay， +600 秒 observation window；`Fixed` 的 request count 为 `ceil(QPS × 600)`。 + +| ID | trace contract | output contract | prefix/KV contract | 可评 latency objectives | +|---|---|---|---|---| +| Fixed-P | 固定 `ISL=2048`、uniform fixed QPS=`4 req/s`，token vector deterministic | `OSL=1` | no shared logical prefix；prefix cache off | mean/p90 TTFT；mean/p90 E2E；TPOT=`N/A` | +| Fixed-PD | 固定 `ISL=2048`、uniform fixed QPS=`4 req/s` | `OSL=128`，`min=max=128`，`ignore_eos=true` | no shared logical prefix；prefix cache off | mean/p90 TTFT、TPOT、E2E | +| Trace-P | held-out production trace 的逐 request prompt-token vector、arrival time、request order、session root 与 prefix block relation 原样保留 | **只将 replay output 强制为 `OSL=1`** | source prefix equivalence、warm/cold state、cache policy 原样保留 | mean/p90 TTFT；mean/p90 E2E；TPOT=`N/A` | +| Trace-PD | held-out production trace 的逐 request input/output lengths、arrival time、order、session root 与 prefix block relation 原样保留 | 原始 `min=max=source OSL`，`ignore_eos=true` | source prefix equivalence、warm/cold state、cache policy 原样保留 | mean/p90 TTFT、TPOT、E2E | + +`Trace-P` 应明确称为 **trace-derived prefill-only**：它忠实保留输入、到达、会话与 cache +state，但不忠实保留原始 output-length distribution；不能被描述成完整 trace-faithful replay。 +`Trace-PD` 才是完整 input/output/arrival/prefix-KV joint replay。 + +### Trace、correctness 与 cache 的不可变契约 + +- 每个 trace manifest 列出：request ID、arrival timestamp、input/output token counts、session + ID、runtime block identities、trace vector SHA-256、source trace SHA-256、requested and + resolved physical page/block size、cache policy、initial cache state、warmup trace digest 与 + model tokenizer/version。 +- 当 simulator 要模拟 vLLM prefix caching 时，`block_hash_ids` 只能描述可复用的**完整** + runtime blocks：adapter 必须导出 `floor(ISL / block_size)` 个 prompt identities，不能将 + `ceil` 得到的 partial final block 当作 cache hit。该 simulator-facing projection 有独立 + digest；private real replay 仍保留完整 prompt 与原始 token vector。 +- `Trace-P` 产生单独 manifest：source OSL 和 replay OSL 都写入 digest；只允许 source + OSL → 1 这一个变换。`Trace-PD`、real client 与 simulator CSV 的 row-vector digest 必须完全 + 相同。 +- 真实端逐 request 核验 prompt/completion usage 与 manifest 一致；simulator 核验 request ID、 + input/output shapes、完成数与 trace digest。一项不符即该 cell 的 metric 无效。 +- 每个 `(config, workload, trial)` fresh server；server warmup 用独立 manifest，warmup request + 不进入指标，也不得改变 evaluation cache 初态。prefix-on trace 以同一明示 initial cache + state重放，不能用随机前史替代。 + +## Candidate-surface 与 baseline 规则 + +1. 对每个 model 冻结 `C = TP∈{1,2,4} × MNS∈{8,16,32,64}`（12 个 configs），并固定 + `DP=PP=EP=1`、MBT=8192、chunked prefill on。每个 candidate 还必须冻结 **真实 engine + resolved** page/block and state-cache contract；requested CLI `--block-size 16` 不能替代它。 + Qwen3.6 smoke 在所有 TP 都将其改写为 784 tokens。任何不被某 model/engine 静态支持的 + config 要在运行前从该 model 的 `C` 整体删除，并记录理由；不能看结果后删除。 +2. `C_real-valid` 是 server launch 成功、所有三个 fresh trials 全部请求完成且 usage/hash 合法的 + configs。真机 OOM、timeout 或 request error 保留为 real infeasible/safety outcome，而非给它 + 赋无限 latency。 +3. 对每个 simulator，selection coverage 必须涵盖全部 `C_real-valid`。crash、scheduler stall、 + trace mismatch、缺 request metrics 都是 simulator **coverage failure**，不可当成高 latency 或 + infeasible 来产生排序。只有完整 coverage 才能形成 tuning verdict。 +4. Frontier、官方 Vidur 与 APEX 分别独立报告版本、patch、profile 与 capability surface。APEX + 若只支持 DP/PP/TP parallel plan，则它只评 `parallel-plan tuning` 交集；不得用该子问题结果 + 宣称解决 MNS/MBT/cache 等完整 engine tuning。 + +## Metrics、统计与 selection score + +每个完整 request 使用同一 client-boundary 定义： + +- `TTFT`：客户端发出请求至收到第一个 generated token；包含 serving queue。 +- `E2E`：客户端发出请求至完成 `[DONE]`。 +- `TPOT`：`(E2E - TTFT) / (completion_tokens - 1)`；只对 `completion_tokens > 1` 的请求定义。 +- 每 trial 的 `mean` 是 request-level arithmetic mean；`p90` 是 nearest-rank + `ceil(0.90 × N)`。P-only 的 TPOT 是 JSON `null` / table `N/A`，绝不写 0。 + +同一 config 的 real 运行三次 fresh-server trial，随机化 config order；每个 trial 都对整个 +trace 计算六项或四项指标。real ranking point estimate 是三个 trial metric 的 median，且报告 +pooled request statistic、每-trial statistic 和 95% hierarchical bootstrap interval（先重采样 +trial，再在 trial 内重采样 request）。simulator 的 seed、determinism 和 repeat policy同样记录。 + +对每一个 objective，输出：absolute real/sim surface、real bootstrap top set、simulator exact +top set、top-set hit、sim-chosen config 的 best/worst tie-break regret、Kendall τ-b、non-tied +pair direction accuracy、coverage 与 error/infeasible table。Latency regret 定义为 +`(real_metric(sim_choice) - real_metric(real_best)) / real_metric(real_best)`。 + +## 预注册 decision gates + +| Gate | Pass | Failure handling | +|---|---|---| +| Trace/metric validity | 100% planned requests have matching usage/shape and finite applicable metrics | cell invalid；诊断 harness，不作 performance conclusion | +| Simulator coverage | 100% of `C_real-valid` output legal request metrics | `FAIL: coverage`；不把 missing cell 排成最差 | +| Selection | simulator top set 与 bootstrap real top set 有交集 | `FAIL: selection` | +| Risk | simulator top-set 的 **worst** real tie-break regret `≤5%` | `FAIL: regret`，即使 best tie-break 碰巧为 0 | +| Global rank diagnostic | τ-b `≥0.8`；若没有非并列 pair 则写 `N/A`，不当通过 | `NEEDS EVIDENCE` / `FAIL` 与 top-set gate 分开报告 | + +这里不使用 SLO、capacity 或 throughput 作 selection objective；它们最多作为安全/诊断附录指标。 +P-only 共 4 tasks，PD 共 6 tasks，所以完整 `non-MoE-hybrid/MoE × 4 workloads` matrix 共有 40 个 +独立 latency-selection tasks。 + +## Calibration 与 information boundary + +- 允许 simulator 使用同硬件/同 model/同 runtime 的 offline operator、collective 或 memory + profiles，但 profile command、shape grid、cost 和 hash 必须在 evaluation 前冻结。 +- calibration 不得读取 evaluation 的 real request latency、top config 或 E2E scalar；禁止 per-TP + 或 per-config serving E2E scale。若有 trace-driven calibration，必须只使用按 session root 切分 + 的 calibration fold；Trace-P/PD 使用完全不相交的 held-out root sessions。 +- Fixed traces 与 trace evaluation 都不能同时作为 calibration workload。任何 required simulator + patch 在查看该 cell real result 前冻结；patched baseline 与 original release 分开作图。 + +## Minimal Qwen3.6-27B bring-up and profile plan + +This is the only permitted path to add the dense case. It intentionally separates cheap provenance +work from GPU work, and it does not authorize either stage by itself. + +1. **Preflight and topology smoke (complete):** the snapshot/runtime are locked; fresh TP1/2/4 + servers completed exact-usage text requests. The actual vLLM state/page setting is 784 tokens, + not the requested 16. This establishes the profile input contract only. +2. **Hybrid Frontier contract (not run):** verify Frontier can represent Gated DeltaNet/linear-state, + gated attention, FFN, true mixed execution, TP collectives, the 784-token page rule and capacity, + without substituting Qwen3.5/Qwen30 fields. +3. **Minimal profile set (GPU, separately approved):** on the exact Qwen3.6/community-vLLM/dash0 + stack, collect per-TP Gated-DeltaNet, gated-attention prefill/decode, state/KV and TP2/TP4 all-reduce + rows. The shape grid must cover the pre-registered Fixed-P/PD and held-out Trace-P/PD ranges, but + must not contain evaluation E2E latency or a fitted time scale. Freeze CSV/schema/command hashes + before running any evaluation trace. +4. **Profile acceptance (CPU):** verify model/runtime identity in every profile manifest; require + requested shape coverage, finite values, and no fallback to a MoE or Qwen3.5 profile. Only then may + a parent-approved real/simulator latency surface be scheduled. + +## 预期产物、复现与成本 + +- **Figure prototype：** [latency-selection-matrix-schematic.svg](latency-selection-matrix-schematic.svg) + （明确标注 `SCHEMATIC — NOT MEASURED DATA`）。最终图替换为每个 task 的 top-set hit / regret / + coverage heatmap，并附 per-config latency surfaces。 +- **Raw layout：** `artifacts/////trial-*/`；每个目录含 + command、environment/machine fingerprint、server flags、trace/profile hash、raw request records、 + stdout/stderr、duration 与 SHA-256 inventory。 +- **Existing reuse audit：** MoE F0 可检验 P-only raw metric reader；现有 F2 可检验 fixed-PD + trace materialization；T1 可检验 Trace-PD digest and simulator coverage reporting。它们并不构成 +这个 600-s no-SLO matrix 的完整 real surface。Qwen3.6 requires a new same-stack trace, + model contract and profile root; old Qwen3.5 internal-vLLM artifacts are excluded. +- **Cost estimate, not authorization：** 单 config×trial 假定 `10 min replay + 5 min fresh-server + startup/warmup = 15 min`。完整 12-config sweep 的 GPU allocation 为 + `4 × (TP1 + TP2 + TP4) = 28 GPU`; 因而每个 `model×workload` 的三 trial 约 + `28 × 0.25 × 3 = 21 H20-GPUh`。8 cells 的 real serving core is **168 H20-GPUh**. The Qwen3.6 + bring-up/profile set adds at most **6 H20-GPUh** (three topology smokes plus dense/attention/ + collective rows), giving a **174 H20-GPUh** core estimate and a **190–215 H20-GPUh** planning + envelope with retries. Every launch must recompute this from the staged trace duration and actual + server preflight; this paragraph is not authorization. + +### Required remote launch echo (not executed) + +```text +NOT EXECUTED — awaiting explicit approval after Qwen3.6 model/runtime/profile gates +host=dash0; remote_root=/home/admin/cpfs/wjh/aituner/aituner +models=non-moe-hybrid:Qwen/Qwen3.6-27B@,moe:Qwen3-30B-A3B-BF16 +engine=community-open-source-vLLM@; simulator=Frontier/Vidur-class, +workloads=Fixed-P(ISL2048,OSL1,QPS4,600s),Fixed-PD(ISL2048,OSL128,QPS4,600s), + Trace-P(heldout exact input/arrival/prefix,OSL1),Trace-PD(heldout exact joint trace) +surface=TP{1,2,4}×MNS{8,16,32,64}; trials=3; fresh_server=true +real_paths=runs/simulator-tuning-latency-matrix-v0/artifacts///real/ +expected_cost=174 H20-GPUh core; 190–215 H20-GPUh planning envelope; expected_wall=multiple queued hours +gate=Qwen3.6 snapshot + community-vLLM fingerprint + resolved-784 hybrid Frontier profile completeness + trace/profile hashes + user/parent approval +``` + +## Expected outcomes and stop/pivot rules + +- If a simulator passes every supported task, stop diagnosing that simulator and look for a new problem + outside its evaluated scope (e.g., unsupported engine knobs or alignment cost). +- If it fails coverage, repair/liveness is a separate intervention; original capability result remains + frozen. First rerun simulator only; do not spend GPU-hours until it can produce a complete surface. +- If it covers but selects incorrectly, retain the counterexample and run one minimal discriminative + mechanism experiment (scheduler state, batch composition, graph/kernel family, or MoE/collective), + changing one factor at a time. +- If Qwen3.6 cannot be profiled on its exact community-vLLM stack, stop the non-MoE-hybrid half rather + than treating any internal-vLLM or Qwen3.5 result as a substitute. If the headline requires a pure + all-attention dense-transformer case, add that model explicitly rather than relabelling Qwen3.6. + +## Benchmark-design audit + +| Risk | Pre-registered control | +|---|---| +| SLO/capacity result misrepresented as tuning | selection uses only per-request mean/p90 TTFT, TPOT, E2E on the same trace | +| P-only judged by nonexistent TPOT | OSL=1 and TPOT=`N/A` are explicit workload/metric contracts | +| Trace simplification hidden | Trace-P is labelled output-normalized; only Trace-PD claims full joint replay | +| calibration equals evaluation | microprofile/trace calibration folds are disjoint; no E2E correction on evaluation | +| simulator crash treated as a bad config | coverage gate separates missing metrics from latency ranking | +| cross-stack or wrong-architecture reuse | only Qwen3.6-27B hybrid profiles collected on its exact community-vLLM stack are admissible; old Qwen3.5/internal-vLLM or Qwen30/MoE artifacts are excluded | +| selected winning cells only | all 8 cells and all legal configs are frozen before real results | +| tail/noise omitted | p90, three fresh trials, per-trial values and bootstrap top sets are mandatory | diff --git a/runs/simulator-tuning-latency-matrix-v0/materialize_qwen30_tp_normalized_trace.py b/runs/simulator-tuning-latency-matrix-v0/materialize_qwen30_tp_normalized_trace.py new file mode 100644 index 0000000..cc7fe98 --- /dev/null +++ b/runs/simulator-tuning-latency-matrix-v0/materialize_qwen30_tp_normalized_trace.py @@ -0,0 +1,240 @@ +#!/usr/bin/env python3 +"""Derive a per-GPU-normalized real/Frontier Qwen30 trace pair. + +Arrival times change as ``t' = t / TP``. The private JSONL preserves the +complete prompt identity vector for real replay; the Frontier CSV exports only +the completed 16-token prompt blocks that vLLM can legally reuse from prefix +cache. Prompt bodies remain exclusively in the private output. +""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +from pathlib import Path +from typing import Any + + +CSV_FIELDS = ( + "arrived_at", + "num_prefill_tokens", + "num_decode_tokens", + "session_id", + "block_hash_ids", +) + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def update_digest(digest: Any, values: list[Any]) -> None: + digest.update(json.dumps(values, separators=(",", ":")).encode()) + digest.update(b"\n") + + +def full_prompt_block_ids( + runtime_block_ids: list[int], input_tokens: int, block_size: int +) -> list[int]: + """Project exact prompt identities to vLLM-cacheable full blocks only.""" + if input_tokens <= 0 or block_size <= 0: + raise ValueError("input_tokens and block_size must be positive") + expected_runtime_blocks = math.ceil(input_tokens / block_size) + if len(runtime_block_ids) != expected_runtime_blocks: + raise ValueError( + "runtime block identity count does not match the exact prompt: " + f"got {len(runtime_block_ids)}, expected {expected_runtime_blocks}" + ) + return runtime_block_ids[: input_tokens // block_size] + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--base-public-csv", type=Path, required=True) + parser.add_argument("--base-private-jsonl", type=Path, required=True) + parser.add_argument("--tp", type=int, required=True) + parser.add_argument("--output-root", type=Path, required=True) + parser.add_argument("--base-window-seconds", type=float, default=600.0) + parser.add_argument("--runtime-block-size", type=int, default=16) + return parser.parse_args() + + +def main() -> None: + args = parse_args() + if ( + args.tp <= 0 + or args.runtime_block_size <= 0 + or not math.isfinite(args.base_window_seconds) + or args.base_window_seconds <= 0 + ): + raise ValueError("TP, runtime block size, and base window must be positive") + + public_rows = list(csv.DictReader(args.base_public_csv.open(newline=""))) + private_rows = [ + json.loads(line) for line in args.base_private_jsonl.open() if line.strip() + ] + if not public_rows or len(public_rows) != len(private_rows): + raise ValueError("public/private request counts must match and be non-zero") + if set(public_rows[0]) != set(CSV_FIELDS): + raise ValueError("unexpected public trace schema") + + normalized_digest = hashlib.sha256() + semantic_digest = hashlib.sha256() + normalized_private: list[dict[str, Any]] = [] + normalized_public: list[dict[str, str]] = [] + original_arrivals: list[float] = [] + frontier_prefix_digest = hashlib.sha256() + partial_prompt_count = 0 + raw_runtime_block_count = 0 + frontier_full_block_count = 0 + + for index, (public, private) in enumerate( + zip(public_rows, private_rows, strict=True) + ): + original_arrival = float(public["arrived_at"]) + if not math.isfinite(original_arrival) or original_arrival < 0: + raise ValueError(f"invalid arrival at row {index}") + if not math.isclose(original_arrival, float(private["arrived_at"]), abs_tol=1e-9): + raise ValueError(f"public/private arrival mismatch at row {index}") + runtime_block_ids = [int(value) for value in private["runtime_block_ids"]] + expected = ( + int(public["num_prefill_tokens"]), + int(public["num_decode_tokens"]), + int(public["session_id"]), + public["block_hash_ids"], + ) + actual = ( + int(private["input_length"]), + int(private["output_length"]), + int(private["session_id"]), + "|".join(str(value) for value in runtime_block_ids), + ) + if expected != actual: + raise ValueError(f"public/private payload mismatch at row {index}") + + input_tokens = int(private["input_length"]) + frontier_block_ids = full_prompt_block_ids( + runtime_block_ids, input_tokens, args.runtime_block_size + ) + if not frontier_block_ids: + raise ValueError( + "Frontier prefix-cache replay requires at least one complete " + f"runtime block; request {index} has ISL={input_tokens}" + ) + partial_prompt_count += int(input_tokens % args.runtime_block_size != 0) + raw_runtime_block_count += len(runtime_block_ids) + frontier_full_block_count += len(frontier_block_ids) + + normalized_arrival = original_arrival / args.tp + updated = dict(private) + updated["arrived_at"] = normalized_arrival + normalized_private.append(updated) + normalized_public.append( + { + "arrived_at": f"{normalized_arrival:.12f}", + "num_prefill_tokens": public["num_prefill_tokens"], + "num_decode_tokens": public["num_decode_tokens"], + "session_id": public["session_id"], + "block_hash_ids": "|".join(str(value) for value in frontier_block_ids), + } + ) + update_digest( + semantic_digest, + [ + int(private["source_index"]), + int(private["input_length"]), + int(private["output_length"]), + int(private["session_id"]), + runtime_block_ids, + ], + ) + update_digest( + normalized_digest, + [ + int(private["source_index"]), + normalized_arrival, + int(private["input_length"]), + int(private["output_length"]), + int(private["session_id"]), + runtime_block_ids, + ], + ) + update_digest( + frontier_prefix_digest, + [ + int(private["source_index"]), + normalized_arrival, + input_tokens, + frontier_block_ids, + ], + ) + original_arrivals.append(original_arrival) + + if any( + right < left for left, right in zip(original_arrivals, original_arrivals[1:]) + ): + raise ValueError("base arrival order is not monotonic") + + public_dir = args.output_root / "public" + private_dir = args.output_root / "private" + public_dir.mkdir(parents=True, exist_ok=True) + private_dir.mkdir(parents=True, exist_ok=True) + public_path = public_dir / "frontier.csv" + private_path = private_dir / "real_requests.jsonl" + with public_path.open("w", newline="") as output: + writer = csv.DictWriter(output, fieldnames=CSV_FIELDS, lineterminator="\n") + writer.writeheader() + writer.writerows(normalized_public) + with private_path.open("w") as output: + for row in normalized_private: + output.write(json.dumps(row, separators=(",", ":")) + "\n") + + manifest = { + "schema": "qwen30-tp-normalized-trace-v2", + "privacy": "prompt text exists only under private/ and must not be harvested", + "transform": "arrival_seconds_prime = arrival_seconds / tensor_parallel_size", + "tensor_parallel_size": args.tp, + "requests": len(normalized_private), + "base_window_seconds": args.base_window_seconds, + "global_offered_request_rate": len(normalized_private) + / args.base_window_seconds + * args.tp, + "per_gpu_offered_request_rate": len(normalized_private) + / args.base_window_seconds, + "frontier_prefix_block_projection": { + "runtime_block_size": args.runtime_block_size, + "rule": "full_prompt_blocks_only=floor(input_tokens/runtime_block_size)", + "partial_prompt_count": partial_prompt_count, + "raw_runtime_block_count": raw_runtime_block_count, + "frontier_full_block_count": frontier_full_block_count, + }, + "original_first_arrival_s": original_arrivals[0], + "original_last_arrival_s": original_arrivals[-1], + "normalized_first_arrival_s": original_arrivals[0] / args.tp, + "normalized_last_arrival_s": original_arrivals[-1] / args.tp, + "base_public_csv": str(args.base_public_csv.resolve()), + "base_public_csv_sha256": sha256(args.base_public_csv), + "base_private_jsonl": str(args.base_private_jsonl.resolve()), + "base_private_jsonl_sha256": sha256(args.base_private_jsonl), + "public_csv": str(public_path.resolve()), + "public_csv_sha256": sha256(public_path), + "private_jsonl": str(private_path.resolve()), + "private_jsonl_sha256": sha256(private_path), + "semantic_vector_sha256": semantic_digest.hexdigest(), + "normalized_row_vector_sha256": normalized_digest.hexdigest(), + "frontier_prefix_vector_sha256": frontier_prefix_digest.hexdigest(), + } + manifest_path = public_dir / "manifest.json" + manifest_path.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n") + print(json.dumps(manifest, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/simulator-tuning-latency-matrix-v0/qwen30-tracepd-u0p01-card.md b/runs/simulator-tuning-latency-matrix-v0/qwen30-tracepd-u0p01-card.md new file mode 100644 index 0000000..9d768b8 --- /dev/null +++ b/runs/simulator-tuning-latency-matrix-v0/qwen30-tracepd-u0p01-card.md @@ -0,0 +1,221 @@ +# Qwen3-30B-A3B Trace-PD `u0p01` latency-selection comparison + +> **Status:** the 36-run real surface is valid. The initially reported Frontier +> surface is **invalidated for simulator fidelity/selection** by a post-hoc +> prefix-cache trace-contract audit (2026-07-17): its simulator-facing block +> metadata includes non-cacheable partial final blocks. The raw CPU artifacts +> are retained for diagnosis, but are not Frontier evidence. This remains the +> first MoE case of `EXP-SIM-TUNING-LATENCY-MATRIX-V0`, not a claim about the +> remaining three workload classes. + +## Question and decision + +Can the frozen Frontier surface select the same low-latency Qwen3-30B-A3B +configuration as a fresh community-vLLM deployment when both replay exactly +the same trace-derived prefill+decode request vector? The objective contains +no SLO or capacity threshold: lower mean/p90 TTFT, TPOT, and E2E is better. + +The result is a bounded Trace-PD result only. It cannot establish a claim +about Fixed-P, Fixed-PD, Trace-P, other trace rates, another model, or a +modified Frontier. + +## Frozen contract + +| Item | Value | +|---|---| +| Hardware | `dash0`, H20, one replica; fresh vLLM server per config/trial | +| Model / engine | `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`; community `vLLM 0.20.0+cu129`, BF16 | +| Config surface | `TP ∈ {1,2,4} × MNS ∈ {8,16,32,64}`; `DP=PP=EP=1`, `MBT=8192`, GPU-memory utilization `0.92`, chunked prefill and prefix caching on | +| Trace | `trace-exact-v1`, anchor `u0p01`; **129 requests**, original input/output length, order and session are preserved. The first Frontier materialization also copied every 16-token runtime identity, including a final partial block; that is not a legal vLLM prefix-cache-hit representation and invalidates this simulator attempt. For a config of tensor parallelism `TP`, every arrival is transformed as `t′ = t / TP`; the TP1/2/4 traces therefore end at 597.037/298.519/149.259 s. | +| Offered load | Base rate is `129 / 600 = 0.215 req/s`. TP1/2/4 replay at 0.215/0.430/0.860 req/s globally, respectively, so all three topologies receive 0.215 req/s/GPU. This is a **per-GPU-normalized trace**, not an unchanged-wall-clock trace. | +| Simulator baseline | The existing same-global-rate T1 surface is excluded: it evaluates TP2/TP4 at lower per-GPU load. A new Frontier surface must use the identical TP-normalized arrival vector for each topology before coverage or latency can be reported. | +| Trials | Three fresh-server trials per config; independent warmup excluded from the measured trace | + +## Validity gates + +1. Each real trial must complete all 129 requests, preserve exact prompt and + completion-token usage, and have the TP-normalized row-vector digest. +2. A simulator metric is valid only if its cell has all 129 finite request + metrics with matching request ID, prompt/output lengths and transformed + arrival vector. A simulator `scheduler_stall` is a coverage result, never + an infeasible-latency label. +3. TTFT, TPOT and E2E are kept in milliseconds. `TPOT` is recorded only for + outputs longer than one token (all requests in this Trace-PD anchor + qualify). +4. A failed real run remains an infeasible/safety observation; it is not + silently converted to infinite latency or omitted from ranking. + +## Analysis and decision rule + +For every `config × trial`, retain raw per-request records and compute mean +and nearest-rank p90 for TTFT, TPOT and E2E. Report pooled 387-request values +alongside the three per-trial values. A selection-stability bootstrap is +deferred; it is not needed to evaluate Frontier's coverage gate. + +The primary gate is Frontier decision-valid coverage on all 12 cells. If any +real-valid candidate lacks a legal simulator metric, Frontier receives no +full-surface selection, top-set overlap, regret, or Kendall-tau score. We +will still show all valid per-cell latency comparisons and the complete real +surface, so any coverage failure and its ground truth remain auditable. + +The simulator's pre-existing SLO fields do not influence request generation, +scheduling, or this no-SLO analysis; they are excluded from scoring. + +## Post-hoc Frontier trace-contract audit (2026-07-17) + +The CPU attempt initially passed only a syntactic trace check (row count, +IDs, lengths, and SHA). It did **not** validate the semantic invariant needed +by vLLM prefix caching: only completed 16-token blocks may be cache hits. + +- The trace exporter emitted `ceil(ISL / 16)` identities. Thus 122/129 prompts + contained one partial final block; the public trace has 36,443 identities, + while a full-block Frontier/vLLM cache adapter must expose 36,321. +- Exact community-vLLM 0.20 source (`vllm.v1.core.kv_cache_manager`) documents + that computed prefix blocks "must be full" and bounds hits by + `prompt_length - 1`. Frontier's fixture contract only requires a non-empty + `block_hash_ids` field and does not validate this full-block invariant. +- Frontier's prefix manager converts every hit to `len(blocks) * 16`. For the + final TP2/MNS16 state, requests 89/111/117 had `(ISL, IDs)` of + `(722,46)`, `(678,43)`, and `(706,45)`. If all supplied identities hit, it + obtains 736/688/720 computed tokens and hence `num_new_tokens` of + -14/-10/-14. The scheduler preserves zero-or-negative-token requests in its + waiting queue; no event remains, so the simulator reports + `Sequential simulation ended with non-empty scheduler state`. + +This proves a trace-adapter/Frontier prefix-cache semantic mismatch, not a +GPU capacity failure: a targeted CPU hook found no failed KV allocation and +no PP-admission deferral, only the three non-admitted requests with a full +8192-token batch budget. The correct next input is a **separate Frontier +adapter** that exports only `floor(ISL / 16)` complete prompt-block identities; +it must not modify the private real trace or its input/output/arrival vector. + +## Invalid raw Frontier attempt (2026-07-17; retained for diagnosis only) + +The TP-normalized surface completed in 156 seconds of CPU-only wall time on +`dash0`, with Frontier commit `deadc4a321f0baaa534c6ebd17f974123733cdc2`. +All twelve input traces passed the 129-request, ID/shape, and TP-specific +trace-SHA checks. Only two cells reached valid request metrics; the other ten +ended in `scheduler_stall`. + +| Config | Status | mean / p90 TTFT (ms) | mean / p90 TPOT (ms) | mean / p90 E2E (ms) | +|---|---|---:|---:|---:| +| TP1/MNS32 | complete | 440032 / 972628 | 145.8 / 151.5 | 936301 / 1645895 | +| TP1/MNS64 | complete | 174335 / 510962 | 163.6 / 178.9 | 724328 / 1285374 | +| remaining 10 configs | `scheduler_stall` | N/A | N/A | N/A | + +The apparent **2/12 (16.7%)** coverage is not a Frontier coverage result: +both the ten stalls and the two completed cells consumed the invalid partial +block metadata. In particular, the two completed cells may receive false +prefix hits. No latency, ranking, coverage, top-set overlap, regret, or +Kendall-tau statement about Frontier may be derived from this attempt. + +The legacy surface runner writes an offered-rate summary assuming a fixed +600-second window. That field is not used here because TP2/TP4 have compressed +arrival clocks. The materialized trace manifests are authoritative: they +record global 0.215/0.430/0.860 req/s and the matched 0.215 req/s/GPU rate. + +## Corrected-adapter CPU liveness probe (2026-07-17) + +One deliberately narrow CPU-only rerun tested the causal fix before launching +another surface: TP2/MNS16, the same 129-request Trace-PD input, MBT=8192, +prefix caching and chunked prefill enabled, and the original Frontier commit. +It changed only the simulator-facing prefix projection from +`ceil(ISL / 16)` to `floor(ISL / 16)`; the private real replay remained bytewise +unchanged. The materialized trace records 36,321 complete blocks (36,443 raw +runtime identities; 122 prompts with a discarded partial final block), global +arrival rate 0.430 req/s, and the required 0.215 req/s/GPU. + +The run completed **129/129** requests in 105.8 CPU wall-clock seconds with no +stall. This is a direct liveness validation of the diagnosis above: the +partial-block adapter, rather than KV capacity, caused the old deadlock. + +It is **not** a valid latency or selection result. Even after the repair, +Frontier reports p50 TTFT=946,973 ms, p50 TPOT=96.37 ms, p50 E2E=1,305,128 ms, +and 112.2 generated decode tokens/s. The trace supplies 460,490 output tokens +over a 298.5-s TP2 arrival horizon (about 1,543 output tokens/s), so the +simulated service rate is below the offered decode work and its queue must +grow. The corresponding real TP2/MNS16 replay has mean/p90 TPOT 14.1/16.1 ms +and mean/p90 TTFT 34,443/77,922 ms. Thus the post-fix discrepancy is a +decode-service-model / runtime-semantics question, not residual evidence from +the invalid cache metadata. The probe artifact is +`/home/admin/cpfs/wjh/aituner/simulator-tuning-latency-q30-tp-normalized-u0p01-20260717/frontier-contract-probe-v1`. + +## Real result (2026-07-17) + +All **36/36** `config × trial` replays completed with exit code zero. Each +trial passed all validity gates: 129/129 successful requests, exact requested +and observed input/output token usage, the TP-specific private-trace SHA and +normalized arrival-vector SHA, and HTTP routing to the explicit alias +`qwen3-30b-exact-trace`. The cache-populating smoke and earlier failed launch +attempts are excluded from this table. Values pool 387 requests/config (three +trials); p90 is nearest-rank. + +| Config | mean / p90 TTFT (ms) | mean / p90 TPOT (ms) | mean / p90 E2E (ms) | +|---|---:|---:|---:| +| TP1/MNS8 | 88479.9 / 168185.5 | 14.3 / 16.3 | 137996.4 / 222136.0 | +| TP1/MNS16 | 17378.2 / 42739.8 | 20.6 / 25.3 | 87355.2 / 156177.2 | +| TP1/MNS32 | 620.5 / 1931.5 | 23.8 / 30.0 | 81051.2 / 162908.6 | +| TP1/MNS64 | 619.8 / 1956.7 | 24.1 / 30.4 | 82253.7 / 165355.3 | +| TP2/MNS8 | 95920.1 / 183878.8 | 9.2 / 10.2 | 128296.9 / 215848.3 | +| TP2/MNS16 | 34443.0 / 77922.4 | 14.1 / 16.1 | 82936.7 / 137726.4 | +| TP2/MNS32 | 1050.6 / 2521.8 | 18.0 / 21.3 | 62267.5 / 117239.3 | +| TP2/MNS64 | 375.8 / 1148.6 | 18.2 / 21.6 | 62353.6 / 118392.3 | +| TP4/MNS8 | 102605.9 / 193359.7 | **6.8 / 7.2** | 126405.4 / 209092.4 | +| TP4/MNS16 | 35042.6 / 74281.3 | 8.8 / 9.6 | 65779.1 / 110374.3 | +| TP4/MNS32 | 6296.2 / 19159.9 | 12.2 / 13.8 | 47944.5 / 83825.7 | +| TP4/MNS64 | **246.0 / 685.5** | 13.2 / 15.4 | **44985.1 / 83763.6** | + +Thus real-vLLM chooses TP4/MNS64 for mean/p90 TTFT and E2E, and TP4/MNS8 +for mean/p90 TPOT. These are different single-metric objectives, not a +claim that one configuration simultaneously optimizes all three. + +## Correct conclusion and next gate + +This run establishes two bounded findings, neither of which is a full Frontier +selection verdict: (1) the initial simulator-facing Trace-PD prefix contract +was invalid, and (2) its corrected TP2/MNS16 liveness probe estimates a much +smaller decode service rate than the real system. The valid real surface remains +a frozen ground truth, but all raw Frontier numbers above are excluded from the +research claim. We therefore retract the prior bounded counterexample and do +not yet know whether Frontier selects the correct config on this case. + +Before a CPU-only Frontier rerun, the comparison must additionally record +these remaining alignment gaps: + +1. real vLLM uses `FULL_AND_PIECEWISE` CUDA graphs, whereas the raw Frontier + command explicitly used `--decode_cuda_graph_mode none`; Frontier exposes + `full_decode_only` and `piecewise`, not the identical combined mode. More + importantly, Frontier chooses kernel-only predictor families whenever this + mode is non-`none`, so a graph-compatible rerun also requires corresponding + kernel-only profile coverage rather than merely changing the CLI flag; +2. Frontier used one explicit KV-block count per TP, while real vLLM's graph + capture changes the count with MNS (the mismatch is small but measurable); +3. Frontier skipped CPU-overhead modeling, and its MoE CSV has only + `standalone_legacy` gating rows, so the code warned and fell back rather + than training the requested `prefill_hot` pseudo-model; +4. the runner's rate *metadata* incorrectly fixed the TP2/TP4 numerator to a + 600-second window. Arrival timestamps supplied to the simulator were + correct, so this did not alter the raw execution, but the metadata must be + fixed before reporting a rerun. + +The next admissible Frontier result uses complete-block prefix metadata, +per-`TP×MNS` real KV capacity, correct TP-normalized rate metadata, and an +explicit CUDA-graph compatibility decision. Only a complete, semantically +valid surface may receive the selection metrics. Any remaining mismatch after +that CPU rerun is then evidence about Frontier's scheduler/profile fidelity, +not this adapter. + +## Cost and output + +The measured replay horizons are 597/299/149 s at TP1/2/4. Including five +minutes of per-launch server start/warmup, 36 launches consume about **13 +H20-GPUh nominally**; applying the launcher's historical 12--35 minute +fresh-server envelope yields an upper bound of **41 H20-GPUh**. A separate +CPU-only simulator rerun is mandatory before this GPU stage; no profile +collection is included. + +The remote, experiment-specific output root is +`/home/admin/cpfs/wjh/aituner/simulator-tuning-latency-q30-tp-normalized-u0p01-20260717`. +It contains the prompt-free real audit JSON/Markdown, commands, environment, +GPU/trace/model hashes, raw prompt-free request records, and cache inventory. +Private prompt text remains on `dash0` and is never copied into the repository. diff --git a/runs/telemetry-residual/analyze_p1_state.py b/runs/telemetry-residual/analyze_p1_state.py new file mode 100644 index 0000000..2c2e7b0 --- /dev/null +++ b/runs/telemetry-residual/analyze_p1_state.py @@ -0,0 +1,174 @@ +#!/usr/bin/env python3 +"""Pair P1 engine intervals with detailed Frontier state summaries.""" + +from __future__ import annotations + +import argparse +import json +import math +import sys +from pathlib import Path +from typing import Any + + +HERE = Path(__file__).resolve().parent +sys.path.insert(0, str(HERE)) + +from common_state import load_jsonl, numeric, residual, summarize_engine # noqa: E402 + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + temporary.replace(path) + + +def adjudicated_labels(path: Path) -> dict[tuple[str, str], bool]: + payload = json.loads(path.read_text(encoding="utf-8")) + return { + (row["cell"], row["level"]): bool(row["adjudicated_feasible"]) + for row in payload["examples"] + } + + +def state_runs(root: Path) -> list[Path]: + candidates = list(root.glob("*/result.json")) + if (root / "result.json").is_file(): + candidates.append(root / "result.json") + return sorted( + path + for path in candidates + if json.loads(path.read_text(encoding="utf-8")).get("schema") + == "telemetry-residual-frontier-state-result-v1" + ) + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + labels = adjudicated_labels(args.pilot_metrics) + examples = [] + red_flags = [] + for state_result_path in state_runs(args.sim_state_root): + run_root = state_result_path.parent + manifest = json.loads((run_root / "run_manifest.json").read_text(encoding="utf-8")) + cell = manifest["entry"]["cell"] + role = manifest["entry"]["role"] + level = "low" if role.startswith("low") else "high" + real_run = args.real_root / "cells" / cell / f"{level}-rep1" + real_result = json.loads((real_run / "result.json").read_text(encoding="utf-8")) + stream_paths = sorted((args.real_root / "cells" / cell / "opprof").glob("*.jsonl")) + if len(stream_paths) != 1: + raise ValueError(f"expected one engine stream for {cell}, found {len(stream_paths)}") + engine = summarize_engine( + load_jsonl(stream_paths[0]), + start_ns=int(real_result["interval"]["start_mono_ns"]), + end_ns=int(real_result["interval"]["end_mono_ns"]), + request_count=int(real_result["selection"]["count"]), + ) + simulator = json.loads((run_root / "common-state.json").read_text(encoding="utf-8")) + scorer = json.loads((run_root / "scorer_output.json").read_text(encoding="utf-8")) + if engine["interval"]["request_count"] != simulator["interval"]["request_count"]: + red_flags.append(f"request_count_mismatch:{cell}:{role}") + difference = residual(engine, simulator) + if any(not math.isfinite(float(value)) for value in difference["values"].values()): + red_flags.append(f"nonfinite_residual:{cell}:{role}") + real_feasible = labels[(cell, level)] + sim_feasible = bool(scorer["slo"]["feasible"]) + examples.append( + { + "cell": cell, + "role": role, + "level": level, + "tp": int(cell.split("_")[0][2:]), + "mns": int(cell.split("_")[1][3:]), + "request_count": engine["interval"]["request_count"], + "real_feasible": real_feasible, + "sim_feasible": sim_feasible, + "simulator_error": sim_feasible != real_feasible, + "simulator_false_feasible": sim_feasible and not real_feasible, + "real_pass_rate_rep1": float(real_result["pass_rate"]), + "offered_req_s": float(real_result["selection"]["offered_req_s"]), + "offered_req_s_per_gpu": float( + real_result["selection"]["offered_req_s_per_gpu"] + ), + "sim_pass_rate": float(scorer["slo"]["pass_rate"]), + "pass_rate_residual": float(real_result["pass_rate"]) + - float(scorer["slo"]["pass_rate"]), + "engine": engine, + "simulator": simulator, + "state_residual": difference, + "paths": { + "real_result": str((real_run / "result.json").resolve()), + "engine_stream": str(stream_paths[0].resolve()), + "simulator_result": str(state_result_path.resolve()), + }, + } + ) + if not examples: + red_flags.append("no_state_examples") + request_counts = [int(row["request_count"]) for row in examples] + pass_rate_residuals = [float(row["pass_rate_residual"]) for row in examples] + result = { + "schema": "telemetry-residual-p1-state-pairs-v1", + "status": "PASS" if not red_flags else "STOP", + "scope": "P1 development plumbing; not held-out contribution evidence", + "examples": examples, + "red_flags": red_flags, + "sanity": { + "n": len(examples), + "request_count": numeric(request_counts) if request_counts else None, + "pass_rate_residual": numeric(pass_rate_residuals) + if pass_rate_residuals + else None, + "simulator_errors": sum(bool(row["simulator_error"]) for row in examples), + "invariants": { + "request_counts_match": not any( + flag.startswith("request_count_mismatch") for flag in red_flags + ), + "finite_residuals": not any( + flag.startswith("nonfinite_residual") for flag in red_flags + ), + "ratios_bounded": all( + 0.0 <= row["real_pass_rate_rep1"] <= 1.0 + and 0.0 <= row["sim_pass_rate"] <= 1.0 + for row in examples + ), + "per_config_not_identical": len(set(pass_rate_residuals)) > 1 + if len(pass_rate_residuals) > 1 + else None, + }, + }, + } + atomic_json(args.output, result) + if result["status"] != "PASS": + raise RuntimeError(red_flags) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--real-root", type=Path, required=True) + result.add_argument("--sim-state-root", type=Path, required=True) + result.add_argument("--pilot-metrics", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "examples": len(result["examples"]), + "simulator_errors": result["sanity"]["simulator_errors"], + "sanity": result["sanity"], + "red_flags": result["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/analyze_r0_gate.py b/runs/telemetry-residual/analyze_r0_gate.py new file mode 100644 index 0000000..d9912e5 --- /dev/null +++ b/runs/telemetry-residual/analyze_r0_gate.py @@ -0,0 +1,292 @@ +#!/usr/bin/env python3 +"""Make the registered R0 go/no-go decision from development artifacts.""" + +from __future__ import annotations + +import argparse +import json +import math +from pathlib import Path +from typing import Any + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + temporary.replace(path) + + +def load_pass(path: Path, name: str) -> dict[str, Any]: + payload = json.loads(path.read_text(encoding="utf-8")) + if payload.get("status") != "PASS": + raise RuntimeError(f"{name} is not a valid PASS artifact: {path}") + return payload + + +def reduction(reference: float, candidate: float) -> float: + if reference <= 0.0 or candidate < 0.0 or not math.isfinite(candidate): + raise ValueError("costs must be finite and non-negative with positive reference") + return 1.0 - candidate / reference + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + paired = load_pass(args.paired_state, "paired state") + transfer = load_pass(args.transfer, "transfer diagnostic") + e2e = load_pass(args.pilot_e2e, "P1 E2E replay") + red_flags = [] + + if len(paired.get("examples", [])) != 12: + red_flags.append("paired_state_not_12_anchors") + if transfer.get("sanity", {}).get("transitions") != 120: + red_flags.append("transfer_not_120_cross_config_transitions") + if transfer.get("red_flags"): + red_flags.append("transfer_has_red_flags") + + examples = paired["examples"] + state_available = all( + row["state_residual"]["coverage"]["missing"] == 0 + and row["state_residual"]["coverage"]["available"] > 0 + for row in examples + ) + state_vectors = { + tuple(sorted(row["state_residual"]["values"].items())) for row in examples + } + state_varies = len(state_vectors) > 1 + error_examples = [row for row in examples if row["simulator_error"]] + simulator_errors = len(error_examples) + error_state_discrepancy = any( + any( + abs(float(value)) > 1e-12 + for value in row["state_residual"]["values"].values() + ) + for row in error_examples + ) + + simulator = transfer["simulator"] + prior_safe = [] + direct_sensitivity = [] + hybrid_incremental = [] + for regularization, detail in transfer["regularization_sensitivity"].items(): + for weight, models in detail["hybrid"]["prior_shrinkage"].items(): + if float(weight) == 0.0: + continue + telemetry = models["raw_simulator_prior"][ + "sim_plus_outcome_plus_telemetry" + ] + decision_safe = ( + telemetry["simulator_errors_corrected"] >= 1 + and telemetry["simulator_errors_corrected"] + >= telemetry["simulator_correct_corrupted"] + ) + continuous_safe = ( + telemetry["rmse"] <= simulator["rmse"] + 1e-12 + and telemetry["mae"] <= simulator["mae"] + 1e-12 + ) + if decision_safe and continuous_safe: + prior_safe.append( + { + "regularization": float(regularization), + "prior_weight": float(weight), + "simulator_errors_corrected": telemetry[ + "simulator_errors_corrected" + ], + "simulator_correct_corrupted": telemetry[ + "simulator_correct_corrupted" + ], + "rmse": telemetry["rmse"], + "mae": telemetry["mae"], + } + ) + + direct = detail["direct"] + direct_cmp = direct["comparison"] + direct_sensitivity.append( + { + "regularization": float(regularization), + "accuracy_delta": direct_cmp["delta_telemetry_minus_baseline"][ + "feasibility_accuracy" + ], + "rmse_delta": direct_cmp["delta_telemetry_minus_baseline"]["rmse"], + "mae_delta": direct_cmp["delta_telemetry_minus_baseline"]["mae"], + "errors_corrected": direct_cmp["baseline_errors_corrected"], + "correct_corrupted": direct_cmp["baseline_correct_corrupted"], + "telemetry_accuracy": direct["telemetry_only"][ + "feasibility_accuracy" + ], + } + ) + hybrid_cmp = detail["hybrid"]["comparison"] + hybrid_incremental.append( + { + "regularization": float(regularization), + "accuracy_delta": hybrid_cmp["delta_telemetry_minus_baseline"][ + "feasibility_accuracy" + ], + "rmse_delta": hybrid_cmp["delta_telemetry_minus_baseline"]["rmse"], + "mae_delta": hybrid_cmp["delta_telemetry_minus_baseline"]["mae"], + "errors_corrected": hybrid_cmp["baseline_errors_corrected"], + "correct_corrupted": hybrid_cmp["baseline_correct_corrupted"], + } + ) + + k1 = e2e["by_k"]["1"]["sim_top_k_plus_real_final"] + k2 = e2e["by_k"]["2"]["sim_top_k_plus_real_final"] + headroom = { + "interpretation": ( + "oracle correction stops after the simulator top-1 real final instead " + "of evaluating the frozen safety top-2" + ), + "online": { + "reference_k2_h20_hours": k2["online_h20_hours"], + "oracle_k1_h20_hours": k1["online_h20_hours"], + "absolute_h20_hours": k2["online_h20_hours"] - k1["online_h20_hours"], + "fraction": reduction(k2["online_h20_hours"], k1["online_h20_hours"]), + }, + "with_prior_failure": { + "reference_k2_h20_hours": k2["conservative_h20_hours_with_prior_failure"], + "oracle_k1_h20_hours": k1["conservative_h20_hours_with_prior_failure"], + "absolute_h20_hours": k2["conservative_h20_hours_with_prior_failure"] + - k1["conservative_h20_hours_with_prior_failure"], + "fraction": reduction( + k2["conservative_h20_hours_with_prior_failure"], + k1["conservative_h20_hours_with_prior_failure"], + ), + }, + "versus_observed_safe_k1_fraction": 0.0, + "k1_zero_regret": k1["real_regret"] == 0.0, + "k2_zero_regret": k2["real_regret"] == 0.0, + } + + condition_1 = state_available and state_varies + condition_2 = simulator_errors >= 1 and error_state_discrepancy + condition_3 = bool(prior_safe) + condition_4 = headroom["online"]["fraction"] >= 0.15 + conditions = { + "state_available_and_varies": condition_1, + "known_simulator_error_has_state_discrepancy": condition_2, + "prior_preserving_safe_correction_exists": condition_3, + "oracle_online_headroom_at_least_15pct": condition_4, + } + gate_pass = not red_flags and all(conditions.values()) + direct_incremental = all( + row["accuracy_delta"] >= -1e-12 + and row["errors_corrected"] >= row["correct_corrupted"] + for row in direct_sensitivity + ) + result = { + "schema": "telemetry-residual-r0-gate-v1", + "status": "STOP" if red_flags else "PASS", + "scope": "P1 development premise/headroom audit; not headline evidence", + "decision": "PROCEED_TO_R1" if gate_pass else "STOP_BEFORE_R1", + "r0_gate_pass": gate_pass, + "conditions": conditions, + "route_findings": { + "hybrid_prior_safe_candidates": prior_safe, + "hybrid_incremental_regularization": hybrid_incremental, + "direct_incremental_regularization": direct_sensitivity, + "direct_incremental_decision_signal_all_lambdas": direct_incremental, + "direct_best_absolute_accuracy": max( + row["telemetry_accuracy"] for row in direct_sensitivity + ), + "raw_simulator_accuracy": simulator["feasibility_accuracy"], + }, + "headroom": headroom, + "red_flags": red_flags, + "sanity": { + "anchors": { + "n": len(examples), + "min": min(row["real_pass_rate_rep1"] for row in examples), + "max": max(row["real_pass_rate_rep1"] for row in examples), + "distinct_n": len( + {row["real_pass_rate_rep1"] for row in examples} + ), + }, + "state_vectors": { + "n": len(examples), + "min": min(len(row["state_residual"]["values"]) for row in examples), + "max": max(len(row["state_residual"]["values"]) for row in examples), + "distinct_n": len(state_vectors), + }, + "costs_h20_hours": { + "n": 4, + "min": min( + k1["online_h20_hours"], + k2["online_h20_hours"], + k1["conservative_h20_hours_with_prior_failure"], + k2["conservative_h20_hours_with_prior_failure"], + ), + "max": max( + k1["online_h20_hours"], + k2["online_h20_hours"], + k1["conservative_h20_hours_with_prior_failure"], + k2["conservative_h20_hours_with_prior_failure"], + ), + "distinct_n": len( + { + k1["online_h20_hours"], + k2["online_h20_hours"], + k1["conservative_h20_hours_with_prior_failure"], + k2["conservative_h20_hours_with_prior_failure"], + } + ), + }, + "invariants": { + "no_data_red_flags": not red_flags, + "state_nonempty_and_varied": condition_1, + "pass_rates_bounded": all( + 0.0 <= row["real_pass_rate_rep1"] <= 1.0 + and 0.0 <= row["sim_pass_rate"] <= 1.0 + for row in examples + ), + "costs_nonnegative": all( + value >= 0.0 + for value in ( + k1["online_h20_hours"], + k2["online_h20_hours"], + k1["conservative_h20_hours_with_prior_failure"], + k2["conservative_h20_hours_with_prior_failure"], + ) + ), + "per_config_not_identical": len( + {row["real_pass_rate_rep1"] for row in examples} + ) + > 1, + }, + }, + } + atomic_json(args.output, result) + if result["status"] != "PASS": + raise RuntimeError(red_flags) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--paired-state", type=Path, required=True) + result.add_argument("--transfer", type=Path, required=True) + result.add_argument("--pilot-e2e", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "decision": result["decision"], + "r0_gate_pass": result["r0_gate_pass"], + "conditions": result["conditions"], + "headroom": result["headroom"], + "sanity": result["sanity"], + "red_flags": result["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/analyze_residual_transfer.py b/runs/telemetry-residual/analyze_residual_transfer.py new file mode 100644 index 0000000..a07283a --- /dev/null +++ b/runs/telemetry-residual/analyze_residual_transfer.py @@ -0,0 +1,480 @@ +#!/usr/bin/env python3 +"""Development-only cross-config telemetry transfer diagnostic for P1. + +Each example asks whether state observed at one source anchor helps predict the +pass rate at a different target config. The hybrid branch predicts the +real-minus-simulator residual; the direct branch never reads simulator state or +outcomes. Folds exclude both the source and target config identities. The two +offered-load anchors belong to one trace/SLO task, so this is a premise check +rather than generalization evidence. +""" + +from __future__ import annotations + +import argparse +import json +import math +from pathlib import Path +from typing import Any, Sequence + +import numpy as np + + +REGULARIZATION = (0.1, 1.0, 10.0, 100.0) +PRIOR_SHRINKAGE = (0.0, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0) + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + temporary.replace(path) + + +def finite(value: Any, *, name: str) -> float: + result = float(value) + if not math.isfinite(result): + raise ValueError(f"non-finite feature {name}={value!r}") + return result + + +def flatten_residual_state(example: dict[str, Any]) -> tuple[float, ...]: + residual = example["state_residual"]["values"] + values = [finite(residual[name], name=name) for name in sorted(residual)] + engine_only = example["engine"]["engine_only"] + values.extend( + finite(engine_only[name], name=name) + for name in sorted(engine_only) + ) + return tuple(values) + + +def flatten_engine_state(example: dict[str, Any]) -> tuple[float, ...]: + values = [] + for name in sorted(example["engine"]["common"]): + value = example["engine"]["common"][name] + if isinstance(value, dict): + values.extend( + finite(value[statistic], name=f"{name}.{statistic}") + for statistic in ("mean", "max", "cv") + ) + elif value is not None: + values.append(finite(value, name=name)) + engine_only = example["engine"]["engine_only"] + values.extend( + finite(engine_only[name], name=name) + for name in sorted(engine_only) + ) + return tuple(values) + + +def config_transition_features( + source: dict[str, Any], target: dict[str, Any] +) -> tuple[float, ...]: + source_rate = finite( + source["offered_req_s_per_gpu"], name="source_offered_req_s_per_gpu" + ) + target_rate = finite( + target["offered_req_s_per_gpu"], name="target_offered_req_s_per_gpu" + ) + return ( + math.log2(float(source["tp"])), + math.log2(float(source["mns"])), + math.log2(float(target["tp"])), + math.log2(float(target["mns"])), + math.log2(float(target["tp"]) / float(source["tp"])), + math.log2(float(target["mns"]) / float(source["mns"])), + math.log2(source_rate), + math.log2(target_rate), + math.log2(target_rate / source_rate), + ) + + +def hybrid_base_features( + source: dict[str, Any], target: dict[str, Any] +) -> tuple[float, ...]: + return ( + finite(source["pass_rate_residual"], name="source_pass_rate_residual"), + finite(source["sim_pass_rate"], name="source_sim_pass_rate"), + finite(target["sim_pass_rate"], name="target_sim_pass_rate"), + ) + config_transition_features(source, target) + + +def direct_base_features( + source: dict[str, Any], target: dict[str, Any] +) -> tuple[float, ...]: + return ( + finite(source["real_pass_rate_rep1"], name="source_real_pass_rate"), + ) + config_transition_features(source, target) + + +def transitions(examples: Sequence[dict[str, Any]]) -> list[dict[str, Any]]: + rows = [] + for source in examples: + residual_state = flatten_residual_state(source) + engine_state = flatten_engine_state(source) + for target in examples: + # Low/high are offered-load anchors inside one workload/SLO task. + # Cross-load transitions are legal; same-cell transitions are + # excluded because R0 asks about transfer to a new configuration. + if source["cell"] == target["cell"]: + continue + hybrid_base = hybrid_base_features(source, target) + direct_base = direct_base_features(source, target) + rows.append( + { + "source_cell": source["cell"], + "source_level": source["level"], + "target_cell": target["cell"], + "target_level": target["level"], + "hybrid_base": hybrid_base, + "hybrid_telemetry": hybrid_base + residual_state, + "direct_base": direct_base, + "direct_telemetry": direct_base + engine_state, + "target_residual": finite( + target["pass_rate_residual"], name="target_residual" + ), + "target_residual_delta": finite( + target["pass_rate_residual"], name="target_residual" + ) + - finite(source["pass_rate_residual"], name="source_residual"), + "target_real_pass_rate": finite( + target["real_pass_rate_rep1"], name="target_real_pass_rate" + ), + "target_real_pass_rate_delta": finite( + target["real_pass_rate_rep1"], name="target_real_pass_rate" + ) + - finite( + source["real_pass_rate_rep1"], name="source_real_pass_rate" + ), + "source_residual": finite( + source["pass_rate_residual"], name="source_residual" + ), + "source_real_pass_rate": finite( + source["real_pass_rate_rep1"], name="source_real_pass_rate" + ), + "target_sim_pass_rate": finite( + target["sim_pass_rate"], name="target_sim_pass_rate" + ), + "target_real_feasible": bool(target["real_feasible"]), + "target_sim_feasible": bool(target["sim_feasible"]), + } + ) + return rows + + +def fit_predict( + train_x: np.ndarray, + train_y: np.ndarray, + test_x: np.ndarray, + regularization: float, +) -> np.ndarray: + mean = train_x.mean(axis=0) + std = train_x.std(axis=0) + std[std == 0.0] = 1.0 + train = (train_x - mean) / std + test = (test_x - mean) / std + train = np.column_stack((np.ones(len(train)), train)) + test = np.column_stack((np.ones(len(test)), test)) + penalty = np.eye(train.shape[1], dtype=np.float64) + penalty[0, 0] = 0.0 + weights = np.linalg.lstsq( + train.T @ train + regularization * penalty, + train.T @ train_y, + rcond=None, + )[0] + return test @ weights + + +def grouped_predictions( + rows: Sequence[dict[str, Any]], + *, + feature_name: str, + target_name: str, + regularization: float, +) -> np.ndarray: + predictions = np.zeros(len(rows), dtype=np.float64) + groups = sorted({(row["source_cell"], row["target_cell"]) for row in rows}) + for source_cell, target_cell in groups: + held_out_cells = {source_cell, target_cell} + test_indexes = [ + index + for index, row in enumerate(rows) + if row["source_cell"] == source_cell and row["target_cell"] == target_cell + ] + train_indexes = [ + index + for index, row in enumerate(rows) + if row["source_cell"] not in held_out_cells + and row["target_cell"] not in held_out_cells + ] + if not test_indexes or not train_indexes: + raise ValueError(f"empty grouped fold for {source_cell}->{target_cell}") + train_x = np.asarray([rows[index][feature_name] for index in train_indexes]) + train_y = np.asarray([rows[index][target_name] for index in train_indexes]) + test_x = np.asarray([rows[index][feature_name] for index in test_indexes]) + predictions[test_indexes] = fit_predict( + train_x, train_y, test_x, regularization + ) + return predictions + + +def metrics(rows: Sequence[dict[str, Any]], predicted_pass: np.ndarray) -> dict[str, Any]: + truth = np.asarray( + [row["target_real_pass_rate"] for row in rows], dtype=np.float64 + ) + real_feasible = np.asarray( + [row["target_real_feasible"] for row in rows], dtype=bool + ) + sim_feasible = np.asarray( + [row["target_sim_feasible"] for row in rows], dtype=bool + ) + clipped_pass = np.clip(predicted_pass, 0.0, 1.0) + predicted_feasible = clipped_pass >= 0.95 + baseline_correct = sim_feasible == real_feasible + model_correct = predicted_feasible == real_feasible + return { + "rmse": float(np.sqrt(np.mean((predicted_pass - truth) ** 2))), + "mae": float(np.mean(np.abs(predicted_pass - truth))), + "feasibility_accuracy": float(np.mean(model_correct)), + "false_feasible": int(np.sum(predicted_feasible & ~real_feasible)), + "false_infeasible": int(np.sum(~predicted_feasible & real_feasible)), + "simulator_errors_corrected": int(np.sum(~baseline_correct & model_correct)), + "simulator_correct_corrupted": int(np.sum(baseline_correct & ~model_correct)), + "predicted_pass_rate": { + "n": len(clipped_pass), + "min": float(clipped_pass.min()), + "max": float(clipped_pass.max()), + "distinct_n": len(set(float(value) for value in clipped_pass)), + }, + } + + +def compare( + rows: Sequence[dict[str, Any]], + baseline_prediction: np.ndarray, + telemetry_prediction: np.ndarray, + baseline: dict[str, Any], + telemetry: dict[str, Any], +) -> dict[str, Any]: + truth = np.asarray([row["target_real_feasible"] for row in rows], dtype=bool) + baseline_feasible = np.clip(baseline_prediction, 0.0, 1.0) >= 0.95 + telemetry_feasible = np.clip(telemetry_prediction, 0.0, 1.0) >= 0.95 + baseline_correct = baseline_feasible == truth + telemetry_correct = telemetry_feasible == truth + return { + "delta_telemetry_minus_baseline": { + "rmse": telemetry["rmse"] - baseline["rmse"], + "mae": telemetry["mae"] - baseline["mae"], + "feasibility_accuracy": telemetry["feasibility_accuracy"] + - baseline["feasibility_accuracy"], + }, + "baseline_errors_corrected": int( + np.sum(~baseline_correct & telemetry_correct) + ), + "baseline_correct_corrupted": int( + np.sum(baseline_correct & ~telemetry_correct) + ), + } + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + paired = json.loads(args.paired_state.read_text(encoding="utf-8")) + if paired.get("status") != "PASS" or len(paired["examples"]) != 12: + raise RuntimeError("paired P1 state evidence is incomplete") + rows = transitions(paired["examples"]) + residual_truth = np.asarray( + [row["target_residual"] for row in rows], dtype=np.float64 + ) + pass_truth = np.asarray( + [row["target_real_pass_rate"] for row in rows], dtype=np.float64 + ) + sim_pass = np.asarray( + [row["target_sim_pass_rate"] for row in rows], dtype=np.float64 + ) + simulator = metrics(rows, sim_pass) + sensitivity = {} + for regularization in REGULARIZATION: + hybrid_base_delta = grouped_predictions( + rows, + feature_name="hybrid_base", + target_name="target_residual_delta", + regularization=regularization, + ) + hybrid_telemetry_delta = grouped_predictions( + rows, + feature_name="hybrid_telemetry", + target_name="target_residual_delta", + regularization=regularization, + ) + source_residual = np.asarray( + [row["source_residual"] for row in rows], dtype=np.float64 + ) + hybrid_base_correction = source_residual + hybrid_base_delta + hybrid_telemetry_correction = source_residual + hybrid_telemetry_delta + hybrid_base_prediction = sim_pass + hybrid_base_correction + hybrid_telemetry_prediction = sim_pass + hybrid_telemetry_correction + hybrid_base = metrics(rows, hybrid_base_prediction) + hybrid_telemetry = metrics(rows, hybrid_telemetry_prediction) + prior_shrinkage = {} + for weight in PRIOR_SHRINKAGE: + prior_shrinkage[str(weight)] = { + "raw_simulator_prior": { + "sim_plus_outcome": metrics( + rows, sim_pass + weight * hybrid_base_correction + ), + "sim_plus_outcome_plus_telemetry": metrics( + rows, sim_pass + weight * hybrid_telemetry_correction + ), + }, + "anchor_offset_prior": { + "sim_plus_outcome": metrics( + rows, + sim_pass + source_residual + weight * hybrid_base_delta, + ), + "sim_plus_outcome_plus_telemetry": metrics( + rows, + sim_pass + source_residual + weight * hybrid_telemetry_delta, + ), + }, + } + + direct_base_prediction = grouped_predictions( + rows, + feature_name="direct_base", + target_name="target_real_pass_rate_delta", + regularization=regularization, + ) + direct_telemetry_prediction = grouped_predictions( + rows, + feature_name="direct_telemetry", + target_name="target_real_pass_rate_delta", + regularization=regularization, + ) + source_real_pass = np.asarray( + [row["source_real_pass_rate"] for row in rows], dtype=np.float64 + ) + direct_base_prediction = source_real_pass + direct_base_prediction + direct_telemetry_prediction = source_real_pass + direct_telemetry_prediction + direct_base = metrics(rows, direct_base_prediction) + direct_telemetry = metrics(rows, direct_telemetry_prediction) + sensitivity[str(regularization)] = { + "hybrid": { + "sim_plus_outcome": hybrid_base, + "sim_plus_outcome_plus_telemetry": hybrid_telemetry, + "comparison": compare( + rows, + hybrid_base_prediction, + hybrid_telemetry_prediction, + hybrid_base, + hybrid_telemetry, + ), + "prior_shrinkage": prior_shrinkage, + }, + "direct": { + "real_outcome_only": direct_base, + "telemetry_only": direct_telemetry, + "comparison": compare( + rows, + direct_base_prediction, + direct_telemetry_prediction, + direct_base, + direct_telemetry, + ), + }, + } + red_flags = [] + if any(not math.isfinite(value) for value in residual_truth): + red_flags.append("nonfinite_residual_target") + if any(not math.isfinite(value) for value in pass_truth): + red_flags.append("nonfinite_pass_rate_target") + result = { + "schema": "telemetry-residual-cross-config-diagnostic-v1", + "status": "PASS" if not red_flags else "STOP", + "scope": ( + "single P1 trace/SLO-task development diagnostic; ordered transitions " + "are not independent tasks and cannot support a generalization claim" + ), + "split": ( + "hold out both ordered source config and target config identities; " + "each fold contains both offered-load anchors" + ), + "features": { + "sim_plus_outcome": len(rows[0]["hybrid_base"]), + "sim_plus_outcome_plus_telemetry": len(rows[0]["hybrid_telemetry"]), + "real_outcome_only": len(rows[0]["direct_base"]), + "telemetry_only": len(rows[0]["direct_telemetry"]), + }, + "simulator": simulator, + "regularization_sensitivity": sensitivity, + "red_flags": red_flags, + "sanity": { + "transitions": len(rows), + "load_levels": len( + {row["source_level"] for row in rows} + | {row["target_level"] for row in rows} + ), + "cross_load_transitions": sum( + row["source_level"] != row["target_level"] for row in rows + ), + "ordered_cell_pairs": len( + {(row["source_cell"], row["target_cell"]) for row in rows} + ), + "target_residual": { + "n": len(residual_truth), + "min": float(residual_truth.min()), + "max": float(residual_truth.max()), + "distinct_n": len(set(float(value) for value in residual_truth)), + }, + "target_real_pass_rate": { + "n": len(pass_truth), + "min": float(pass_truth.min()), + "max": float(pass_truth.max()), + "distinct_n": len(set(float(value) for value in pass_truth)), + }, + "invariants": { + "finite_targets": not red_flags, + "ratios_bounded": all( + 0.0 <= row["target_sim_pass_rate"] <= 1.0 for row in rows + ), + "source_differs_from_target": all( + row["source_cell"] != row["target_cell"] for row in rows + ), + "per_config_not_identical": len( + set(float(value) for value in pass_truth) + ) + > 1, + }, + }, + } + atomic_json(args.output, result) + if result["status"] != "PASS": + raise RuntimeError(red_flags) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--paired-state", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "transitions": result["sanity"]["transitions"], + "sensitivity": result["regularization_sensitivity"], + "sanity": result["sanity"], + "red_flags": result["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/common_state.py b/runs/telemetry-residual/common_state.py new file mode 100644 index 0000000..9850e1c --- /dev/null +++ b/runs/telemetry-residual/common_state.py @@ -0,0 +1,400 @@ +#!/usr/bin/env python3 +"""Extract directly measurable engine/simulator state into one schema. + +The schema deliberately keeps common, engine-only, and simulator-only fields +separate. Missing simulator mechanisms must remain missing; they must not be +filled by a heuristic bottleneck label. +""" + +from __future__ import annotations + +import csv +import json +import math +from pathlib import Path +from statistics import fmean +from typing import Any, Iterable, Mapping, Sequence + + +SCHEMA = "telemetry-common-state-v1" + + +def numeric(values: Iterable[float | int]) -> dict[str, Any]: + finite = [float(value) for value in values] + if not finite: + raise ValueError("numeric summary requires at least one value") + if any(not math.isfinite(value) for value in finite): + raise ValueError("numeric summary received a non-finite value") + mean = fmean(finite) + variance = fmean((value - mean) ** 2 for value in finite) + return { + "n": len(finite), + "min": min(finite), + "max": max(finite), + "mean": mean, + "cv": math.sqrt(variance) / abs(mean) if mean else 0.0, + "distinct_n": len(set(finite)), + } + + +def load_jsonl(path: Path) -> list[dict[str, Any]]: + rows = [] + with path.open(encoding="utf-8") as source: + for line_number, line in enumerate(source, start=1): + if not line.strip(): + continue + row = json.loads(line) + if not isinstance(row, dict): + raise ValueError(f"{path}:{line_number}: expected a JSON object") + rows.append(row) + if not rows: + raise ValueError(f"{path}: no JSONL records") + return rows + + +def _as_number(value: Any, *, name: str) -> float: + if isinstance(value, bool) or not isinstance(value, (int, float)): + raise ValueError(f"{name} must be numeric, got {value!r}") + result = float(value) + if not math.isfinite(result): + raise ValueError(f"{name} must be finite, got {value!r}") + return result + + +def _time_weighted_mean( + records: Sequence[Mapping[str, Any]], + *, + start_ns: int, + end_ns: int, + value, +) -> float: + if end_ns <= start_ns: + raise ValueError("time-weighted interval must be positive") + selected = [ + record + for record in records + if start_ns <= int(record["submit_mono_ns"]) <= end_ns + ] + if not selected: + raise ValueError("time-weighted interval contains no records") + selected.sort(key=lambda record: int(record["submit_mono_ns"])) + cursor = start_ns + total = 0.0 + current = _as_number(value(selected[0]), name="time-weighted value") + for record in selected[1:]: + timestamp = int(record["submit_mono_ns"]) + if timestamp < cursor: + raise ValueError("telemetry timestamps are not monotonic") + total += current * (timestamp - cursor) + cursor = timestamp + current = _as_number(value(record), name="time-weighted value") + total += current * (end_ns - cursor) + return total / (end_ns - start_ns) + + +def summarize_engine( + records: Sequence[Mapping[str, Any]], + *, + start_ns: int, + end_ns: int, + request_count: int, +) -> dict[str, Any]: + """Summarize a measured engine interval from Layer-1 records.""" + + if request_count <= 0: + raise ValueError("request_count must be positive") + layer1 = [record for record in records if "step_index" in record] + if not layer1: + raise ValueError("engine stream has no Layer-1 records") + step_indexes = [int(record["step_index"]) for record in layer1] + if len(step_indexes) != len(set(step_indexes)): + raise ValueError("engine Layer-1 step indexes are not unique") + if step_indexes != sorted(step_indexes): + raise ValueError("engine Layer-1 step indexes are not ordered") + if any(int(record.get("dropped_records_before", 0)) != 0 for record in layer1): + raise ValueError("engine Layer-1 stream reports dropped records") + + interval = [ + record + for record in layer1 + if start_ns <= int(record["submit_mono_ns"]) <= end_ns + ] + if not interval: + raise ValueError("engine interval has no Layer-1 records") + executed = [record for record in interval if bool(record["model_executed"])] + if not executed: + raise ValueError("engine interval has no executed model steps") + duration_s = (end_ns - start_ns) / 1e9 + if duration_s <= 0: + raise ValueError("engine interval duration must be positive") + + batch_sizes = [int(record["scheduled_requests"]) for record in executed] + prefill_tokens = [int(record["prefill_tokens"]) for record in executed] + decode_tokens = [int(record["decode_tokens"]) for record in executed] + batch_tokens = [ + prefill + decode + for prefill, decode in zip(prefill_tokens, decode_tokens, strict=True) + ] + decode_batches = [int(record["decode_batch_size"]) for record in executed] + if any(value < 0 for value in batch_sizes + batch_tokens + decode_batches): + raise ValueError("engine batch counters must be non-negative") + if any( + int(record["prefill_tokens"]) + int(record["decode_tokens"]) + <= 0 + for record in executed + ): + raise ValueError("executed engine step has no scheduled tokens") + + waiting_mean = _time_weighted_mean( + interval, + start_ns=start_ns, + end_ns=end_ns, + value=lambda record: record["queues"]["waiting"], + ) + running_mean = _time_weighted_mean( + interval, + start_ns=start_ns, + end_ns=end_ns, + value=lambda record: record["queues"]["running"], + ) + kv_mean = _time_weighted_mean( + interval, + start_ns=start_ns, + end_ns=end_ns, + value=lambda record: record["kv"]["usage"], + ) + kv_values = [float(record["kv"]["usage"]) for record in interval] + if any(not 0.0 <= value <= 1.0 for value in kv_values): + raise ValueError("engine KV usage must be in [0, 1]") + total_prefill = sum(prefill_tokens) + total_decode = sum(decode_tokens) + graph_modes = [str(record["cudagraph"]["runtime_mode"]) for record in executed] + bucket_tokens = sum(int(record["cudagraph"]["bucket_tokens"]) for record in executed) + padding_tokens = sum(int(record["cudagraph"]["padding_tokens"]) for record in executed) + + common = { + "scheduler_steps_per_s": len(executed) / duration_s, + "batch_size": numeric(batch_sizes), + "batch_tokens": numeric(batch_tokens), + "decode_batch_size": numeric(decode_batches), + "prefill_token_fraction": total_prefill / (total_prefill + total_decode), + "queue_waiting_mean": waiting_mean, + "queue_running_mean": running_mean, + "queue_waiting_time_per_request_ms": waiting_mean * duration_s * 1000.0 / request_count, + "queue_running_time_per_request_ms": running_mean * duration_s * 1000.0 / request_count, + "preemptions": sum(int(record["preemptions"]) for record in executed), + } + result = { + "schema": SCHEMA, + "source": "engine_layer1", + "interval": { + "start_ns": start_ns, + "end_ns": end_ns, + "duration_s": duration_s, + "request_count": request_count, + }, + "common": common, + "engine_only": { + "kv_usage_mean": kv_mean, + "kv_usage_max": max(kv_values), + "kv_usage_end_minus_start": kv_values[-1] - kv_values[0], + "graph_none_share": graph_modes.count("NONE") / len(graph_modes), + "graph_full_share": graph_modes.count("FULL") / len(graph_modes), + "graph_padding_fraction": padding_tokens / max(1, bucket_tokens), + }, + "simulator_only": {}, + "sanity": { + "records": len(interval), + "executed_steps": len(executed), + "step_index_min": min(int(record["step_index"]) for record in interval), + "step_index_max": max(int(record["step_index"]) for record in interval), + "invariants": { + "positive_duration": duration_s > 0, + "positive_request_count": request_count > 0, + "zero_drops": True, + "nonnegative_counters": True, + "kv_bounded": True, + "batch_values_not_all_identical": any( + summary["distinct_n"] > 1 + for summary in ( + common["batch_size"], + common["batch_tokens"], + common["decode_batch_size"], + ) + ), + }, + }, + } + return result + + +def _csv_rows(path: Path) -> list[dict[str, str]]: + with path.open(encoding="utf-8", newline="") as source: + rows = list(csv.DictReader(source)) + if not rows: + raise ValueError(f"{path}: CSV contains no rows") + return rows + + +def _column(rows: Sequence[Mapping[str, str]], name: str) -> list[float]: + if name not in rows[0]: + raise ValueError(f"CSV is missing required column {name!r}") + values = [] + for row in rows: + text = row.get(name, "") + if text == "": + raise ValueError(f"CSV column {name!r} contains an empty value") + values.append(_as_number(float(text), name=name)) + return values + + +def summarize_frontier( + *, + system_metrics_path: Path, + request_metrics_path: Path, + batch_metrics_path: Path | None = None, + ledger_path: Path | None = None, +) -> dict[str, Any]: + """Summarize a Frontier run, retaining unavailable state as null.""" + + system = json.loads(system_metrics_path.read_text(encoding="utf-8")) + throughput = system["throughput_metrics"] + duration_s = _as_number( + throughput["total_duration_seconds"], name="total_duration_seconds" + ) + if duration_s <= 0: + raise ValueError("Frontier duration must be positive") + request_rows = _csv_rows(request_metrics_path) + waiting_ms = _column(request_rows, "request_waiting_time_total") + e2e_ms = _column(request_rows, "request_e2e_time") + running_ms = [max(0.0, e2e - waiting) for e2e, waiting in zip(e2e_ms, waiting_ms, strict=True)] + duration_ms = duration_s * 1000.0 + request_count = len(request_rows) + + common: dict[str, Any] = { + "scheduler_steps_per_s": None, + "batch_size": None, + "batch_tokens": None, + "decode_batch_size": None, + "prefill_token_fraction": None, + "queue_waiting_mean": sum(waiting_ms) / duration_ms, + "queue_running_mean": sum(running_ms) / duration_ms, + "queue_waiting_time_per_request_ms": fmean(waiting_ms), + "queue_running_time_per_request_ms": fmean(running_ms), + "preemptions": sum( + int(float(row.get("request_total_preemption_count") or 0)) + for row in request_rows + ), + } + batch_rows: list[dict[str, str]] = [] + if batch_metrics_path is not None: + batch_rows = _csv_rows(batch_metrics_path) + batch_sizes = _column(batch_rows, "batch_size") + batch_tokens = _column(batch_rows, "batch_num_tokens") + prefill_tokens = _column(batch_rows, "batch_num_prefill_tokens") + decode_tokens = _column(batch_rows, "batch_num_decode_tokens") + if any(value < 0 for value in batch_sizes + batch_tokens + prefill_tokens + decode_tokens): + raise ValueError("Frontier batch counters must be non-negative") + common.update( + { + "scheduler_steps_per_s": len(batch_rows) / duration_s, + "batch_size": numeric(batch_sizes), + "batch_tokens": numeric(batch_tokens), + "decode_batch_size": numeric(decode_tokens), + "prefill_token_fraction": sum(prefill_tokens) + / max(1.0, sum(prefill_tokens) + sum(decode_tokens)), + } + ) + + ledger_rows: list[dict[str, Any]] = [] + if ledger_path is not None: + ledger_rows = load_jsonl(ledger_path) + for row in ledger_rows: + start = _as_number(row["stage_start_ts"], name="stage_start_ts") + end = _as_number(row["stage_end_ts"], name="stage_end_ts") + if end < start: + raise ValueError("Frontier ledger has a negative stage duration") + + batch_distinct = ( + max( + summary["distinct_n"] + for summary in ( + common["batch_size"], + common["batch_tokens"], + common["decode_batch_size"], + ) + ) + if batch_rows + else None + ) + return { + "schema": SCHEMA, + "source": "frontier", + "interval": { + "duration_s": duration_s, + "request_count": request_count, + }, + "common": common, + "engine_only": { + "kv_usage_mean": None, + "kv_usage_max": None, + "kv_usage_end_minus_start": None, + "graph_none_share": None, + "graph_full_share": None, + "graph_padding_fraction": None, + }, + "simulator_only": { + "request_waiting_time_ms": numeric(waiting_ms), + "request_running_time_ms": numeric(running_ms), + "ledger_rows": len(ledger_rows) if ledger_path is not None else None, + }, + "sanity": { + "request_rows": request_count, + "batch_rows": len(batch_rows), + "ledger_rows": len(ledger_rows), + "invariants": { + "positive_duration": duration_s > 0, + "positive_request_count": request_count > 0, + "nonnegative_counters": True, + "request_values_not_all_identical": max( + numeric(waiting_ms)["distinct_n"], + numeric(running_ms)["distinct_n"], + ) + > 1, + "batch_values_not_all_identical": ( + batch_distinct > 1 if batch_distinct is not None else None + ), + }, + }, + } + + +def residual(real: Mapping[str, Any], simulated: Mapping[str, Any]) -> dict[str, Any]: + if real.get("schema") != SCHEMA or simulated.get("schema") != SCHEMA: + raise ValueError("residual inputs must use the common-state schema") + values = {} + missing = [] + for name, real_value in real["common"].items(): + sim_value = simulated["common"].get(name) + if isinstance(real_value, dict): + if not isinstance(sim_value, dict): + missing.append(name) + continue + for statistic in ("mean", "max", "cv"): + key = f"{name}.{statistic}" + values[key] = float(real_value[statistic]) - float(sim_value[statistic]) + continue + if real_value is None or sim_value is None: + missing.append(name) + continue + values[name] = float(real_value) - float(sim_value) + return { + "schema": "telemetry-state-residual-v1", + "values": values, + "missing_common_fields": sorted(missing), + "coverage": { + "available": len(values), + "missing": len(missing), + "common_field_count": len(real["common"]), + }, + } diff --git a/runs/telemetry-residual/run_frontier_state.py b/runs/telemetry-residual/run_frontier_state.py new file mode 100644 index 0000000..2fde9c8 --- /dev/null +++ b/runs/telemetry-residual/run_frontier_state.py @@ -0,0 +1,316 @@ +#!/usr/bin/env python3 +"""Replay frozen Frontier fixtures with state-detail outputs enabled. + +This runner does not modify the frozen fixture/config inputs. It reuses the +audited SimFid command and only turns on existing Frontier batch/ledger output +flags in a separate result root. +""" + +from __future__ import annotations + +import argparse +import hashlib +import importlib.util +import json +import os +import subprocess +import sys +import time +from pathlib import Path +from typing import Any + + +HERE = Path(__file__).resolve().parent +AITUNER_ROOT = HERE.parents[1] +sys.path.insert(0, str(HERE)) + +from common_state import numeric, summarize_frontier # noqa: E402, F401 + + +def load_module(name: str, path: Path): + module_root = str(path.parent.resolve()) + if module_root not in sys.path: + sys.path.insert(0, module_root) + spec = importlib.util.spec_from_file_location(name, path) + if spec is None or spec.loader is None: + raise ImportError(path) + module = importlib.util.module_from_spec(spec) + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def sha256_file(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as source: + for chunk in iter(lambda: source.read(1 << 20), b""): + digest.update(chunk) + return digest.hexdigest() + + +def git_capture(root: Path, *arguments: str) -> str: + return subprocess.run( + ["git", "-C", str(root), *arguments], + check=True, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + ).stdout + + +def atomic_json(path: Path, payload: Any) -> None: + path.parent.mkdir(parents=True, exist_ok=True) + temporary = path.with_suffix(path.suffix + ".tmp") + temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n") + os.replace(temporary, path) + + +def enable_state_outputs(command: list[str]) -> list[str]: + result = list(command) + disabled_ledger = "--no-metrics_config_store_frontier_stage_batch_ledger" + enabled_ledger = "--metrics_config_store_frontier_stage_batch_ledger" + enabled_batches = "--metrics_config_keep_individual_batch_metrics" + disabled_batches = "--no-metrics_config_keep_individual_batch_metrics" + if result.count(disabled_ledger) != 1 or enabled_ledger in result: + raise ValueError("base command must explicitly disable one full Frontier ledger") + result[result.index(disabled_ledger)] = enabled_ledger + if disabled_batches in result: + if result.count(disabled_batches) != 1: + raise ValueError("duplicate disabled individual-batch flag") + result[result.index(disabled_batches)] = enabled_batches + elif enabled_batches not in result: + result.append(enabled_batches) + if result.count(enabled_ledger) != 1 or result.count(enabled_batches) != 1: + raise ValueError("state-detail flags were not enabled exactly once") + return result + + +def find_state_metrics(run_root: Path) -> dict[str, Path]: + patterns = { + "system": "frontier_metrics/**/system_metrics.json", + "requests": "frontier_metrics/**/request_metrics.csv", + "batches": "frontier_metrics/**/monolithic_batch_metrics.csv", + "ledger": "frontier_metrics/**/frontier_stage_batch_ledger.jsonl", + } + result = {} + for name, pattern in patterns.items(): + matches = sorted(run_root.glob(pattern)) + if len(matches) != 1: + raise ValueError( + f"{run_root}: expected exactly one {name} artifact, found {len(matches)}" + ) + result[name] = matches[0] + if len({path.parent for path in result.values()}) != 1: + raise ValueError("Frontier state artifacts are not colocated") + return result + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + prepared = json.loads(args.prepared_manifest.read_text(encoding="utf-8")) + if prepared.get("status") != "PASS": + raise RuntimeError("prepared simulator manifest did not pass") + matches = [ + entry + for entry in prepared["entries"] + if entry["cell"] == args.cell and entry["role"] == args.role + ] + if len(matches) != 1: + raise ValueError( + f"expected one prepared entry for {args.cell}/{args.role}, found {len(matches)}" + ) + entry = matches[0] + if args.output.exists() and any(args.output.iterdir()): + raise FileExistsError(f"non-empty output already exists: {args.output}") + args.output.mkdir(parents=True, exist_ok=True) + + driver = load_module( + "telemetry_residual_execution_driver", + args.replayserve_root / "runs/simfid_s2rb/results/execution_driver.py", + ) + config_path = Path(entry["config"]) + config = json.loads(config_path.read_text(encoding="utf-8")) + fixture_manifest_path = Path(entry["fixture_manifest"]) + fixture = json.loads(fixture_manifest_path.read_text(encoding="utf-8")) + trace_path = Path(entry["frontier_csv"]) + sidecar_path = Path(entry["sidecar"]) + metrics_root = args.output / "frontier_metrics" + run_id = f"telemetry_residual_r0_{args.cell}_{args.role}" + knobs = config["frontier"]["knobs"] + base_command = driver.build_command( + trace_path=trace_path, + metrics_root=metrics_root, + run_id=run_id, + knobs=knobs, + ) + driver.audit_command(base_command, knobs) + command = enable_state_outputs(base_command) + row = { + "hook_path": config["calibration"]["hook_path"], + "applied_a_tp": config["calibration"]["a_tp"], + "sidecar_path": str(sidecar_path), + "request_count": int(fixture["request_count"]), + "tensor_parallel_size": int(fixture["tensor_parallel_size"]), + } + environment = driver.environment_for(row) + if environment.get("CUDA_VISIBLE_DEVICES") != "": + raise ValueError("Frontier state replay must hide CUDA devices") + manifest = { + "schema": "telemetry-residual-frontier-state-run-v1", + "entry": { + "cell": args.cell, + "role": args.role, + "anchor": entry["anchor"], + "request_count": entry["selected_count"], + }, + "inputs": { + "prepared_manifest": str(args.prepared_manifest.resolve()), + "prepared_manifest_sha256": sha256_file(args.prepared_manifest), + "config": str(config_path.resolve()), + "config_sha256": sha256_file(config_path), + "fixture_manifest": str(fixture_manifest_path.resolve()), + "fixture_manifest_sha256": sha256_file(fixture_manifest_path), + "frontier_csv": str(trace_path.resolve()), + "frontier_csv_sha256": sha256_file(trace_path), + "sidecar": str(sidecar_path.resolve()), + "sidecar_sha256": sha256_file(sidecar_path), + }, + "frontier": { + "root": str(args.frontier_root.resolve()), + "git_head": git_capture(args.frontier_root, "rev-parse", "HEAD").strip(), + "git_status_short": git_capture(args.frontier_root, "status", "--short"), + }, + "runner": { + "script": str(Path(__file__).resolve()), + "script_sha256": sha256_file(Path(__file__).resolve()), + "aituner_git_head": git_capture(AITUNER_ROOT, "rev-parse", "HEAD").strip(), + "aituner_git_status_short": git_capture(AITUNER_ROOT, "status", "--short"), + }, + "environment": { + key: environment[key] + for key in ( + "PYTHONPATH", + "FRONTIER_EXECUTION_TIME_SCALE", + "CUDA_VISIBLE_DEVICES", + "NVIDIA_VISIBLE_DEVICES", + "FRONTIER_LOG_LEVEL", + ) + }, + "command": command, + "state_outputs": { + "individual_batch_metrics": True, + "full_stage_batch_ledger": True, + }, + "contains_prompt_text": False, + } + atomic_json(args.output / "run_manifest.json", manifest) + start = time.monotonic() + with (args.output / "stdout.log").open("w", encoding="utf-8") as stdout, ( + args.output / "stderr.log" + ).open("w", encoding="utf-8") as stderr: + try: + process = subprocess.run( + command, + cwd=args.frontier_root, + env=environment, + stdout=stdout, + stderr=stderr, + timeout=args.timeout_s, + ) + return_code = int(process.returncode) + except subprocess.TimeoutExpired: + return_code = 124 + runtime_s = time.monotonic() - start + if return_code != 0: + failure = { + "status": "STOP", + "return_code": return_code, + "runtime_s": runtime_s, + } + atomic_json(args.output / "failure.json", failure) + raise RuntimeError(f"Frontier state replay failed: {failure}") + + paths = find_state_metrics(args.output) + summary = summarize_frontier( + system_metrics_path=paths["system"], + request_metrics_path=paths["requests"], + batch_metrics_path=paths["batches"], + ledger_path=paths["ledger"], + ) + atomic_json(args.output / "common-state.json", summary) + scorer = driver.score_trial(row, paths["system"], paths["requests"]) + atomic_json(args.output / "scorer_output.json", scorer) + sizes = {name: path.stat().st_size for name, path in paths.items()} + red_flags = [] + if summary["interval"]["request_count"] != int(entry["selected_count"]): + red_flags.append("request_count_mismatch") + if summary["sanity"]["batch_rows"] <= 0: + red_flags.append("no_batch_rows") + if summary["sanity"]["ledger_rows"] <= 0: + red_flags.append("no_ledger_rows") + result = { + "schema": "telemetry-residual-frontier-state-result-v1", + "status": "PASS" if not red_flags else "STOP", + "runtime_s": runtime_s, + "return_code": return_code, + "paths": {name: str(path.resolve()) for name, path in paths.items()}, + "bytes": sizes, + "common_state": str((args.output / "common-state.json").resolve()), + "red_flags": red_flags, + "sanity": { + "request_rows": summary["sanity"]["request_rows"], + "batch_rows": summary["sanity"]["batch_rows"], + "ledger_rows": summary["sanity"]["ledger_rows"], + "artifact_bytes": { + "n": len(sizes), + "min": min(sizes.values()), + "max": max(sizes.values()), + "distinct_n": len(set(sizes.values())), + }, + "invariants": { + "zero_failures": return_code == 0, + "gpu_visibility_disabled": True, + "request_count_match": not red_flags, + "nonnegative_file_sizes": all(size >= 0 for size in sizes.values()), + "state_values_not_all_identical": summary["sanity"]["invariants"][ + "batch_values_not_all_identical" + ], + }, + }, + } + atomic_json(args.output / "result.json", result) + if result["status"] != "PASS": + raise RuntimeError(red_flags) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--prepared-manifest", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + result.add_argument("--replayserve-root", type=Path, required=True) + result.add_argument("--frontier-root", type=Path, required=True) + result.add_argument("--cell", required=True) + result.add_argument("--role", choices=("low1", "high1"), required=True) + result.add_argument("--timeout-s", type=float, default=300.0) + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "runtime_s": result["runtime_s"], + "request_rows": result["sanity"]["request_rows"], + "batch_rows": result["sanity"]["batch_rows"], + "ledger_rows": result["sanity"]["ledger_rows"], + "red_flags": result["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/run_frontier_state_campaign.py b/runs/telemetry-residual/run_frontier_state_campaign.py new file mode 100644 index 0000000..7b9ae0e --- /dev/null +++ b/runs/telemetry-residual/run_frontier_state_campaign.py @@ -0,0 +1,169 @@ +#!/usr/bin/env python3 +"""Run the frozen 12-fixture P1 Frontier state-detail campaign, CPU only.""" + +from __future__ import annotations + +import argparse +import json +import sys +import time +from pathlib import Path +from types import SimpleNamespace +from typing import Any + + +HERE = Path(__file__).resolve().parent +sys.path.insert(0, str(HERE)) + +import run_frontier_state as state_runner # noqa: E402 + + +def atomic_json(path: Path, payload: Any) -> None: + state_runner.atomic_json(path, payload) + + +def scorer_without_runtime(value: dict[str, Any]) -> dict[str, Any]: + result = dict(value) + result.pop("runtime_s", None) + return result + + +def execute(args: argparse.Namespace) -> dict[str, Any]: + prepared = json.loads(args.prepared_manifest.read_text(encoding="utf-8")) + committed = json.loads(args.committed_results.read_text(encoding="utf-8")) + if prepared.get("status") != "PASS" or committed.get("status") != "PASS": + raise RuntimeError("prepared or committed simulator evidence did not pass") + committed_by_key = { + (row["cell"], row["role"]): scorer_without_runtime(row["scorer"]) + for row in committed["results"] + } + entries = prepared["entries"] + if len(entries) != 12 or len(committed_by_key) != 12: + raise ValueError("P1 state campaign requires exactly 12 fixtures") + args.output.mkdir(parents=True, exist_ok=True) + results = [] + campaign_start = time.monotonic() + for index, entry in enumerate(entries, start=1): + key = (entry["cell"], entry["role"]) + output = args.output / f"{entry['cell']}_{entry['role']}" + result_path = output / "result.json" + if args.resume and result_path.is_file(): + result = json.loads(result_path.read_text(encoding="utf-8")) + if result.get("status") != "PASS": + raise RuntimeError(f"cannot resume failed state replay: {result_path}") + resumed = True + else: + if output.exists() and any(output.iterdir()): + raise FileExistsError(f"non-empty state replay output: {output}") + print( + f"RUN {index:02d}/12 {entry['cell']}/{entry['role']}", + flush=True, + ) + result = state_runner.execute( + SimpleNamespace( + prepared_manifest=args.prepared_manifest, + output=output, + replayserve_root=args.replayserve_root, + frontier_root=args.frontier_root, + cell=entry["cell"], + role=entry["role"], + timeout_s=args.timeout_s, + ) + ) + resumed = False + observed_scorer = json.loads( + (output / "scorer_output.json").read_text(encoding="utf-8") + ) + exact_scorer_match = scorer_without_runtime(observed_scorer) == committed_by_key[key] + if not exact_scorer_match: + raise ValueError(f"state-output replay changed the committed scorer: {key}") + results.append( + { + "cell": entry["cell"], + "role": entry["role"], + "runtime_s": result["runtime_s"], + "request_rows": result["sanity"]["request_rows"], + "batch_rows": result["sanity"]["batch_rows"], + "ledger_rows": result["sanity"]["ledger_rows"], + "artifact_bytes": sum(result["bytes"].values()), + "exact_committed_scorer_match": exact_scorer_match, + "resumed": resumed, + "result": str(result_path.resolve()), + } + ) + print( + f"DONE {index:02d}/12 {entry['cell']}/{entry['role']} " + f"runtime={result['runtime_s']:.3f}s batches={result['sanity']['batch_rows']}", + flush=True, + ) + runtimes = [float(row["runtime_s"]) for row in results] + batches = [int(row["batch_rows"]) for row in results] + bytes_values = [int(row["artifact_bytes"]) for row in results] + red_flags = [] + if len(results) != 12: + red_flags.append("runs_not_12") + if not all(row["exact_committed_scorer_match"] for row in results): + red_flags.append("committed_scorer_mismatch") + if any(value <= 0 for value in batches): + red_flags.append("empty_batch_output") + result = { + "schema": "telemetry-residual-frontier-state-campaign-v1", + "status": "PASS" if not red_flags else "STOP", + "prepared_manifest": str(args.prepared_manifest.resolve()), + "committed_results": str(args.committed_results.resolve()), + "campaign_elapsed_s": time.monotonic() - campaign_start, + "results": results, + "red_flags": red_flags, + "sanity": { + "n": len(results), + "runtime_s": state_runner.numeric(runtimes), + "batch_rows": state_runner.numeric(batches), + "artifact_bytes": state_runner.numeric(bytes_values), + "invariants": { + "runs_12": len(results) == 12, + "zero_failures": not red_flags, + "exact_committed_scorers": all( + row["exact_committed_scorer_match"] for row in results + ), + "nonnegative_counts": all(value > 0 for value in batches), + "per_config_not_identical": len(set(batches)) > 1, + "gpu_visibility_disabled": True, + }, + }, + } + atomic_json(args.output / "campaign-metrics.json", result) + if result["status"] != "PASS": + raise RuntimeError(red_flags) + return result + + +def parser() -> argparse.ArgumentParser: + result = argparse.ArgumentParser() + result.add_argument("--prepared-manifest", type=Path, required=True) + result.add_argument("--committed-results", type=Path, required=True) + result.add_argument("--output", type=Path, required=True) + result.add_argument("--replayserve-root", type=Path, required=True) + result.add_argument("--frontier-root", type=Path, required=True) + result.add_argument("--timeout-s", type=float, default=300.0) + result.add_argument("--resume", action="store_true") + return result + + +def main() -> None: + result = execute(parser().parse_args()) + print( + json.dumps( + { + "status": result["status"], + "runs": len(result["results"]), + "elapsed_s": result["campaign_elapsed_s"], + "sanity": result["sanity"], + "red_flags": result["red_flags"], + }, + sort_keys=True, + ) + ) + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/test_common_state.py b/runs/telemetry-residual/test_common_state.py new file mode 100644 index 0000000..175f1c8 --- /dev/null +++ b/runs/telemetry-residual/test_common_state.py @@ -0,0 +1,183 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import csv +import json +import math +import tempfile +from pathlib import Path + +from common_state import numeric, residual, summarize_engine, summarize_frontier +from run_frontier_state import enable_state_outputs + + +def engine_record( + step: int, + timestamp_ns: int, + *, + batch: int, + prefill: int, + decode: int, + waiting: int, + running: int, + kv: float, +) -> dict[str, object]: + return { + "step_index": step, + "submit_mono_ns": timestamp_ns, + "model_executed": True, + "scheduled_requests": batch, + "decode_batch_size": decode, + "prefill_tokens": prefill, + "decode_tokens": decode, + "preemptions": 0, + "queues": {"waiting": waiting, "running": running}, + "kv": {"usage": kv}, + "cudagraph": { + "runtime_mode": "NONE" if step == 0 else "FULL", + "bucket_tokens": prefill + decode + 1, + "padding_tokens": 1, + }, + "dropped_records_before": 0, + } + + +def main() -> None: + assert numeric((-2.0, -1.0))["cv"] >= 0.0 + command = [ + "python3", + "-m", + "frontier.main", + "--no-metrics_config_store_frontier_stage_batch_ledger", + ] + state_command = enable_state_outputs(command) + assert "--metrics_config_store_frontier_stage_batch_ledger" in state_command + assert "--metrics_config_keep_individual_batch_metrics" in state_command + assert "--no-metrics_config_store_frontier_stage_batch_ledger" not in state_command + + engine = summarize_engine( + [ + engine_record( + 0, + 0, + batch=2, + prefill=6, + decode=0, + waiting=1, + running=2, + kv=0.1, + ), + engine_record( + 1, + 1_000_000_000, + batch=4, + prefill=0, + decode=4, + waiting=3, + running=4, + kv=0.3, + ), + ], + start_ns=0, + end_ns=2_000_000_000, + request_count=2, + ) + assert math.isclose(engine["common"]["queue_waiting_mean"], 2.0) + assert math.isclose(engine["common"]["queue_running_mean"], 3.0) + assert math.isclose( + engine["common"]["queue_waiting_time_per_request_ms"], 2000.0 + ) + assert engine["common"]["batch_size"]["distinct_n"] == 2 + assert math.isclose(engine["common"]["prefill_token_fraction"], 0.6) + + with tempfile.TemporaryDirectory() as temporary: + root = Path(temporary) + system = root / "system.json" + requests = root / "requests.csv" + batches = root / "batches.csv" + ledger = root / "ledger.jsonl" + system.write_text( + json.dumps( + { + "throughput_metrics": { + "total_duration_seconds": 2.0, + } + } + ), + encoding="utf-8", + ) + with requests.open("w", encoding="utf-8", newline="") as output: + writer = csv.DictWriter( + output, + fieldnames=[ + "request_e2e_time", + "request_waiting_time_total", + "request_total_preemption_count", + ], + ) + writer.writeheader() + writer.writerow( + { + "request_e2e_time": 2000, + "request_waiting_time_total": 1000, + "request_total_preemption_count": 0, + } + ) + writer.writerow( + { + "request_e2e_time": 3000, + "request_waiting_time_total": 3000, + "request_total_preemption_count": 0, + } + ) + with batches.open("w", encoding="utf-8", newline="") as output: + writer = csv.DictWriter( + output, + fieldnames=[ + "batch_size", + "batch_num_tokens", + "batch_num_prefill_tokens", + "batch_num_decode_tokens", + ], + ) + writer.writeheader() + writer.writerow( + { + "batch_size": 2, + "batch_num_tokens": 6, + "batch_num_prefill_tokens": 6, + "batch_num_decode_tokens": 0, + } + ) + writer.writerow( + { + "batch_size": 4, + "batch_num_tokens": 4, + "batch_num_prefill_tokens": 0, + "batch_num_decode_tokens": 4, + } + ) + ledger.write_text( + json.dumps({"stage_start_ts": 0.0, "stage_end_ts": 1.0}) + "\n" + + json.dumps({"stage_start_ts": 1.0, "stage_end_ts": 2.0}) + + "\n", + encoding="utf-8", + ) + simulator = summarize_frontier( + system_metrics_path=system, + request_metrics_path=requests, + batch_metrics_path=batches, + ledger_path=ledger, + ) + assert math.isclose(simulator["common"]["queue_waiting_mean"], 2.0) + assert math.isclose(simulator["common"]["queue_running_mean"], 0.5) + assert simulator["common"]["batch_size"]["distinct_n"] == 2 + difference = residual(engine, simulator) + assert difference["coverage"]["missing"] == 0 + assert difference["coverage"]["available"] == 16 + assert math.isclose(difference["values"]["queue_waiting_mean"], 0.0) + print("telemetry common state: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/telemetry-residual/test_residual_transfer.py b/runs/telemetry-residual/test_residual_transfer.py new file mode 100644 index 0000000..1944f6a --- /dev/null +++ b/runs/telemetry-residual/test_residual_transfer.py @@ -0,0 +1,67 @@ +#!/usr/bin/env python3 +"""Small structural tests for the cross-config transfer diagnostic.""" + +from __future__ import annotations + +import importlib.util +from pathlib import Path + + +HERE = Path(__file__).resolve().parent +SPEC = importlib.util.spec_from_file_location( + "analyze_residual_transfer", HERE / "analyze_residual_transfer.py" +) +assert SPEC is not None and SPEC.loader is not None +MODULE = importlib.util.module_from_spec(SPEC) +SPEC.loader.exec_module(MODULE) + + +def example(cell: str, level: str, value: float) -> dict: + tp_text, mns_text = cell.split("_") + return { + "cell": cell, + "level": level, + "tp": int(tp_text[2:]), + "mns": int(mns_text[3:]), + "pass_rate_residual": value - 0.5, + "real_pass_rate_rep1": value, + "sim_pass_rate": 0.5, + "offered_req_s_per_gpu": 1.0 if level == "low" else 2.0, + "real_feasible": value >= 0.95, + "sim_feasible": False, + "state_residual": {"values": {"batch.mean": value}}, + "engine": { + "common": {"batch": {"mean": value, "max": value, "cv": 0.0}}, + "engine_only": {"kv": value}, + }, + } + + +def main() -> None: + cells = ("tp1_mns8", "tp1_mns16", "tp2_mns8", "tp2_mns16") + examples = [ + example(cell, level, 0.1 + index / 10.0) + for level in ("low", "high") + for index, cell in enumerate(cells) + ] + rows = MODULE.transitions(examples) + assert len(rows) == 48 + assert any(row["source_level"] != row["target_level"] for row in rows) + assert all(row["source_cell"] != row["target_cell"] for row in rows) + assert len(rows[0]["hybrid_telemetry"]) > len(rows[0]["hybrid_base"]) + assert len(rows[0]["direct_telemetry"]) > len(rows[0]["direct_base"]) + for feature, target in ( + ("hybrid_base", "target_residual_delta"), + ("hybrid_telemetry", "target_residual_delta"), + ("direct_base", "target_real_pass_rate_delta"), + ("direct_telemetry", "target_real_pass_rate_delta"), + ): + prediction = MODULE.grouped_predictions( + rows, feature_name=feature, target_name=target, regularization=1.0 + ) + assert len(prediction) == len(rows) + print("telemetry residual transfer: PASS") + + +if __name__ == "__main__": + main() diff --git a/runs/tuning-cost/analyze.py b/runs/tuning-cost/analyze.py new file mode 100644 index 0000000..be3003a --- /dev/null +++ b/runs/tuning-cost/analyze.py @@ -0,0 +1,507 @@ +#!/usr/bin/env python3 +"""Reconstruct tuning cost and cost-to-oracle curves from existing runs. + +The historical engine logs do not contain controller setup/cleanup timestamps. +Consequently the reported GPU cost is an engine-lifetime lower bound: +parallel_size * (last engine timestamp - first engine timestamp). It must not +be presented as all-in method cost. New experiments should record allocation +start/end directly. +""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +import re +import subprocess +from datetime import datetime +from pathlib import Path +from typing import Any +from urllib.parse import urlsplit + + +SCHEMA = "aituner-tuning-cost-analysis-v1" +TIMESTAMP = re.compile(r"\b(\d{2}-\d{2} \d{2}:\d{2}:\d{2})(?:\.\d+)?\b") + + +def numeric_summary(values: list[float]) -> dict[str, Any]: + values = [float(value) for value in values] + return { + "n": len(values), + "min": min(values) if values else None, + "max": max(values) if values else None, + "distinct_n": len(set(values)), + } + + +def sha256_text(text: str) -> str: + return hashlib.sha256(text.encode("utf-8")).hexdigest() + + +class Reader: + def __init__(self, repo_root: Path): + self.repo_root = repo_root + + def _split(self, locator: str) -> tuple[str | None, str]: + if locator.startswith("ssh://"): + parsed = urlsplit(locator) + if not parsed.hostname or not parsed.path.startswith("/"): + raise ValueError(f"invalid SSH locator: {locator}") + return parsed.hostname, parsed.path + path = Path(locator) + if not path.is_absolute(): + path = self.repo_root / path + return None, str(path) + + def read_text(self, locator: str) -> str: + host, path = self._split(locator) + if host is None: + return Path(path).read_text(encoding="utf-8", errors="replace") + completed = subprocess.run( + ["ssh", host, "cat", "--", path], + check=True, + capture_output=True, + text=True, + ) + return completed.stdout + +def join_locator(root: str, *parts: str) -> str: + return "/".join([root.rstrip("/"), *(part.strip("/") for part in parts)]) + + +def timestamp_span(log_text: str, year: int) -> tuple[float | None, bool, int]: + parsed = [ + datetime.strptime(f"{year}-{match}", "%Y-%m-%d %H:%M:%S") + for match in TIMESTAMP.findall(log_text) + ] + if not parsed: + return None, True, 0 + monotonic = all(right >= left for left, right in zip(parsed, parsed[1:])) + return (max(parsed) - min(parsed)).total_seconds(), monotonic, len(parsed) + + +def load_campaign( + reader: Reader, + root: str, + year: int, + missing_duration_s: dict[str, float] | None = None, + missing_reason: dict[str, str] | None = None, +) -> dict[str, Any]: + missing_duration_s = missing_duration_s or {} + missing_reason = missing_reason or {} + state_text = reader.read_text(join_locator(root, "state.json")) + state = json.loads(state_text) + trials = [] + for trial in state["trials"]: + trial_id = str(trial["trial_id"]) + log_text = reader.read_text(join_locator(root, "trials", trial_id, "engine.log")) + duration_s, monotonic, timestamp_n = timestamp_span(log_text, year) + duration_source = "engine_log_span" + duration_note = "" + if duration_s is None: + duration_s = float(missing_duration_s.get(trial_id, 0.0)) + duration_source = ( + "conservative_fallback" if trial_id in missing_duration_s else "no_engine_timestamp" + ) + duration_note = missing_reason.get(trial_id, "") + parallel_size = int(trial["parallel_size"]) + score = trial.get("best_request_rate_per_gpu") + trials.append( + { + "trial_id": trial_id, + "status": trial["status"], + "failure_stage": trial.get("failure_stage", ""), + "parallel_size": parallel_size, + "score_req_s_per_gpu": None if score is None else float(score), + "config_patch": trial["config_patch"], + "duration_s": duration_s, + "duration_source": duration_source, + "duration_note": duration_note, + "engine_timestamp_n": timestamp_n, + "engine_timestamps_monotonic": monotonic, + "engine_h20_hours_lower_bound": duration_s * parallel_size / 3600.0, + "engine_log_sha256": sha256_text(log_text), + } + ) + return { + "root": root, + "state_sha256": sha256_text(state_text), + "trials": trials, + } + + +def fixed_task_context(spec: dict[str, Any]) -> dict[str, Any]: + flags = dict(spec["engine"]["base_flags"]) + flags.pop("port", None) + return { + "model": spec["model"], + "hardware": spec["hardware"], + "engine_version": spec["engine"]["engine_version"], + "launch_args": spec["engine"]["launch_args"], + "base_flags_without_port": flags, + "search": spec["search"], + "slo": spec["slo"], + "trace": spec["trace"], + } + + +def regret(score: float, reference: float) -> float: + if reference <= 0: + raise ValueError("reference score must be positive") + return max(0.0, 1.0 - float(score) / float(reference)) + + +def sequential_curve( + trials: list[dict[str, Any]], reference: float, thresholds: list[float] +) -> dict[str, Any]: + cumulative_cost = 0.0 + best_score: float | None = None + points = [] + for trial in trials: + cumulative_cost += float(trial["engine_h20_hours_lower_bound"]) + score = trial["score_req_s_per_gpu"] + if score is not None: + best_score = score if best_score is None else max(best_score, score) + points.append( + { + "trial_id": trial["trial_id"], + "cumulative_engine_h20_hours_lower_bound": cumulative_cost, + "best_score_req_s_per_gpu": best_score, + "regret": None if best_score is None else regret(best_score, reference), + } + ) + hits = {} + for threshold in thresholds: + hit = next( + ( + point + for point in points + if point["regret"] is not None + and float(point["regret"]) <= float(threshold) + 1e-12 + ), + None, + ) + hits[f"regret_le_{threshold:g}"] = hit + return { + "reference_score_req_s_per_gpu": reference, + "points": points, + "cost_to_threshold": hits, + "total_engine_h20_hours_lower_bound": cumulative_cost, + } + + +def surface_cell(trial: dict[str, Any]) -> str: + flags = trial["config_patch"]["flag_patch"] + return f"tp{int(flags['tensor-parallel-size'])}_mns{int(flags['max-num-seqs'])}" + + +def tie_expanded_candidates(scores: dict[str, float], nominal_k: int) -> list[str]: + ordered = sorted(scores, key=lambda cell: (-float(scores[cell]), cell)) + if nominal_k <= 0 or nominal_k > len(ordered): + raise ValueError("nominal k outside score surface") + cutoff = float(scores[ordered[nominal_k - 1]]) + tolerance = max(1e-12, abs(cutoff) * 1e-12) + return [cell for cell in ordered if float(scores[cell]) >= cutoff - tolerance] + + +def real_final_policy( + candidates: list[str], real_scores: dict[str, float], cell_costs: dict[str, float] +) -> dict[str, Any]: + oracle = max(real_scores.values()) + selected = max(candidates, key=lambda cell: (real_scores[cell], cell)) + return { + "candidate_cells": candidates, + "real_evaluations": len(candidates), + "selected_cell": selected, + "selected_real_score_req_s_per_gpu": real_scores[selected], + "real_regret": regret(real_scores[selected], oracle), + "engine_h20_hours_lower_bound": sum(cell_costs[cell] for cell in candidates), + } + + +def percentage_saving(new: float, old: float) -> float: + if old <= 0: + raise ValueError("baseline cost must be positive") + return 1.0 - new / old + + +def build_analysis(manifest_path: Path) -> dict[str, Any]: + repo_root = manifest_path.resolve().parents[2] + manifest = json.loads(manifest_path.read_text(encoding="utf-8")) + reader = Reader(repo_root) + year = int(manifest["year"]) + + sequential = {} + task_contexts = {} + for name, run in manifest["sequential_runs"].items(): + sequential[name] = load_campaign( + reader, + run["root"], + year, + run.get("missing_log_duration_s"), + run.get("missing_log_reason"), + ) + spec_text = reader.read_text(join_locator(run["root"], "study_spec.snapshot.json")) + sequential[name]["study_spec_sha256"] = sha256_text(spec_text) + task_contexts[name] = fixed_task_context(json.loads(spec_text)) + + all_sequential_scores = [ + trial["score_req_s_per_gpu"] + for campaign in sequential.values() + for trial in campaign["trials"] + if trial["score_req_s_per_gpu"] is not None + ] + empirical_reference = max(all_sequential_scores) + thresholds = [float(value) for value in manifest["threshold_regrets"]] + for campaign in sequential.values(): + campaign["curve"] = sequential_curve( + campaign["trials"], empirical_reference, thresholds + ) + + surface_manifest = manifest["real_surface"] + primary = load_campaign(reader, surface_manifest["primary_root"], year) + companion = load_campaign(reader, surface_manifest["tp4_companion_root"], year) + completed_surface_trials = [ + trial + for trial in primary["trials"] + companion["trials"] + if trial["status"] == "completed" and trial["score_req_s_per_gpu"] is not None + ] + cells: dict[str, dict[str, Any]] = {} + for trial in completed_surface_trials: + flags = trial["config_patch"]["flag_patch"] + if int(flags["max-num-batched-tokens"]) != int( + surface_manifest["fixed_max_num_batched_tokens"] + ): + raise ValueError("surface MBT invariant failed") + cell = surface_cell(trial) + if cell in cells: + raise ValueError(f"duplicate completed surface cell: {cell}") + cells[cell] = trial + real_scores = {cell: float(trial["score_req_s_per_gpu"]) for cell, trial in cells.items()} + cell_costs = { + cell: float(trial["engine_h20_hours_lower_bound"]) for cell, trial in cells.items() + } + surface_oracle_score = max(real_scores.values()) + surface_oracle_cells = [ + cell for cell, score in real_scores.items() if math.isclose(score, surface_oracle_score) + ] + + simulator_text = reader.read_text(manifest["simulator_metrics"]) + simulator = json.loads(simulator_text) + simulator_real_scores = { + cell: float(score) for cell, score in simulator["real_scores"].items() + } + surface_matches_simulator = set(real_scores) == set(simulator_real_scores) and all( + math.isclose(real_scores[cell], simulator_real_scores[cell], abs_tol=1e-12) + for cell in real_scores + ) + throughput = simulator["analyses"]["frozen-calibrated/throughput-proxy"] + throughput_scores = { + cell: float(score) for cell, score in throughput["simulated_scores"].items() + } + throughput_real_final = { + f"nominal_k_{k}": real_final_policy( + tie_expanded_candidates(throughput_scores, k), real_scores, cell_costs + ) + for k in (1, 2, 3) + } + throughput_top1 = tie_expanded_candidates(throughput_scores, 1) + simulator_only_cell = throughput_top1[0] + slo = simulator["analyses"]["frozen-calibrated/SLO-gated"] + slo_top_bucket = list(slo["metrics"]["top1"]["candidate_cells"]) + slo_diagnostic = real_final_policy(slo_top_bucket, real_scores, cell_costs) + + pure_hits = sequential["pure_llm"]["curve"]["cost_to_threshold"] + guided_hits = sequential["guided_harness"]["curve"]["cost_to_threshold"] + direct_comparison = {} + for threshold in (0.05, 0.02): + key = f"regret_le_{threshold:g}" + pure_hit = pure_hits[key] + guided_hit = guided_hits[key] + if pure_hit is None or guided_hit is None: + continue + pure_cost = float(pure_hit["cumulative_engine_h20_hours_lower_bound"]) + guided_cost = float(guided_hit["cumulative_engine_h20_hours_lower_bound"]) + direct_comparison[key] = { + "pure_llm_h20_hours_lower_bound": pure_cost, + "guided_harness_h20_hours_lower_bound": guided_cost, + "guided_saving_vs_pure_llm": percentage_saving(guided_cost, pure_cost), + } + + five_cost = direct_comparison["regret_le_0.05"][ + "guided_harness_h20_hours_lower_bound" + ] + two_cost = direct_comparison["regret_le_0.02"][ + "guided_harness_h20_hours_lower_bound" + ] + sim_real_cost = float(slo_diagnostic["engine_h20_hours_lower_bound"]) + target_bars = { + "five_percent_regret": { + "twenty_percent_below_current_guided": 0.8 * five_cost, + "thirty_percent_below_posthoc_sim_slo_real_final": 0.7 * sim_real_cost, + "development_target_h20_hours_lower_bound": min( + 0.8 * five_cost, 0.7 * sim_real_cost + ), + }, + "two_percent_regret": { + "twenty_percent_below_current_guided": 0.8 * two_cost, + "thirty_percent_below_posthoc_sim_slo_real_final": 0.7 * sim_real_cost, + "development_target_h20_hours_lower_bound": min( + 0.8 * two_cost, 0.7 * sim_real_cost + ), + }, + } + + all_trial_costs = [ + float(trial["engine_h20_hours_lower_bound"]) + for campaign in sequential.values() + for trial in campaign["trials"] + ] + [ + float(trial["engine_h20_hours_lower_bound"]) + for campaign in (primary, companion) + for trial in campaign["trials"] + ] + all_regrets = [ + float(point["regret"]) + for campaign in sequential.values() + for point in campaign["curve"]["points"] + if point["regret"] is not None + ] + monotonic_logs = all( + trial["engine_timestamps_monotonic"] + for campaign in [*sequential.values(), primary, companion] + for trial in campaign["trials"] + ) + invariants = { + "dash0_task_contexts_equal_except_method_and_port": len( + {json.dumps(value, sort_keys=True) for value in task_contexts.values()} + ) + == 1, + "surface_has_expected_cell_count": len(cells) + == int(surface_manifest["expected_cells"]), + "surface_matches_simulator_real_scores": surface_matches_simulator, + "all_costs_non_negative": all(value >= 0 for value in all_trial_costs), + "all_regrets_in_0_1": all(0 <= value <= 1 for value in all_regrets), + "surface_scores_not_all_identical": len(set(real_scores.values())) > 1, + "sequential_scores_not_all_identical": len(set(all_sequential_scores)) > 1, + "engine_log_timestamps_monotonic": monotonic_logs, + "sequential_trial_counts_match_manifest": all( + len(sequential[name]["trials"]) == int(run["expected_trials"]) + for name, run in manifest["sequential_runs"].items() + ), + "simulator_suite_has_no_failed_runs": int(simulator["execution"]["failed_runs"]) + == 0, + "simulator_scores_not_all_identical": len(set(throughput_scores.values())) > 1, + } + failed_invariants = [name for name, passed in invariants.items() if not passed] + if failed_invariants: + raise RuntimeError(f"data sanity invariant failed: {failed_invariants}") + + failed_primary_attempts = [ + trial for trial in primary["trials"] if trial["status"] != "completed" + ] + return { + "schema": SCHEMA, + "cost_definition": { + "reported_metric": "engine H20-hours lower bound", + "formula": "parallel_size * (last_engine_log_timestamp - first_engine_log_timestamp) / 3600", + "included": ["engine startup after first timestamp", "warm-up/probes until last timestamp"], + "not_reconstructable": [ + "GPU allocation before first engine timestamp", + "controller/LLM latency", + "cleanup after last engine timestamp", + "one-time simulator operator profiling GPU-hours", + ], + "future_all_in_metric": "allocation_start_to_GPU_idle * allocated_GPU_count, including failures", + }, + "comparison_scope": { + "pure_llm_vs_guided_harness": "direct: same dash0 fixed task context", + "simulator_vs_surface": "direct: simulator predictions and exact dash1 12-cell real surface", + "dash0_methods_vs_dash1_simulator": "indicative only: matched model/engine/workload/GPU type, different host and campaign", + }, + "empirical_reference": { + "score_req_s_per_gpu": empirical_reference, + "meaning": "best observed across the two dash0 sequential runs; not a global oracle", + }, + "sequential_runs": sequential, + "direct_dash0_comparison": direct_comparison, + "real_surface": { + "cells": { + cell: { + "score_req_s_per_gpu": real_scores[cell], + "engine_h20_hours_lower_bound": cell_costs[cell], + } + for cell in sorted(cells) + }, + "oracle_score_req_s_per_gpu": surface_oracle_score, + "oracle_cells": surface_oracle_cells, + "completed_annotation_engine_h20_hours_lower_bound": sum(cell_costs.values()), + "failed_primary_attempt_n": len(failed_primary_attempts), + "failed_primary_attempt_engine_h20_hours_lower_bound": sum( + float(trial["engine_h20_hours_lower_bound"]) + for trial in failed_primary_attempts + ), + "primary_state_sha256": primary["state_sha256"], + "tp4_companion_state_sha256": companion["state_sha256"], + }, + "simulator": { + "marginal_gpu_hours_without_real_verification": 0.0, + "observed_fidelity_suite_cpu_hours": float( + simulator["execution"]["suite_elapsed_seconds"] + ) + / 3600.0, + "observed_fidelity_suite_runs": int(simulator["execution"]["attempted_runs"]), + "one_time_profile_gpu_hours": None, + "one_time_profile_cost_status": "not recorded; total cold-start cost is unknown", + "decision_bearing_throughput_proxy_sim_only_top1": { + "selected_cell": simulator_only_cell, + "selected_real_score_req_s_per_gpu": real_scores[simulator_only_cell], + "real_regret": regret(real_scores[simulator_only_cell], surface_oracle_score), + "gpu_hours": 0.0, + }, + "decision_bearing_throughput_proxy_plus_real_final": throughput_real_final, + "posthoc_slo_gated_plus_real_final": { + **slo_diagnostic, + "status": "diagnostic/post-hoc, not a preregistered prospective policy", + "false_feasible": int(slo["false_feasibility"]["overall"]["false_feasible"]), + "false_infeasible": int(slo["false_feasibility"]["overall"]["false_infeasible"]), + }, + "metrics_sha256": sha256_text(simulator_text), + }, + "provisional_development_targets": { + "status": "lower-bound, single-task targets; require same-host prospective validation", + **target_bars, + }, + "data_sanity": { + "invariants": invariants, + "sequential_score_summary": numeric_summary(all_sequential_scores), + "surface_score_summary": numeric_summary(list(real_scores.values())), + "simulator_throughput_score_summary": numeric_summary( + list(throughput_scores.values()) + ), + "trial_cost_summary": numeric_summary(all_trial_costs), + "regret_summary": numeric_summary(all_regrets), + }, + } + + +def main() -> None: + parser = argparse.ArgumentParser() + parser.add_argument("--manifest", type=Path, default=Path(__file__).with_name("manifest.json")) + parser.add_argument("--output", type=Path, default=Path(__file__).with_name("metrics.json")) + args = parser.parse_args() + analysis = build_analysis(args.manifest) + args.output.write_text(json.dumps(analysis, indent=2, sort_keys=True) + "\n", encoding="utf-8") + print(json.dumps({ + "status": "ok", + "output": str(args.output), + "empirical_reference": analysis["empirical_reference"], + "surface_oracle": analysis["real_surface"]["oracle_score_req_s_per_gpu"], + "sanity": analysis["data_sanity"], + }, indent=2, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/tuning-cost/manifest.json b/runs/tuning-cost/manifest.json new file mode 100644 index 0000000..fabfb53 --- /dev/null +++ b/runs/tuning-cost/manifest.json @@ -0,0 +1,42 @@ +{ + "schema": "aituner-tuning-cost-v1", + "year": 2026, + "task": { + "model": "Qwen/Qwen3-30B-A3B", + "engine": "community-vLLM 0.20.0", + "gpu": "NVIDIA H20", + "trace_window": "chat_w20260311_1000", + "input_tokens": [0, 8192], + "output_tokens": 128, + "replay_time_scale": 0.1, + "target_pass_rate": 0.95, + "ttft_ms": [2000, 4000, 6000], + "tpot_ms": 50 + }, + "sequential_runs": { + "pure_llm": { + "root": "ssh://dash0/home/admin/cpfs/wjh/aituner/aituner/.aituner-community-vllm020/dash0-qwen30b-a3b-community-vllm020-chat-0-8k-out128-scale01-high1-noharness", + "expected_trials": 12, + "missing_log_duration_s": { + "trial-0003": 13.0 + }, + "missing_log_reason": { + "trial-0003": "conservative trial_spec-to-result mtime envelope for a pre-ready CLI failure" + } + }, + "guided_harness": { + "root": "ssh://dash0/home/admin/cpfs/wjh/aituner/aituner/.aituner-community-vllm020/dash0-qwen30b-a3b-community-vllm020-chat-0-8k-out128-scale01-high1-harness-guided-v2", + "expected_trials": 4, + "missing_log_duration_s": {}, + "missing_log_reason": {} + } + }, + "real_surface": { + "primary_root": "recovered-stores/aituner-interaction-runs-dash1-20260710/interaction-mixed-qwen30b-tp-mns-surface-high1-dash1-d8899c5-20260701T095858Z/store/interaction-mixed-qwen30b-tp-mns-surface-high1-dash1-d8899c5-20260701T095858Z", + "tp4_companion_root": "recovered-stores/aituner-interaction-runs-dash1-20260710/interaction-mixed-qwen30b-tp4-mns-nocap-qps20-dash1-d8899c5-20260701T161900Z/store/interaction-mixed-qwen30b-tp4-mns-nocap-qps20-dash1-d8899c5-20260701T161900Z", + "expected_cells": 12, + "fixed_max_num_batched_tokens": 8192 + }, + "simulator_metrics": "/home/gahow/phd/replayserve/runs/simfid_s2rb/results/metrics.json", + "threshold_regrets": [0.05, 0.02, 0.01, 0.0] +} diff --git a/runs/tuning-cost/metrics.json b/runs/tuning-cost/metrics.json new file mode 100644 index 0000000..e246bea --- /dev/null +++ b/runs/tuning-cost/metrics.json @@ -0,0 +1,736 @@ +{ + "comparison_scope": { + "dash0_methods_vs_dash1_simulator": "indicative only: matched model/engine/workload/GPU type, different host and campaign", + "pure_llm_vs_guided_harness": "direct: same dash0 fixed task context", + "simulator_vs_surface": "direct: simulator predictions and exact dash1 12-cell real surface" + }, + "cost_definition": { + "formula": "parallel_size * (last_engine_log_timestamp - first_engine_log_timestamp) / 3600", + "future_all_in_metric": "allocation_start_to_GPU_idle * allocated_GPU_count, including failures", + "included": [ + "engine startup after first timestamp", + "warm-up/probes until last timestamp" + ], + "not_reconstructable": [ + "GPU allocation before first engine timestamp", + "controller/LLM latency", + "cleanup after last engine timestamp", + "one-time simulator operator profiling GPU-hours" + ], + "reported_metric": "engine H20-hours lower bound" + }, + "data_sanity": { + "invariants": { + "all_costs_non_negative": true, + "all_regrets_in_0_1": true, + "dash0_task_contexts_equal_except_method_and_port": true, + "engine_log_timestamps_monotonic": true, + "sequential_scores_not_all_identical": true, + "sequential_trial_counts_match_manifest": true, + "simulator_scores_not_all_identical": true, + "simulator_suite_has_no_failed_runs": true, + "surface_has_expected_cell_count": true, + "surface_matches_simulator_real_scores": true, + "surface_scores_not_all_identical": true + }, + "regret_summary": { + "distinct_n": 6, + "max": 0.34328358208955223, + "min": 0.0, + "n": 16 + }, + "sequential_score_summary": { + "distinct_n": 7, + "max": 3.35, + "min": 1.1041666666666667, + "n": 9 + }, + "simulator_throughput_score_summary": { + "distinct_n": 10, + "max": 4.356763578770651, + "min": 1.5449814460277083, + "n": 12 + }, + "surface_score_summary": { + "distinct_n": 8, + "max": 3.283333333333333, + "min": 1.2833333333333334, + "n": 12 + }, + "trial_cost_summary": { + "distinct_n": 26, + "max": 0.49777777777777776, + "min": 0.0, + "n": 32 + } + }, + "direct_dash0_comparison": { + "regret_le_0.02": { + "guided_harness_h20_hours_lower_bound": 0.4458333333333333, + "guided_saving_vs_pure_llm": 0.6109090909090908, + "pure_llm_h20_hours_lower_bound": 1.1458333333333333 + }, + "regret_le_0.05": { + "guided_harness_h20_hours_lower_bound": 0.26805555555555555, + "guided_saving_vs_pure_llm": 0.0585365853658536, + "pure_llm_h20_hours_lower_bound": 0.2847222222222222 + } + }, + "empirical_reference": { + "meaning": "best observed across the two dash0 sequential runs; not a global oracle", + "score_req_s_per_gpu": 3.35 + }, + "provisional_development_targets": { + "five_percent_regret": { + "development_target_h20_hours_lower_bound": 0.21444444444444444, + "thirty_percent_below_posthoc_sim_slo_real_final": 0.36088888888888887, + "twenty_percent_below_current_guided": 0.21444444444444444 + }, + "status": "lower-bound, single-task targets; require same-host prospective validation", + "two_percent_regret": { + "development_target_h20_hours_lower_bound": 0.3566666666666667, + "thirty_percent_below_posthoc_sim_slo_real_final": 0.36088888888888887, + "twenty_percent_below_current_guided": 0.3566666666666667 + } + }, + "real_surface": { + "cells": { + "tp1_mns16": { + "engine_h20_hours_lower_bound": 0.14083333333333334, + "score_req_s_per_gpu": 2.35 + }, + "tp1_mns32": { + "engine_h20_hours_lower_bound": 0.13194444444444445, + "score_req_s_per_gpu": 2.283333333333333 + }, + "tp1_mns64": { + "engine_h20_hours_lower_bound": 0.13527777777777777, + "score_req_s_per_gpu": 2.283333333333333 + }, + "tp1_mns8": { + "engine_h20_hours_lower_bound": 0.13333333333333333, + "score_req_s_per_gpu": 2.1 + }, + "tp2_mns16": { + "engine_h20_hours_lower_bound": 0.29944444444444446, + "score_req_s_per_gpu": 2.275 + }, + "tp2_mns32": { + "engine_h20_hours_lower_bound": 0.2816666666666667, + "score_req_s_per_gpu": 3.283333333333333 + }, + "tp2_mns64": { + "engine_h20_hours_lower_bound": 0.2338888888888889, + "score_req_s_per_gpu": 3.2583333333333333 + }, + "tp2_mns8": { + "engine_h20_hours_lower_bound": 0.2777777777777778, + "score_req_s_per_gpu": 2.275 + }, + "tp4_mns16": { + "engine_h20_hours_lower_bound": 0.4866666666666667, + "score_req_s_per_gpu": 2.441666666666667 + }, + "tp4_mns32": { + "engine_h20_hours_lower_bound": 0.49777777777777776, + "score_req_s_per_gpu": 2.441666666666667 + }, + "tp4_mns64": { + "engine_h20_hours_lower_bound": 0.49666666666666665, + "score_req_s_per_gpu": 2.441666666666667 + }, + "tp4_mns8": { + "engine_h20_hours_lower_bound": 0.48, + "score_req_s_per_gpu": 1.2833333333333334 + } + }, + "completed_annotation_engine_h20_hours_lower_bound": 3.5952777777777776, + "failed_primary_attempt_engine_h20_hours_lower_bound": 0.0, + "failed_primary_attempt_n": 4, + "oracle_cells": [ + "tp2_mns32" + ], + "oracle_score_req_s_per_gpu": 3.283333333333333, + "primary_state_sha256": "790bf6df9045e22c46a8abdc022390bf842d884ef13ade4b627936722b9c3366", + "tp4_companion_state_sha256": "1fc4f584a69a90fb0ff1d64f6a07a8383c368f4e306a0b7a3c19d2304042fd73" + }, + "schema": "aituner-tuning-cost-analysis-v1", + "sequential_runs": { + "guided_harness": { + "curve": { + "cost_to_threshold": { + "regret_le_0": null, + "regret_le_0.01": null, + "regret_le_0.02": { + "best_score_req_s_per_gpu": 3.283333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.4458333333333333, + "regret": 0.01990049751243783, + "trial_id": "trial-0003" + }, + "regret_le_0.05": { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.26805555555555555, + "regret": 0.027363184079602032, + "trial_id": "trial-0002" + } + }, + "points": [ + { + "best_score_req_s_per_gpu": 2.3833333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.09305555555555556, + "regret": 0.2885572139303483, + "trial_id": "trial-0001" + }, + { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.26805555555555555, + "regret": 0.027363184079602032, + "trial_id": "trial-0002" + }, + { + "best_score_req_s_per_gpu": 3.283333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.4458333333333333, + "regret": 0.01990049751243783, + "trial_id": "trial-0003" + }, + { + "best_score_req_s_per_gpu": 3.3, + "cumulative_engine_h20_hours_lower_bound": 0.6230555555555555, + "regret": 0.014925373134328401, + "trial_id": "trial-0004" + } + ], + "reference_score_req_s_per_gpu": 3.35, + "total_engine_h20_hours_lower_bound": 0.6230555555555555 + }, + "root": "ssh://dash0/home/admin/cpfs/wjh/aituner/aituner/.aituner-community-vllm020/dash0-qwen30b-a3b-community-vllm020-chat-0-8k-out128-scale01-high1-harness-guided-v2", + "state_sha256": "76e439d1f20e0f9af54785005258cbe39f205ae107fc2398955e360e263c7718", + "study_spec_sha256": "03241089052f1a07dbbb2ba6100c6737635877b683d275813b3c59f2769fc2d7", + "trials": [ + { + "config_patch": { + "env_patch": {}, + "flag_patch": {} + }, + "duration_note": "", + "duration_s": 335.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.09305555555555556, + "engine_log_sha256": "6d8b908085c03d4796e78679eed13737581fc7d2e18947fb72d7031a9f557c9d", + "engine_timestamp_n": 108, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 1, + "score_req_s_per_gpu": 2.3833333333333333, + "status": "completed", + "trial_id": "trial-0001" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 315.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.175, + "engine_log_sha256": "b09caadd930495c27effea57e4164e52f58c6461cbbc0f4e873a309d45ec3443", + "engine_timestamp_n": 134, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.2583333333333333, + "status": "completed", + "trial_id": "trial-0002" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 16384, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 320.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.17777777777777778, + "engine_log_sha256": "8f6f2637cff8566d39b8e6c17de71ed37a01f2aa6a4460ff36ecc9728d033c3a", + "engine_timestamp_n": 134, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.283333333333333, + "status": "completed", + "trial_id": "trial-0003" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 24576, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 319.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.17722222222222223, + "engine_log_sha256": "29b8fc588b5422ceae60456188d6373070c179610243258fa213aeead9f07a00", + "engine_timestamp_n": 137, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.3, + "status": "completed", + "trial_id": "trial-0004" + } + ] + }, + "pure_llm": { + "curve": { + "cost_to_threshold": { + "regret_le_0": { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.3719444444444444, + "regret": 0.0, + "trial_id": "trial-0007" + }, + "regret_le_0.01": { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.3719444444444444, + "regret": 0.0, + "trial_id": "trial-0007" + }, + "regret_le_0.02": { + "best_score_req_s_per_gpu": 3.3, + "cumulative_engine_h20_hours_lower_bound": 1.1458333333333333, + "regret": 0.014925373134328401, + "trial_id": "trial-0006" + }, + "regret_le_0.05": { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.2847222222222222, + "regret": 0.027363184079602032, + "trial_id": "trial-0002" + } + }, + "points": [ + { + "best_score_req_s_per_gpu": 2.2, + "cumulative_engine_h20_hours_lower_bound": 0.10861111111111112, + "regret": 0.34328358208955223, + "trial_id": "trial-0001" + }, + { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.2847222222222222, + "regret": 0.027363184079602032, + "trial_id": "trial-0002" + }, + { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.29194444444444445, + "regret": 0.027363184079602032, + "trial_id": "trial-0003" + }, + { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.4280555555555555, + "regret": 0.027363184079602032, + "trial_id": "trial-0004" + }, + { + "best_score_req_s_per_gpu": 3.2583333333333333, + "cumulative_engine_h20_hours_lower_bound": 0.9247222222222222, + "regret": 0.027363184079602032, + "trial_id": "trial-0005" + }, + { + "best_score_req_s_per_gpu": 3.3, + "cumulative_engine_h20_hours_lower_bound": 1.1458333333333333, + "regret": 0.014925373134328401, + "trial_id": "trial-0006" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.3719444444444444, + "regret": 0.0, + "trial_id": "trial-0007" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.561388888888889, + "regret": 0.0, + "trial_id": "trial-0008" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.7497222222222222, + "regret": 0.0, + "trial_id": "trial-0009" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 1.9047222222222222, + "regret": 0.0, + "trial_id": "trial-0010" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 2.0930555555555554, + "regret": 0.0, + "trial_id": "trial-0011" + }, + { + "best_score_req_s_per_gpu": 3.35, + "cumulative_engine_h20_hours_lower_bound": 2.2824999999999998, + "regret": 0.0, + "trial_id": "trial-0012" + } + ], + "reference_score_req_s_per_gpu": 3.35, + "total_engine_h20_hours_lower_bound": 2.2824999999999998 + }, + "root": "ssh://dash0/home/admin/cpfs/wjh/aituner/aituner/.aituner-community-vllm020/dash0-qwen30b-a3b-community-vllm020-chat-0-8k-out128-scale01-high1-noharness", + "state_sha256": "342cde3546299cd312ebc078b1c9589773b841d23c4ed4632370883be598fd4a", + "study_spec_sha256": "3238dc89477108405c09b25eee434c3c10c0f5be245d2bfa3b90509a372c088e", + "trials": [ + { + "config_patch": { + "env_patch": {}, + "flag_patch": {} + }, + "duration_note": "", + "duration_s": 391.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.10861111111111112, + "engine_log_sha256": "c5caf9e9f6faa73a4a39adb71a7a6c04bcf623d5c0c91b231116bb596c8a11be", + "engine_timestamp_n": 113, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 1, + "score_req_s_per_gpu": 2.2, + "status": "completed", + "trial_id": "trial-0001" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 317.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.1761111111111111, + "engine_log_sha256": "239ff7349d7b9a1b24334fbe6e2a45a967bfa2d16380e98da4f81520d422cd0d", + "engine_timestamp_n": 134, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.2583333333333333, + "status": "completed", + "trial_id": "trial-0002" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-expert-parallel": true, + "expert-parallel-size": 2, + "tensor-parallel-size": 2 + } + }, + "duration_note": "conservative trial_spec-to-result mtime envelope for a pre-ready CLI failure", + "duration_s": 13.0, + "duration_source": "conservative_fallback", + "engine_h20_hours_lower_bound": 0.007222222222222222, + "engine_log_sha256": "4707735f33c8bf9030054eccd36b66985a3879050d80a853f93f166282414d8f", + "engine_timestamp_n": 0, + "engine_timestamps_monotonic": true, + "failure_stage": "engine_launch", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "failed", + "trial_id": "trial-0003" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "data-parallel-size": 2, + "tensor-parallel-size": 1 + } + }, + "duration_note": "", + "duration_s": 245.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.1361111111111111, + "engine_log_sha256": "215ed31c6d227e4a6771ecb637c2b4e3e38e766d73a8882a923d1fa8b97a23da", + "engine_timestamp_n": 170, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0004" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "data-parallel-size": 2, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 447.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.49666666666666665, + "engine_log_sha256": "6057b3477ccca14e0ebd467ac96b777662df67b57827c851d58caf3a77b95ca4", + "engine_timestamp_n": 201, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 4, + "score_req_s_per_gpu": 1.1041666666666667, + "status": "completed", + "trial_id": "trial-0005" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 16384, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 398.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.22111111111111112, + "engine_log_sha256": "c5c336f74d419645639b469eea9f92575780164637fdda078fa5c84c96915f1f", + "engine_timestamp_n": 138, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.3, + "status": "completed", + "trial_id": "trial-0006" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 24576, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 407.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.22611111111111112, + "engine_log_sha256": "5e2829479ecd4771e7c781d40209f74e4c2843bbca15522974e435e06daa12a2", + "engine_timestamp_n": 142, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": 3.35, + "status": "completed", + "trial_id": "trial-0007" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 32768, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 341.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.18944444444444444, + "engine_log_sha256": "fde628968fba10fccfe5b09675aa57cf725f507be83c10dd3c0b6c109b5d72eb", + "engine_timestamp_n": 136, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0008" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 28672, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 339.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.18833333333333332, + "engine_log_sha256": "fc60765578d57af08e0310983ff42e2a200739a90f11c45f9d4b1f6b7b789cc8", + "engine_timestamp_n": 136, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0009" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "enable-prefix-caching": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 24576, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 279.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.155, + "engine_log_sha256": "22c473b39662df6c8c4c64a616346652ab889d63256efab18257b1f4445ce13b", + "engine_timestamp_n": 133, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0010" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 25600, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 339.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.18833333333333332, + "engine_log_sha256": "19b19de4a0a11a5a112b3e8ef04c3ca4d90d6c6c9097b5180c024d8570c9a608", + "engine_timestamp_n": 136, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0011" + }, + { + "config_patch": { + "env_patch": {}, + "flag_patch": { + "enable-chunked-prefill": true, + "gpu-memory-utilization": 0.95, + "max-num-batched-tokens": 25088, + "tensor-parallel-size": 2 + } + }, + "duration_note": "", + "duration_s": 341.0, + "duration_source": "engine_log_span", + "engine_h20_hours_lower_bound": 0.18944444444444444, + "engine_log_sha256": "40f7307d6677f2bb55c0132d42fc82fce78ffa6a558d5c69b01ec60d1e40b3ae", + "engine_timestamp_n": 136, + "engine_timestamps_monotonic": true, + "failure_stage": "", + "parallel_size": 2, + "score_req_s_per_gpu": null, + "status": "completed", + "trial_id": "trial-0012" + } + ] + } + }, + "simulator": { + "decision_bearing_throughput_proxy_plus_real_final": { + "nominal_k_1": { + "candidate_cells": [ + "tp1_mns64" + ], + "engine_h20_hours_lower_bound": 0.13527777777777777, + "real_evaluations": 1, + "real_regret": 0.30456852791878175, + "selected_cell": "tp1_mns64", + "selected_real_score_req_s_per_gpu": 2.283333333333333 + }, + "nominal_k_2": { + "candidate_cells": [ + "tp1_mns64", + "tp1_mns32" + ], + "engine_h20_hours_lower_bound": 0.26722222222222225, + "real_evaluations": 2, + "real_regret": 0.30456852791878175, + "selected_cell": "tp1_mns64", + "selected_real_score_req_s_per_gpu": 2.283333333333333 + }, + "nominal_k_3": { + "candidate_cells": [ + "tp1_mns64", + "tp1_mns32", + "tp2_mns32", + "tp2_mns64" + ], + "engine_h20_hours_lower_bound": 0.7827777777777778, + "real_evaluations": 4, + "real_regret": 0.0, + "selected_cell": "tp2_mns32", + "selected_real_score_req_s_per_gpu": 3.283333333333333 + } + }, + "decision_bearing_throughput_proxy_sim_only_top1": { + "gpu_hours": 0.0, + "real_regret": 0.30456852791878175, + "selected_cell": "tp1_mns64", + "selected_real_score_req_s_per_gpu": 2.283333333333333 + }, + "marginal_gpu_hours_without_real_verification": 0.0, + "metrics_sha256": "55edb37d5692e979ab6f6dc6c65913a9db0aa0a836c350e4c05d9c38eee78206", + "observed_fidelity_suite_cpu_hours": 2.055026211017717, + "observed_fidelity_suite_runs": 184, + "one_time_profile_cost_status": "not recorded; total cold-start cost is unknown", + "one_time_profile_gpu_hours": null, + "posthoc_slo_gated_plus_real_final": { + "candidate_cells": [ + "tp2_mns32", + "tp2_mns64" + ], + "engine_h20_hours_lower_bound": 0.5155555555555555, + "false_feasible": 21, + "false_infeasible": 7, + "real_evaluations": 2, + "real_regret": 0.0, + "selected_cell": "tp2_mns32", + "selected_real_score_req_s_per_gpu": 3.283333333333333, + "status": "diagnostic/post-hoc, not a preregistered prospective policy" + } + } +} diff --git a/runs/tuning-cost/test_analysis.py b/runs/tuning-cost/test_analysis.py new file mode 100644 index 0000000..ca698c7 --- /dev/null +++ b/runs/tuning-cost/test_analysis.py @@ -0,0 +1,57 @@ +#!/usr/bin/env python3 +from __future__ import annotations + +import importlib.util +import math +import sys +from pathlib import Path + + +HERE = Path(__file__).resolve().parent + + +def load_analysis(): + spec = importlib.util.spec_from_file_location("tuning_cost", HERE / "analyze.py") + module = importlib.util.module_from_spec(spec) + assert spec.loader is not None + sys.modules[spec.name] = module + spec.loader.exec_module(module) + return module + + +def main() -> None: + analysis = load_analysis() + duration, monotonic, count = analysis.timestamp_span( + "INFO 07-01 10:00:00 x\nINFO 07-01 10:05:30 y\n", 2026 + ) + assert duration == 330.0 + assert monotonic + assert count == 2 + + trials = [ + {"trial_id": "t1", "engine_h20_hours_lower_bound": 0.1, "score_req_s_per_gpu": 8.0}, + {"trial_id": "t2", "engine_h20_hours_lower_bound": 0.2, "score_req_s_per_gpu": None}, + {"trial_id": "t3", "engine_h20_hours_lower_bound": 0.3, "score_req_s_per_gpu": 9.6}, + ] + curve = analysis.sequential_curve(trials, 10.0, [0.05, 0.04]) + assert curve["cost_to_threshold"]["regret_le_0.05"]["trial_id"] == "t3" + assert curve["cost_to_threshold"]["regret_le_0.04"]["trial_id"] == "t3" + assert math.isclose(curve["total_engine_h20_hours_lower_bound"], 0.6) + + candidates = analysis.tie_expanded_candidates( + {"a": 3.0, "b": 2.0, "c": 2.0, "d": 1.0}, 2 + ) + assert candidates == ["a", "b", "c"] + policy = analysis.real_final_policy( + candidates, + {"a": 1.0, "b": 4.0, "c": 2.0, "d": 3.0}, + {"a": 0.1, "b": 0.2, "c": 0.3, "d": 0.4}, + ) + assert policy["selected_cell"] == "b" + assert policy["real_regret"] == 0.0 + assert math.isclose(policy["engine_h20_hours_lower_bound"], 0.6) + print("tuning cost analysis: PASS") + + +if __name__ == "__main__": + main() diff --git a/simulator-fidelity.md b/simulator-fidelity.md new file mode 100644 index 0000000..b646d5b --- /dev/null +++ b/simulator-fidelity.md @@ -0,0 +1,443 @@ +# Frontier simulator fidelity：以 config ranking 为目标的阶段性评测 + +更新日期：2026-07-17。 + +统一实验平台：所有新增与重跑实验只使用 `dash0` 的 8×NVIDIA H20。Qwen30B 的真实 P1 artifacts 也来自 `dash0`；早期文档中的主机 provenance 标注错误，本版已按实验主机和远端 artifact 路径更正。 + +## 结论摘要 + +本文统一使用同一个优化目标：**在给定 workload 和 SLO 下，最大被测试可行 offered request throughput/GPU**。我们关心 simulator 是否保持 config 的相对排序、是否覆盖真机最优集合，以及部署 simulator 所选 config 后的真实 regret；绝对 latency/throughput 误差不是首要判据。 + +当前证据不支持“Frontier 普遍找不到最优 config”这一判断： + +- 在 Qwen3-30B mixed serving artifact 上，未经端到端校准的 Frontier 排序失败，选择 regret 为 `25.63%`；但使用独立 workload 上冻结的 per-TP calibration 后，Frontier 的候选 top set 包含真机最优 config，Kendall τ-b 为 `0.9668`，最坏 tie-break regret 为 `0.76%`。真实 serving 与本轮补测 profiles 均运行在 `dash0`。 +- 将 operator profile 更新为与真实 serving 一致的 community vLLM `0.20.0`、BF16、H20、TP1/2/4 栈后，新的 profile-only Frontier **没有恢复排序**：92 个真实 anchors 在 simulator 中全部 SLO-infeasible，12 个 config 因而全部并列，最坏 tie-break regret 为 `60.91%`。这否证了“旧 profile 版本不一致是主要原因”这一简单解释，并把问题收敛到 execution context、operator composition 与 mixed-state schema。 +- 在 Qwen3-235B-A22B-FP8 prefill-only 上，补齐 FP8/MoE profile 与 serving semantics、但不做端到端 action calibration 后，Frontier 的最优集合与真机完全一致，Spearman ρ 为 `0.9487`，20/20 个可比较非 tie config pair 同序，选择 regret 为 `0`。 +- 在 Qwen3-235B-A22B-FP8 fixed-shape mixed 上，Frontier 同样给出了与真机完全相同的 TP4 top set，Kendall τ-b=`0.8944`、worst tie-break regret=`0`。但这次成功掩盖了一个明确的机制错误：Frontier 认为同一 TP family 内四个 MNS/MBT config 完全等价，真机 TP8 capacity 却形成对角为 0.30、非对角为 0.20 req/s/GPU 的 checkerboard interaction。34 个 config-load labels 中有 10 个 false-infeasible；20 个 real non-tie pairs 中只保持 16 个方向。 +- 新增的 Qwen3-30B-A3B BF16 prefill-only 对照在没有 decode、prefix reuse 和 true-mixed attention 的情况下仍然失败：真机 top set 是四个 TP4 config，Frontier top set 却是全部 TP1/TP2 config，top set 无交集，worst regret=`12.5%`、Kendall τ-b=`-1.0`，32 个 real non-tie pairs 全部反向。这直接否证了“prefill-only 是 simulator ranking 充分容易条件”。 +- 现有结果更支持一个 **margin-aware fidelity** 解释：action-differential residual 可以很大，但只有当它足以穿过 real decision margin 时才会导致选错。Qwen235 mixed 中真机 TP4 相对最好 TP8 的 margin 是 `2×`，足以掩盖同 TP family 内的机制错误；Qwen30 prefill-only 中 TP scaling 差分更细，Frontier 的 residual 直接反转了 topology ordering。模型大小或 execution phase 单独都不足以解释现有结果。 + +因此，现阶段最准确的 research statement 不是“simulator 排序一定错误”，而是： + +> Simulator 的 usefulness 取决于 action-differential residual 是否小于真实 decision margin。问题不是简单区分 prefill 与 mixed，而是定位哪些 scheduler-state-conditioned execution residual 会随 TP、batch composition 和负载被放大，并用最少的真机信息判断 ranking 是否足够可信。 + +## 评测口径 + +对系统 `s ∈ {real, sim}` 和 config `c`，定义： + +```text +capacity_s(c) = max { tested offered rate r | empirical SLO pass rate(c, r) >= 0.95 } +``` + +所有结果使用 `capacity/实际占用 GPU 数` 作为 primary score。我们只报告实际测试到的最大可行点，不对未测试区间做连续插值，也不因观察到局部非单调就强制单调化。 + +核心指标如下： + +- **Top set**：达到各自最大 capacity 的全部 config，保留并列。 +- **Top-set hit/match**：simulator top set 是否与 real top set 有交集/完全相同。 +- **Worst tie-break regret**：若 simulator 给出多个并列最优，任取其中最差一个在真机上的损失： + + ```text + max[c in SimTop] (RealBest - capacity_real(c)) / RealBest + ``` + +- **Rank correlation**：有并列时优先使用 Kendall τ-b；Qwen235B 已有冻结分析使用 Spearman ρ，同时报告非 tie pair direction。 +- **Pairwise direction**：两两 config 的 `> / = / <` 是否一致。只报告 non-tied accuracy 时，会同时给出可比较 pair 数量。 + +这个定义也修正了之前的口径混用：内部 `throughput proxy` 不是“满足 SLO 的最大 throughput”，不能用来决定部署 config。本文的 Qwen30B 主结论因此取代 [旧 simulator fidelity 总结](docs/simulator-fidelity-frontier-20260711.md) 中基于 throughput-proxy 的 config-selection 结论；该旧读法仍可作为 simulator 内部机制误差的诊断证据。 + +## 实验 setup + +### Case A：Qwen3-30B-A3B mixed serving + +| 项目 | 设置 | +|---|---| +| 真机 | `dash0`，8×NVIDIA H20 | +| model | `Qwen/Qwen3-30B-A3B` | +| serving runtime | community vLLM `0.20.0`，CUDA 12.9 环境 | +| simulator | Frontier commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e` | +| trace | `chat_w20260311_1000`，replay time scale=`0.1`，input 0--8192 tokens，output override=128 tokens，max concurrency=64 | +| SLO | 至少 95% requests 同时满足 TTFT 阶梯阈值（input ≤4096: 2s；≤32768: 4s；其余 6s）和 TPOT ≤50ms | +| config surface | `TP∈{1,2,4} × MNS∈{8,16,32,64}`，MBT=8192，共 12 cells | +| ground truth | 每个 cell 最多 8 个搜索 probes；92 个实测 anchors，共 27,583 个 request observations | +| simulator workload | 对每个 real anchor 重放同一冻结 request cohort 的 token length、arrival、request count 与 prefix block identities；old profile-only、calibrated、vLLM 0.20 profile-only 各 92 runs，合计 276/276 成功 | +| primary objective | SLO-feasible request throughput/GPU | + +我们区分两个 Frontier 条件： + +1. **Profile-only**：直接使用已有 H20 operator profiles，不用 serving E2E 数据修正时间尺度。 +2. **Frozen per-TP calibration**：使用 `TP1=0.723481, TP2=0.468089, TP4=0.352137` 三个 execution-time scales。它们在独立的 `coder_200_ts2` workload 上拟合，并在 `coder_200_ts3` 上冻结验证；本 `chat` surface 不重新拟合。该 scale 由外部 hook 在 Frontier 组成 batch execution time 后注入，并不是 Frontier 原生的 per-TP operator profiling 功能。 + +Frontier 原生 profile schema 可以按 `num_tensor_parallel_workers` 选择 TP-specific kernel rows；这与上面的 calibration 不同。第二个条件不是 zero-measurement prediction：它使用了同 model/hardware/runtime family 的真机 serving 数据，并能同时吸收 CUDA graph、fusion、routing、CPU launch 和 profile provenance 等残差。另一个限制是已有 H20 profiles 的生成栈与 ground-truth vLLM 0.20.0 并非严格同版本；per-TP calibration 吸收了相当一部分 execution-scale mismatch。 + +### Case B：Qwen3-235B-A22B-FP8 prefill-only + +| 项目 | 设置 | +|---|---| +| 真机 | `dash0`，8×NVIDIA H20（97,871 MiB/GPU，全互联 NV18），driver `580.95.05` | +| model | `Qwen/Qwen3-235B-A22B-FP8`，block-wise FP8 weights，BF16 KV cache | +| serving runtime | community vLLM `0.10.2`；Python 3.12.3、torch 2.8.0+cu128、FlashInfer 0.3.1.post1、Transformers 4.55.2；eager execution | +| simulator | Frontier commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + 冻结的 FP8/MoE best-effort patches | +| source trace | `thinking_w20260327_1000`；从 15,409 个 eligible requests 中固定选择 64 个 length-stratified prompts | +| fixed cohort | seed=`2026071501`；input p50=1478、p95=18,439、max=29,385、mean=3511.55 tokens；output override=1 token | +| offered rates | 0.15, 0.20, 0.25, 0.30, 0.35, 0.40, 0.45, 0.50, 0.75, 1.00, 1.50 req/s；只缩放同一 cohort 的 arrival timeline | +| SLO | 至少 61/64 requests 满足 `TTFT ≤ 1000ms + input_tokens/8000 tokens/s` | +| config surface | `TP∈{4,8} × MNS∈{64,128} × MBT∈{8192,16384}`；实际 TP4 为 EP off，TP8 为 EP on | +| serving controls | prefix cache/speculative decoding off，chunked prefill on；每个 config fresh server，第二轮反转 config/rate 顺序 | +| primary objective | 最大被测试可行 offered rate / TP | + +这里比较的不是 stock Frontier。为建立公平的 FP8 compatibility envelope，我们先用同一 community-vLLM 0.10.2 栈在 H20 上测量 FP8 linear/attention/MoE/collective profiles，修复了 MoE FP8 tuning-key、Qwen MoE serving plan、TP/EP-aware cache key 与 critical-lane prediction，并输入真机测得的 KV block capacity（TP4=26,101；TP8=62,351）。Frontier refined outputs 在真机 v2/v3 ranking 运行前冻结，**没有使用本 case 的 serving capacity 做 end-to-end calibration**。 + +该 case 是有意设计的 prefill mechanism-isolation workload：output 固定为 1，prefix cache 关闭，并使用固定 64-request cohort。它不保持原 trace 的 output-length distribution 和 prefix reuse，因此不能被称为 trace-faithful workload，也不能代表 mixed serving。 + +### Case D：Qwen3-30B-A3B BF16 prefill-only + +| 项目 | 设置 | +|---|---| +| 真机 | `dash0`，8×NVIDIA H20；每个 job 按 TP 独占 GPU | +| model/runtime | `Qwen/Qwen3-30B-A3B`；community vLLM `0.20.0+cu129`（source `88d34c6409e9fb3c7b8ca0c04756f061d2099eb1`），BF16 weights/activation/KV，FA3，runtime 默认 CUDA graph | +| workload | 64 个不同 token-chain prompts；ISL=2,048、OSL=1、uniform open-loop QPS、prefix cache off | +| config surface | `TP∈{1,2,4} × MNS∈{8,16,32,64}`，MBT=8,192，chunked prefill on，共 12 cells | +| simulator | Frontier commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + 当前 compatibility patches + vLLM 0.20 same-stack `profile-v2`；无 E2E calibration、decode graph=`none`、analytical collective | +| load lattice | base system rates `{4,8,16,32,64}` req/s；事先声明的 common refinement 为 `{5,6,7}` req/s/GPU | +| real anchor | 每个 `(config,rate,round)` fresh server；两轮独立运行且第二轮反转顺序；两轮都 pass 才算 feasible | +| SLO | 至少 61/64 requests 满足 TTFT≤1,256 ms | +| primary objective | 最大被测试可行 offered req/s / 实际 TP GPUs | + +这个 case 是对“prefill-only 是 simulator 的容易区间”的直接可证伪测试。它刻意去掉 decode、true-mixed batch、prefix reuse 和 KV-residency feedback，但保留 TP 改变下的 local shard、collective、batching 与 scheduler queue。Frontier surface 在查看真机 ranking 前冻结，没有使用本 surface 的 serving latency 拟合 scale。 + +## Workload fidelity 修正 + +`thinking_w20260327_1000` 的 600 秒原始窗口包含 15,479 requests，字段包括 exact prompt、arrival、`input_length`、`output_length`、session parent/turn 和 block-size=64 的 `hash_ids`。只读审计得到： + +| 属性 | 数值 | +|---|---:| +| natural offered rate | 25.798 req/s | +| input tokens | mean=3,660.0，p50=1,491，p95=19,610.6，p99=29,944.6，max=51,827 | +| output tokens | mean=3,924.6，p50=3,435，p95=8,945.1，p99=17,705.5，max=74,867 | +| total context | mean=7,584.6，p50=5,317，p95=25,102.1，p99=37,110.5，max=74,973 | +| rows with `hash_ids` | 15,479/15,479 | +| `len(hash_ids) == ceil(input_length/64)` | 15,479/15,479 | +| arrival-order repeated block ratio | 40.88% | +| input+output 超过 community model 40,960 context | 72 requests | +| output length=0 | 6 requests | + +因此，input/output/hash filtering 不是 trace fidelity 的默认要求。此前 fixed output length 是为了分别隔离 prefill 或 decode 机制；如果用它回答 production-trace fidelity，就属于错误 benchmark。新的实验矩阵固定为： + +| 层次 | Workload contract | Prefix contract | 回答的问题 | +|---|---|---|---| +| T0 fixed-shape | 固定 ISL、OSL、uniform QPS；所有 config 使用相同 request IDs/order | prefix off，使用等长但不同 token 的 prompts | 最简单条件下 execution/scheduler composition 是否匹配 | +| T1 trace-faithful mixed | 保留 trace 的 arrival、input、output、session；不使用 completion override | real 由 exact prompt tokens 自然计算 cache key；Frontier 消费等价 `block_hash_ids` | 在真实 joint distribution 下能否保持 config rank | +| T2 decode-only | output 保留 trace 或受控固定值；input length 表示 request 到达时已存在的 KV residency | initial-KV state 单独冻结；不得用一次 1-token prefill 冒充 | KV residency、decode batching、TP/DP/EP coupling 下能否保持 rank | + +T1 中只允许两类有记录的排除：模型 context 上限不支持的 72 requests，以及 API 无法形成 completion 的 6 个 zero-output rows；二者不重叠，最终 eligible universe 为 15,401/15,479 requests（99.50%）。排除后 input/output/total 的均值分别为 3,575.0/3,823.3/7,398.2 tokens，p95 分别为 18,887/8,768/23,697，说明没有把 workload 人为变成 short-request case。其余请求不能因为“跑得慢”或“方便形成 cohort”被选择性移除。capacity search 只通过 trace 已有、同 session 共享的 `sampling_u` 做与长度无关的 session-coherent thinning；每个被选 request 的 arrival、input、output、prompt、hash 和相对次序均保持不变。完整 eligible universe 是抽样母体和最终 workload 声明，不用 length-stratified cohort 代替。 + +全量 tokenizer audit 进一步验证了 15,401/15,401 个 eligible prompts、共 55,057,919 tokens 的实际 community-Qwen token 数与 trace `input_length` 完全一致。在 867,538 个 source blocks 上，`hash_id` 与 `(parent_hash_id, 64-token chunk)` 的双向映射也没有发现冲突。因此 exact prompt 和 source hash 足以保留该 trace 的 prefix-equivalence relation，不需要用 synthetic length-only prompt 替代。 + +trace 的 source hash block size 为 64，而 community vLLM 0.10.2 的 CUDA KV block size 不能直接设置为 64。下一轮实验冻结 real 与 Frontier 的 block size 都为 16：从 exact prompt token IDs 重新生成 block-16 parent/content identities，并用 source `hash_ids` 验证每四个完整 block-16 所对应的 64-token equality/reuse relation。tokenized length、request order、cache query/hit/allocated blocks 仍必须在两侧逐请求核对；上面的离线 audit 不能代替 runtime counter parity。 + +## 总体结果 + +| Case / simulator 条件 | configs | real optimum | simulator top set | top-set 结论 | rank / pairwise | worst regret | 判断 | +|---|---:|---|---|---|---|---:|---| +| Qwen30 mixed / old profile-only | 12 | TP2,MNS32 | TP4,MNS32/64 | miss | τ-b=0.000；exact sign=37.88% | 25.63% | 排序错误 | +| Qwen30 mixed / vLLM 0.20 profile-only | 12 | TP2,MNS32 | 全部 12 configs | 不可辨识 | τ-b=0.000；exact sign=7.58% | 60.91% | 同栈 raw profile 仍不足 | +| Qwen30 mixed / frozen per-TP calibration | 12 | TP2,MNS32 | TP2,MNS32/64 | hit，非 exact | τ-b=0.9668；exact sign=93.94% | 0.76% | dash0 | +| Qwen30 BF16 prefill-only / vLLM 0.20 profile-only | 12 | 全部 TP4 configs | 全部 TP1/TP2 configs | miss，无交集 | τ-b=-1.0000；real non-tie=0/32 | 12.50% | topology order 反转 | +| Qwen235 FP8 prefill / best-effort | 8 | TP4,MBT16K,MNS64/128 | 与 real 完全相同 | exact match | ρ=0.9487；non-tied 20/20 | 0 | 足以选最优 config | +| Qwen235 FP8 fixed-shape mixed / frozen full profile | 8 | 四个 TP4 configs | 与 real 完全相同 | exact match | τ-b=0.8944；exact sign=24/28；real non-tie=16/20 | 0 | 选对 topology；漏掉 TP8 MNS×MBT interaction | + +最重要的区别是：Qwen30 的 high-fidelity mixed 结果依赖 per-TP E2E calibration，而新 prefill-only profile-only 实验证明“去掉 decode/mixed 状态”也不足以恢复 ranking。Qwen235 的 high-fidelity 结果不依赖本 case 的 E2E calibration，但依赖同 runtime/hardware 的 operator profile、真实 KV capacity 和多处 compatibility fixes。两个 model 的 stack 和精度不同，所以不能把差异直接归因为 model size；二者也都不能表述为“拿 stock Frontier 零成本预测即可”。 + +## Case A baseline 结果：Qwen30 mixed serving + +![Qwen3-30B mixed serving simulator-vs-real config ranking](docs/assets/simulator-fidelity/qwen30-mixed-config-ranking.png) + +图中 config 按真机 capacity 从高到低排列。上图显示 profile-only Frontier 选错 TP family；下图显示冻结 per-TP calibration 后，simulator 基本恢复了有用排序。 + +| Config | Real | Frontier profile-only | Frontier calibrated | Calibrated − real | +|---|---:|---:|---:|---:| +| TP1, MNS8 | 2.1000 | 1.1000 | 1.7167 | -0.3833 | +| TP1, MNS16 | 2.3500 | 1.1000 | 2.3833 | +0.0333 | +| TP1, MNS32 | 2.2833 | 1.1000 | 2.3833 | +0.1000 | +| TP1, MNS64 | 2.2833 | 1.1000 | 2.3833 | +0.1000 | +| TP2, MNS8 | 2.2750 | 0.0000 | 1.7417 | -0.5333 | +| TP2, MNS16 | 2.2750 | 1.1917 | 2.3000 | +0.0250 | +| **TP2, MNS32** | **3.2833** | 0.0000 | **3.7500** | +0.4667 | +| TP2, MNS64 | 3.2583 | 0.0000 | **3.7500** | +0.4917 | +| TP4, MNS8 | 1.2833 | 0.0000 | 1.3208 | +0.0375 | +| TP4, MNS16 | 2.4417 | 0.0000 | 2.5000 | +0.0583 | +| TP4, MNS32 | 2.4417 | **1.3208** | 2.5000 | +0.0583 | +| TP4, MNS64 | 2.4417 | **1.3208** | 2.5000 | +0.0583 | + +单位均为 SLO-feasible req/s/GPU。`0.0000` 表示该 config 在复用的 tested anchors 中没有被 Frontier 判为 SLO-feasible，不代表硬件在零负载下也无法运行。 + +相对排序方面： + +- Profile-only 的 top set 为 TP4/MNS32、TP4/MNS64；两者真机 capacity 都是 2.4417，而 real optimum TP2/MNS32 是 3.2833，因此 regret 为 `1−2.4417/3.2833=25.63%`。 +- Calibrated Frontier 把 TP2/MNS32 和 TP2/MNS64 判为并列。前者是真机唯一最优，后者低 `0.025 req/s/GPU`；因此 optimistic regret=0，worst tie-break regret=`0.76%`。 +- Calibrated absolute error 并不是一个全局常数：TP2/MNS8 被低估 0.5333，而 TP2/MNS32/64 被高估约 0.47--0.49。**选对 config 不代表 simulator 已正确解释每个 state/action 的性能。** +- 92 个 anchor-level SLO 判定中，calibrated Frontier 仍有 21 个 false-feasible、7 个 false-infeasible；aggregate top set 正确部分来自 cell boundary 上误差抵消。因此该条件足以做本 surface 的 near-optimal selection，但不能替代真机 SLO certification。 + +### 同栈 vLLM 0.20 per-TP profile 消融 + +为了判断上述差异是否只是旧 operator profile 与真实 serving 版本不一致，我们在 `dash0` 上使用 ground-truth 相同的 community vLLM `0.20.0` source commit `88d34c6409e9fb3c7b8ca0c04756f061d2099eb1`、Qwen3-30B-A3B BF16 与 H20，重新测量 TP1/2/4 profiles。Frontier code、92 个 trace fixtures、SLO、KV capacity、config surface、communication backend 与随机种子保持不变；这轮只替换 compute profiles，不使用本 surface 的 serving latency 拟合 scale。 + +| Profile component | 冻结行数 | 测量与使用方式 | +|---|---:|---| +| attention + KV | 132 | 102 个 pure prefill/decode rows；30 个 true-mixed rows由同 TP pure profile 比例做 total-conserving split | +| fused mixed diagnostic | 30 | 保存 vLLM FA3 实测 fused total；不作为 Frontier 的第三个 target | +| linear / norm / RoPE / router | 36 | vLLM 0.20 CUDA operator | +| MoE | 72 | TP1/2/4 local-shard shapes × 12 token sizes × uniform/random/hotset8 routing | +| all-reduce | 24 | TP2/4 实测；本 base comparison 不注入，用来诊断 communication，保持历史 analytical CC backend 不变 | + +attention、linear 和 MoE 的 `TP` 字段改变的是单张 H20 上的 local-shard shape；它们不是一次 distributed full-engine graph replay。只有 all-reduce 使用真实 TP2/4 ranks 测量。该限制是实验设计的一部分：本消融回答“把相同 runtime 的 isolated operator profiles 换进去是否足够”，而不是预先假设 isolated profile 已等价于真实 execution counterfactual。 + +![Qwen3-30B vLLM 0.20 profile-only ablation and execution-context diagnostics](docs/assets/simulator-fidelity/qwen30-vllm020-profile-ablation.png) + +完整 92-probe 结果如下。`0` 表示复用的 anchors 中没有一个被 simulator 判为 SLO-feasible；它既不表示零负载硬件 capacity，也不能被解释成“因为真实最优 config 位于并列集合中所以 top-set hit”。当所有 config 都并列时 simulator 没有选择信息，故 primary 指标是 worst tie-break regret。 + +| Frontier 条件 | Sim top set | τ-b | Exact pair sign | Worst regret | Anchor labels：agree / false feasible / false infeasible | +|---|---|---:|---:|---:|---:| +| old profile-only | TP4,MNS32/64 | 0.0000 | 37.88% | 25.63% | 37 / 0 / 55 | +| vLLM 0.20 profile-only | 全部 12 configs | 0.0000 | 7.58% | 60.91% | 30 / 0 / 62 | +| frozen per-TP calibration | TP2,MNS32/64 | 0.9668 | 93.94% | 0.76% | 64 / 21 / 7 | + +若忽略 SLO，仅按 simulator 完成 throughput proxy 排序，新 profile 的 top-1 是 TP1/MNS64，真机 regret 仍为 `30.46%`，τ-b=`0.3940`，non-tied pair concordance=`70.49%`。这只是 failure diagnosis，不能替代本文定义的 SLO-feasible capacity objective。两路 CPU shard 都是 46/46 PASS，合计 simulator runtime 为 2,394.9 s（0.665 CPU-hour）；所以结果不是 crash、missing row 或未完成运行造成的。 + +#### 为什么相同 operator provenance 仍不等于 execution fidelity + +证据把 mismatch 收敛到以下三个相互耦合的层面： + +1. **Execution regime 随 scheduler state 改变。** 6 个真实 P1 cells 共记录 161,161 个 model steps。151,471 个 pure-decode steps 全部使用 FULL CUDA graph；224 个 pure-prefill steps 中 192 个为 NONE、30 个 PIECEWISE、仅 2 个 FULL；9,466 个 true-mixed steps 中 8,623 个为 NONE、842 个 PIECEWISE、仅 1 个 FULL。本轮 isolated profiles 则关闭 CUDA graph，Frontier comparison 也固定 `decode_cuda_graph_mode=none`。同一个 nominal operator 在不同 phase/state 下并不是同一个 launch/fusion regime。 +2. **Mixed attention schema 不保真。** vLLM 0.20 的 FA3 对 true-mixed batch 执行一个 fused varlen call；Frontier 必须分别得到 `attn_prefill` 与 `attn_decode_in_mixed`。不做 split 时 simulator 在第一个 mixed batch 明确报 missing prediction；采用 pure-profile ratio split 虽保持 `prefill + decode = fused total`，但 split 是兼容性归因而非真实观测。scheduler 使用这个虚构分解推进后续状态，误差会累积。 +3. **MoE routing 不是固定先验。** 对 exact trace 的 8 个 requests、48 层 native routing capture，实际 median load CV 在 prefill/decode 分别为 `1.076/1.107`，median max/mean expert load 为 `6.12/5.64`；Frontier 固定的 per-layer Uniform prior 分别只有 `0.485` 与 `1.84`。实际与 Frontier expert allocation 的逐层 Pearson median 接近 0（prefill `-0.017`、decode `-0.024`），而 Frontier 在所有 phase 复用相同 allocation，prefill/decode correlation 被固定为 1.0；真实 median correlation 为 `0.667`。8-request capture 是 mechanism probe，不是 workload population estimate，但足以否证“固定 routing vector 等价于该 trace execution”的假设。 + +更直接地，在同一个 `tp1_mns64` fixture、`batch_id=0/layer_id=0`、两条 trajectory 尚未分叉时，新 profile 的 layer component sum 已是旧 profile 的 `2.70×`（1.148→3.101 ms）：MoE grouped GEMM 为 `3.76×`、RoPE `7.39×`、attention prefill `2.67×`。这说明版本更新没有产生一个小而一致的 correction；不同 components 的误差方向和幅度不同，随后又通过 scheduler queue、batch composition、graph mode 与 routing 反馈到下一个 state。 + +因此本实验否证 H1：**Frontier 原生支持按 `num_tensor_parallel_workers` 选择 per-TP profile rows，而且 attention/MoE kernel 也确实可以逐 TP 测量；真正缺失的是 scheduler state 与 execution regime 条件化的 joint counterfactual。** TP 和 MNS 不能拆开：TP 改变 shard、collective、replica capacity、routing placement 与 graph shape，MNS 改变 batching/queue evolution；两者共同决定 phase mixture 和 execution mode。历史 per-TP calibration 是 composition 之后的外部 E2E scale，因此能吸收这些残差，但它不是机制解释，也不是 Frontier 原生 profiling。 + +下一轮不应继续盲目加 isolated rows，而应依次做可证伪的 context ablation：按 FULL pure-decode、NONE/PIECEWISE prefill/mixed 建立 graph-conditioned step profiles；保留 fused mixed layer/step total 而不是要求不可观测的拆分；用 trace-conditioned per-layer/request/phase routing histogram 替换固定先验；再单独注入 measured collective。每次只改变一个条件，最终仍在完整 `TP×MNS` joint surface 上以 SLO ranking 验证，frozen calibration 只作为可达到的上界。 + +本次 accepted profile jobs 使用约 0.118 H20-GPU-hour；包括 smoke 和失败 attempt 的完整 campaign 为 0.410 H20-GPU-hour。冻结 profile、manifest、condensed artifacts 与作图脚本位于 `runs/frontier-qwen30-vllm020-profile-v1/`。 + +## Case B 结果：Qwen235 FP8 prefill-only + +![Qwen3-235B FP8 prefill-only simulator-vs-real config ranking](docs/assets/simulator-fidelity/qwen235-prefill-config-ranking.png) + +| TP | EP | MNS | MBT | Real | Frontier | Frontier − real | +|---:|---:|---:|---:|---:|---:|---:| +| 4 | off | 64 | 8192 | 0.05000 | 0.03750 | -0.01250 | +| 4 | off | 128 | 8192 | 0.05000 | 0.03750 | -0.01250 | +| 4 | off | 64 | 16384 | **0.07500** | **0.06250** | -0.01250 | +| 4 | off | 128 | 16384 | **0.07500** | **0.06250** | -0.01250 | +| 8 | on | 64 | 8192 | 0.05625 | 0.05000 | -0.00625 | +| 8 | on | 128 | 8192 | 0.05625 | 0.05000 | -0.00625 | +| 8 | on | 64 | 16384 | 0.05625 | 0.05625 | 0.00000 | +| 8 | on | 128 | 16384 | 0.05625 | 0.05625 | 0.00000 | + +单位均为 SLO-feasible req/s/GPU。这里 Frontier 对 capacity 有明显、且依赖 topology/action family 的绝对 gap： + +- TP4/MBT8K 与 TP4/MBT16K 都低估 0.0125 req/s/GPU，但正确预测了 MBT8K→16K 的相对增益。 +- TP8/MBT8K 低估 0.00625，而 TP8/MBT16K 恰好匹配；Frontier 因而预测 MBT 在 TP8 上有收益，真机中四个 TP8 config 实际并列。 +- 尽管这个 action differential 是错的，TP4/MBT16K 相对其余 config 有足够 margin，所以 simulator top set 仍与 real top set 完全相同，最终 regret=0。 + +细化边界实验中，Frontier/real 的 33 个 config-load labels 有 6 个 false-infeasible；这些误差在 fresh-server v3 复测中稳定重复。因此这个 case 的结论应严格限定为：**Frontier 是有效的 config ranker / conservative screener，但不是精确的 capacity estimator，也没有完全学对 topology-dependent MBT effect。** + +## Case C 结果：Qwen235 FP8 fixed-shape mixed + +2026-07-16 在 `dash0` GPU 0--3 上完成了 community vLLM 0.10.2、TP4/DP1、MNS64、MBT8192、eager、BF16 KV、prefix/spec/CUDA graph off 的 pilot。所有请求实际 usage 和 streaming token count 都严格为 ISL=2,048、OSL=128： + +| load | requests | TTFT mean/p95 | TPOT mean/p95 | 40 ms joint SLO | +|---|---:|---:|---:|---:| +| single | 1 | 586.2 / 586.2 ms | 136.2 / 136.2 ms | 0/1 | +| concurrency=2 | 2 | 310.2 / 434.3 ms | 124.3 / 124.9 ms | 0/2 | + +Frontier 使用与成功 prefill ranking 相同的 frozen source fingerprint、TP4 KV blocks=26,101 和 block size=16。第一次运行完成 prefill 并提交第一个 decode token后，在纯 decode batch 抛出 `attention decode prediction cache not found for cluster monolithic`。初始化日志给出直接原因：`no standard decode rows`，而 final attention CSV 的 726 rows 全部是 prefill rows。该 crash 没有被记为 SLO fail，也没有使用 dummy prediction。 + +随后在 `dash0` 单张 H20 上用同一 community-vLLM/FlashInfer 栈补测 TP4、batch `{1,2}`、KV `{2048,2176}` 的 4 个 CUDA-event decode points。profile-closed Frontier smoke 精确完成了相同请求： + +| load | Real TTFT | Frontier TTFT | Real TPOT | Frontier TPOT | +|---|---:|---:|---:|---:| +| single | 586.2 ms | 267.7 ms | 136.2 ms | 88.3 ms | +| concurrency=2 | 186.1--434.3 ms | 470.7 ms | 123.8--124.9 ms | 90.5 ms | + +因此 missing-row blocker 已关闭，但 absolute error 明显且不是固定比例；这正是需要 config surface 而不能只看单点 latency 的原因。steady-QPS 下还会出现 prefill+decode true-mixed batches，最小 4-row profile 不足以支持完整 sweep。完整 closure 另外测量了 TP4/TP8 各 81 个 standard-decode points 和 108 个 true-mixed points，覆盖 decode batch 到 128、KV 2,048--2,175；合并 root 共 1,104 rows。Frontier 已确认用 108 个 TP-specific samples 训练 `attn_decode_in_mixed`，没有使用本 case 的 E2E latency calibration。 + +完整 8 configs × 8 rates simulator surface 已在任何真机 surface cell 之前冻结，manifest SHA256 为 `801aa36451c8647f71cc87011144622d2203786e82f189ed1375d964399b106a`。64/64 cells 都有 64 个 finite、非负、shape-exact request records，累计 simulator CPU time 为 11,675.2 s。冻结结果如下，单位为最大被测试可行 req/s/GPU： + +| config family | TPOT 40 ms | TPOT 120 ms | TPOT 150 ms | TPOT 180 ms | +|---|---:|---:|---:|---:| +| TP4，任意 MNS/MBT | none | 0.20 | 0.40 | 0.60 | +| TP8，任意 MNS/MBT | none | 0.05 | 0.15 | 0.20 | + +Frontier 在同一 TP family 内给四个 MNS/MBT config 生成了逐 request 完全相同的结果，即它预测这些 controls 在当前 load range 都不 binding；150 ms sensitivity 下则预测 TP4 family 的 per-GPU capacity 是 TP8 的 2.67×。这两个判断均已用下面的完整真机 surface 检验。 + +第一次真机 surface attempt 还暴露了一个 measurement-contract 问题：一个 fresh server 内连续测多个 rates 会把前一 anchor 的 execution warm state 带入后一 anchor。两个 TP4/MNS64 config 在正向 `0.10→1.60→2.40` 的 1.60 点分别只有 36/64、41/64 requests 通过 150 ms SLO；反向先运行 2.40 后，同一 1.60 点都变成 64/64。逐段看，正向 1.60 的前 24 requests TPOT 从约 234 ms 递减到 169 ms，后半段才稳定到 114--134 ms；反向 1.60 从第一个 request 起就在 113--128 ms。这个 attempt 整体排除，不能用“两个 round 都通过”的保守聚合掩盖 real 与 simulator 的 state contract 不同。 + +修正后的 ground-truth contract 对每个 `(config, rate, round)` 独立启动 server，并在目标 rate 下丢弃 `min(32, max(4, ceil(rate×20)))` 个 exact-shape warmup requests。anchor-isolation smoke 在两个 config、两个 fresh-server repeats 上均为 64/64 completed、0 failure、150 ms SLO 64/64；四次 p95 TPOT 为 127.8--138.0 ms，p95 TTFT 为 295.8--305.5 ms。只有这个通过 gate 的 contract 会进入最终 real capacity/rank。 + +### 完整评测 setup + +| 项目 | 设置 | +|---|---| +| 真机 | `dash0`，8×NVIDIA H20（97,871 MiB/GPU、driver 580.95.05、NV18）；TP4 cells 各独占 4 GPU，TP8 cells 独占全部 8 GPU | +| model/runtime | `Qwen/Qwen3-235B-A22B-FP8`；community vLLM 0.10.2、Python 3.12.3、torch 2.8.0+cu128、FlashInfer 0.3.1.post1；eager、FP8 weights、BF16 KV | +| workload | 64 个不同 token 的固定长度 requests；ISL=2,048、OSL=128、uniform QPS、prefix/speculative decoding/CUDA graph off | +| config surface | `TP∈{4,8} × MNS∈{64,128} × MBT∈{8192,16384}`；TP4 为 EP off，TP8 为 EP on | +| simulator | Frontier commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + frozen best-effort patches；1,104-row attention profile；无本 case E2E calibration | +| simulator anchors | 每个 config 都运行 `{0.10,0.20,0.40,0.80,1.20,1.60,2.40,3.20}` req/s；64/64 cells 在真机 surface 前冻结 | +| real anchors | 围绕冻结 simulator boundary 运行并自适应补齐相邻点；共 34 config-rate cells × 2 independent rounds | +| 每个 real anchor | fresh vLLM server；先按目标 rate 丢弃 `min(32,max(4,ceil(rate×20)))` 个 warmup requests，再从空 scheduler queue 测 64 requests | +| primary SLO | ≥61/64 requests 同时满足 TTFT≤1,256 ms、TPOT≤150 ms；两个 rounds 均通过才判定 feasible | +| primary objective | 最大被测试可行 offered req/s / 实际占用 GPU 数 | + +150 ms 是看到最低负载真实 TPOT 已达 136.2 ms 后公开记录的 post-pilot amendment；原 40 ms 阈值仍保留，但所有 config 在最低 anchor 都不可行。120 ms sensitivity 的 lower boundary 尚未补齐，180 ms sensitivity 在最高 3.20 anchor 仍有 censored cells，因此本文只对 150 ms primary SLO 报告完整 ranking。 + +### 完整结果 + +![Qwen3-235B fixed-shape mixed simulator-vs-real config ranking](docs/assets/simulator-fidelity/qwen235-t0-fixed-shape-ranking.png) + +| TP | MNS | MBT | Real | Frontier | Real boundary：pass → fail（两个 rounds） | +|---:|---:|---:|---:|---:|---| +| 4 | 64 | 8192 | **0.60** | **0.40** | 2.40: 64/64,64/64 → 3.20: 57/64,59/64 | +| 4 | 64 | 16384 | **0.60** | **0.40** | 2.40: 64/64,64/64 → 3.20: 62/64,57/64 | +| 4 | 128 | 8192 | **0.60** | **0.40** | 2.40: 64/64,64/64 → 3.20: 53/64,55/64 | +| 4 | 128 | 16384 | **0.60** | **0.40** | 2.40: 64/64,64/64 → 3.20: 52/64,54/64 | +| 8 | 64 | 8192 | 0.30 | 0.15 | 2.40: 64/64,64/64 → 3.20: 46/64,48/64 | +| 8 | 64 | 16384 | 0.20 | 0.15 | 1.60: 64/64,64/64 → 2.40: 45/64,53/64 | +| 8 | 128 | 8192 | 0.20 | 0.15 | 1.60: 64/64,64/64 → 2.40: 40/64,56/64 | +| 8 | 128 | 16384 | 0.30 | 0.15 | 2.40: 64/64,64/64 → 3.20: 50/64,46/64 | + +Real 与 Frontier 单位均为 SLO-feasible req/s/GPU；boundary 的 offered rate 单位为 system req/s。所有 8 个 real boundaries 都由相邻 lattice points 闭合。四个 TP4 configs 是 real 与 simulator 的共同、完全相同的 top set，因此 optimistic/worst tie-break regret 都是 0。Kendall τ-b=`0.8944`；28 个 config pairs 中 exact `>/< /=` sign 命中 24 个，20 个 real non-tie pairs 中命中 16 个方向。34 个实际测量的 config-load labels 中，Frontier 命中 24 个，有 10 个 false-infeasible、0 个 false-feasible。 + +这个结果同时给出一个比“absolute error”更重要的反例。TP8 真机在 2.40 system req/s 上的 feasibility matrix 是： + +```text + MBT 8K MBT 16K +MNS 64 pass fail +MNS 128 fail pass +``` + +换成 capacity，`ΔMBT | MNS64 = -0.10`，而 `ΔMBT | MNS128 = +0.10 req/s/GPU`，difference-in-differences 为 `+0.20 req/s/GPU`。也就是说，**MBT 的作用方向取决于 MNS；它们不能被看成两个独立、可加的 controls。** Frontier 对四个 TP8 cells 都预测 0.15，interaction 恰好为 0。相同四个 configs 在 1.60 及以下都通过，差异只在持续高负载下出现，说明问题不只是单个 operator 在空载下的固定时间偏差，而发生在 batching/scheduling state 与 execution topology 的组合上。现有数据还不能区分具体是 batch composition、MoE EP communication、memory/KV pressure，还是 runtime 的其他动态状态；下一步应对此边界做 scheduler batch trace、per-step kernel/NCCL timeline 与 KV/queue counters，而不是先给出未经验证的根因。 + +因此本 case 的严格结论是：**Frontier 足以在这个 8-config surface 上选择全局最优 topology，但不足以解释或优化 TP8 family 内的 MNS/MBT controls。** 成功来自 TP4-vs-TP8 的真实 margin 足够大，能够容忍 TP4 33.3%、TP8 25%--50% 的 capacity 低估和内部 action-ordering error；它不是 simulator 已获得 scheduling/execution 深层认知的证据。 + +接受的 ground-truth 共 18 个 fleet jobs、68 个 fresh-server anchors、4,352 个 measured requests 和 1,664 个 warmup requests，消耗 36.26 H20-GPU-hours；从第一个 accepted job 到最后一个完成的墙钟跨度为 5.74 小时。增量 full-attention closure 在单张 H20 上运行 12 秒（0.0033 H20-GPU-hours），但它复用了此前已有的 726-row prefill profile，因此这个数字不是构建全部 compatibility envelope 的总成本。另有共享 GPU/endpoint 的污染 attempts 和暴露 cross-anchor warm-state leakage 的 diagnostic attempts 被整目录隔离，不进入上述统计。Frontier 64-cell surface 消耗 11,675.2 CPU-seconds。完整 hash、逐 anchor input 与排除 ledger 见本节末的复现链接。 + +完整证据见 [T0 smoke report](runs/frontier-multicase-sufficiency-v1/t0-smoke-report.md) 与 [预注册协议](runs/frontier-multicase-sufficiency-v1/protocol.md)。 + +原预注册的 TPOT 40 ms 因最低负载已不可行而失去 capacity-ranking 可辨识性；本文已把 150 ms 明确记录为 post-pilot protocol amendment，并保留 40 ms 全部失败的 sensitivity 结果，没有把 SLO change 隐藏成预注册结论。 + +## Case D 结果：Qwen30 BF16 prefill-only + +![Qwen3-30B BF16 prefill-only simulator-vs-real config ranking](docs/assets/simulator-fidelity/qwen30-prefill-ranking.png) + +结果不是“绝对 capacity 有偏差但 rank 可用”，而是完整的 topology ordering reversal： + +| TP | MNS | Real | Frontier profile-only | 差异 | +|---:|---|---:|---:|---:| +| 1 | 8/16/32/64 | 7.0 | **8.0** | +1.0 | +| 2 | 8/16/32/64 | 7.0 | **8.0** | +1.0 | +| 4 | 8/16/32/64 | **8.0** | 6.0 | -2.0 | + +单位为最大被测试 SLO-feasible req/s/GPU。四个 TP4 configs 是真机 top set；Frontier 却把全部八个 TP1/TP2 configs 判为 top set，两者无交集。部署任一 simulator top config 都只能在真机上得到 7 req/s/GPU，因此 optimistic 与 worst tie-break regret 都为 `1-7/8=12.5%`。Kendall τ-b=`-1.0`；32 个 real non-tie pairs 中 0 个同序。96 个 config-load decisions 中 80 个一致，但同时有 8 个 false-feasible 和 8 个 false-infeasible，它们刚好跨过了不同 TP family 的 capacity boundary。 + +### 错误从哪里开始放大 + +代表性 `MNS=8` anchors 表明，低负载单次 execution time 并没有数倍偏差，错误主要在接近饱和时被 queue 放大： + +| Config / system rate | Real TTFT p95（两轮） | Frontier TTFT p95 | Real / sim SLO | +|---|---:|---:|---| +| TP1 @ 4 req/s | 154.1 / 153.6 ms | 171.6 ms | pass / pass | +| TP1 @ 8 req/s | 1289.2 / 1254.4 ms | 516.4 ms | fail / pass | +| TP2 @ 8 req/s | 97.7 / 93.8 ms | 122.2 ms | pass / pass | +| TP2 @ 16 req/s | 1389.9 / 1365.0 ms | 962.8 ms | fail / pass | +| TP4 @ 4 req/s | 82.3 / 69.6 ms | 93.1 ms | pass / pass | +| TP4 @ 32 req/s | 1136.6 / 1134.0 ms | 1787.9 ms | pass / fail | + +这排除了“只要给所有 kernel 乘一个 global scale 就能修好”的解释：相同 stack 在低负载下接近，但在不同 TP 的饱和边界上向相反方向偏移。更符合数据的抽象是：小的 per-step composition residual 改变 service rate，再通过 scheduler queue 的非线性反馈放大成 TTFT 和 capacity 边界错误。 + +当前证据可以排除一些原因，但还不能唯一定位根因： + +1. **Decode/true-mixed schema 不是必要条件。** 这个 workload 没有 decode、FULL decode CUDA graph、fused true-mixed attention 或 initial-KV state，排序仍然失败。这些机制可能在 mixed/decode 中进一步增大误差，但不是本次失败的前提。 +2. **Communication 可能解释部分 TP scaling，但不能单独解释全部误差。** `profile-v2/allreduce.json` 已有 24 个 TP2/TP4 实测 rows，但 base Frontier 为了保持历史对照使用 analytical 600-Gbps/1-µs collective，没有注入这些数据。下一步必须单独做 measured-collective ablation；但 TP1 无 collective 仍在 8 req/s 出现 773 ms 的 p95 低估，所以 collective 不会是唯一根因。 +3. **Profile 覆盖了 token count，却未必覆盖 scheduler 实际产生的 batch composition。** linear/MoE rows 覆盖到 8,192 tokens，但 exact-2,048 pure-prefill attention 在每个 TP 下只有 batch=1 的直接样本；实际 MBT=8,192 可以产生多个 2,048-token sequence 的组合。这是一个需要补测的 coverage hypothesis，不是已证明根因。 +4. **Routing、step composition 和 scheduler batch state 仍然是联合候选。** pure-prefill 只排除了 phase mixing，没有排除 MoE routing、TP-specific local shape、launch/fusion 和持续负载下 batch/queue trajectory 的耦合。 + +### 为什么 235B 能选对，30B 却选错 + +现有数据不支持“235B 大所以 simulator 更容易”这种模型大小因果。两个 case 的 precision、vLLM 版本、attention backend、graph mode、EP 和 profile closure 都不同。能被数据直接支持的区别是 **decision margin**： + +- 235B mixed 真机中 TP4 最优 capacity 为 0.60 req/s/GPU，最好 TP8 为 0.30，有 `2×` margin；Frontier 虽然漏掉 TP8 的 MNS×MBT interaction，仍预测 TP4/TP8 为 0.40/0.15，所以 residual 没有穿过全局拓扑边界。 +- 30B prefill-only 真机 TP4 相对 TP1/TP2 只有 8 vs. 7 req/s/GPU 的 margin。Frontier 预测为 6 vs. 8，TP-dependent residual 大于真实 margin，因而把最优 topology 完整反转。 + +因此，“prefill-only 容易，decode/mixed 困难”的强假设已被否证。phase 仍可能改变 residual 的幅度，但它不是 fidelity 的充分条件。按预注册 decision rule，此时不应继续无区分度地扩展更多 phase cases，而应在 Qwen30 同一 model/workload 上按单变量顺序做：`measured collective injection` → `batch-composition-conditioned pure-prefill attention/step profile` → 在 TP1@8、TP2@16、TP4@32 对齐 real/sim scheduler batch、queue 与 per-step critical path → 最后再测 routing/graph。 + +本 case 接受的 ground truth 共 24 个 fleet jobs、192 个 fresh-server anchors、12,288 个 measured requests 和 4,512 个 warmup requests，消耗 12.07 H20-GPU-hours。运行中曾因 fleet controller 在 fresh-server 空窗误判 GPU 为空闲而产生重叠 launch；这些 attempts 未被合并到 accepted artifact root，已整体隔离并用独立 queue state 重跑。最终 comparison SHA256 为 `c9a9cac9f60c7be804d1cb9466c455f8fe9e3a8dc60b9cec3329bde6a8c19334`。 + +## 尚不能纳入 simulator ranking 的 case + +### 原 internal-runtime Qwen235 prefill-only + +原实验有 8/8 valid real cells，但使用不同的 serving/runtime contract，real top set 是四个 TP8 config。它没有与本文 best-effort Frontier 使用同一个 community-vLLM FP8 profile/serving contract，不能和新 simulator 数值直接拼接。本文的 community-vLLM Case B 是目前可辩护的 aligned prefill comparison。 + +### 原 Qwen235 decode-only + +真实 surface 有 7/8 valid cells，且 capacity 搜索区间大量重叠：经验 top set 为 5/8，而考虑 bracket uncertainty 时 8/8 都可能最优。更关键的是，该 case 依赖 initial KV state、DecodeBenchConnector、EAGLE3、FP8 KV、DeepEP/NVSHMEM 与 decode CUDA graphs；Frontier 当前没有等价的 initial-KV/EP8 execution contract。因此现在给出 simulator-vs-real rank 或 regret 都会把 contract mismatch 误写成 fidelity 结果。 + +下一步不再用 decode-dominant mixed workload 回避问题，而是在 community vLLM 0.10.2 与 Frontier 两侧增加显式、相同的 initial-KV contract：request 到达时分配与 input length 对应的 KV blocks、标记 prefill complete，再从第一步 decode 开始调度。首先关闭 speculative decode、prefix cache 和 decode CUDA graph，建立 fixed-shape decode-only baseline；随后按单变量顺序启用 trace lengths/prefix、EP8、EAGLE3 和 FULL_DECODE_ONLY graph。每一层都重新报告 ranking，不能把后一层的 mismatch 归因给前一层。 + +## 当前判断与下一步 gate + +当前证据的结论是： + +1. **绝对 gap 不是否决 simulator 的理由。** Qwen235 中 11%--50% 的 capacity 低估仍可保持 zero-regret config selection。 +2. **rank 成功不能证明模型机制正确，rank 失败则表明 residual 穿过了 decision margin。** Qwen235 prefill 的 TP8 MBT differential 错误，fixed-shape mixed 又漏掉 TP8 的 MNS×MBT 非加性交互，但大 topology margin 保住了 top set；Qwen30 prefill-only 的较小 margin 被 TP-dependent saturation residual 穿过，导致 τ-b=-1 和 12.5% regret。 +3. **prefill-only 不是 fidelity 的充分容易条件。** Qwen30 在没有 decode、prefix reuse、initial KV 和 true-mixed batch 时仍选错 topology。phase 可以改变 residual，但不能单独作为 compatibility-envelope 边界。 +4. **alignment 是结果的一部分。** Qwen30 mixed 需要 per-TP E2E calibration 才能恢复排序，同栈 operator profile 或去掉 mixed phase 都不足；Qwen235 不需要本 case E2E calibration，但需要同栈 FP8 profiles、真机 KV capacity 和 simulator patches。论文必须报告这部分真机成本，不能把它隐藏在“offline profiles”中。 +5. **现在有了可复现的全局选优反例，但尚不足以声称 simulator 普遍失败。** Qwen30 BF16 prefill-only 是一个无 top-set overlap 的稳定反例；Qwen235 两个 case 又证明 Frontier 在某些 envelope 内足以选优。更有研究价值的 premise 是：**什么 state-conditioned action residual 决定 ranking 是否能跨过 real margin，以及如何在不做全 surface 真机 sweep 的情况下检测这个风险。** + +后续每个 case 建议使用同一 gate：worst selected-config regret ≤5%、tie-aware rank correlation ≥0.8、足够数量的 informative pairs、ground-truth bracket 不足以反转最优决策，并单独报告达到该结果所需的 real-GPU profiling/calibration cost。当前下一 gate 是在 Qwen30 prefill-only 上做同模型单变量 context ablation，而不是继续扩展跨模型 phase 矩阵。 + +## 数据与复现 + +- 图表输入：[data.json](docs/assets/simulator-fidelity/data.json) +- 画图脚本：[plot_simulator_fidelity.py](scripts/plot_simulator_fidelity.py) +- Qwen30 audit：[report.md](runs/frontier-multicase-sufficiency-v0/results/qwen30-baseline/report.md) +- Qwen30 aligned metrics：[metrics.json](runs/frontier-slo-alignment-v0/results/metrics.json) +- Qwen30 prefill-only experiment card：[experiment-card.md](runs/frontier-phase-factorial-v0/experiment-card.md) +- Qwen30 prefill-only comparison：[comparison.json](runs/frontier-phase-factorial-v0/results/final/comparison.json) +- Qwen30 prefill-only capacity table：[capacity.csv](runs/frontier-phase-factorial-v0/results/final/capacity.csv) +- Qwen30 prefill-only analyzer：[analyze_qwen30_prefill_fidelity.py](runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py) +- Qwen235 fixed-cohort comparison：[v2_refined_comparison.json](runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json) +- Qwen235 full report：[report.md](runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/report.md) +- Fixed-shape mixed comparison：[comparison.json](runs/frontier-multicase-sufficiency-v1/results/t0-final/comparison.json) +- Fixed-shape mixed capacity table：[capacity.csv](runs/frontier-multicase-sufficiency-v1/results/t0-final/capacity.csv) +- Frozen Frontier surface：[frontier_surface_frozen.json](runs/frontier-multicase-sufficiency-v1/artifacts/frontier-t0-surface-v1/frontier_surface_frozen.json) +- Executed real plan：[real-plan-executed.json](runs/frontier-multicase-sufficiency-v1/artifacts/t0-real-surface-v1/real-plan-executed.json) +- Real exclusions/acceptance ledger：[exclusions.json](runs/frontier-multicase-sufficiency-v1/artifacts/t0-real-surface-v1/exclusions.json) +- Fixed-shape / trace-faithful / decode-only 分阶段协议：[protocol.md](runs/frontier-multicase-sufficiency-v1/protocol.md) +- Trace tokenizer/hash contract audit：[trace-contract-audit.json](runs/frontier-multicase-sufficiency-v1/trace-contract-audit.json) +- Trace contract audit reproducer：[audit_trace_contract.py](runs/frontier-multicase-sufficiency-v1/audit_trace_contract.py) +- T0 real smoke 与 Frontier decode-profile failure：[t0-smoke-report.md](runs/frontier-multicase-sufficiency-v1/t0-smoke-report.md) +- 尚未对齐 case 的 real-only audit：[report.md](runs/frontier-multicase-sufficiency-v0/results/ground-truth/report.md) + +重新生成 PNG/SVG： + +```bash +python3 scripts/plot_simulator_fidelity.py + +python3 runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py \ + --fleet-artifacts runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json \ + --output-root runs/frontier-phase-factorial-v0/results/final + +cp runs/frontier-phase-factorial-v0/results/final/qwen30-prefill-ranking.png \ + docs/assets/simulator-fidelity/qwen30-prefill-ranking.png +``` diff --git a/src/aituner/http_client.py b/src/aituner/http_client.py index 186cf0d..0adfa80 100644 --- a/src/aituner/http_client.py +++ b/src/aituner/http_client.py @@ -254,10 +254,11 @@ def stream_chat_completion( base_url: str, body: dict[str, Any], timeout_s: float, + api_path: str = "/v1/chat/completions", ) -> StreamMetrics: data = json.dumps(body).encode("utf-8") request = urllib.request.Request( - url=_openai_url(base_url, "/v1/chat/completions"), + url=_openai_url(base_url, api_path), headers=_auth_headers(None), data=data, method="POST", @@ -285,10 +286,11 @@ def stream_chat_completion( choices = payload.get("choices") if not isinstance(choices, list) or not choices: continue - delta = choices[0].get("delta", {}) - if not isinstance(delta, dict): - continue - content = delta.get("content") + choice = choices[0] + delta = choice.get("delta", {}) + content = delta.get("content") if isinstance(delta, dict) else None + if not isinstance(content, str): + content = choice.get("text") if isinstance(content, str) and content: now = time.monotonic() if first_token_at is None: diff --git a/src/aituner/slo.py b/src/aituner/slo.py index 1a86a8d..fcbb5c5 100644 --- a/src/aituner/slo.py +++ b/src/aituner/slo.py @@ -16,6 +16,7 @@ class RequestOutcome: completion_tokens: int | None error: str = "" completion_tokens_source: str = "" + completed_mono_ns: int | None = None @dataclass(frozen=True) diff --git a/src/aituner/spec.py b/src/aituner/spec.py index a9c5720..a881e8e 100644 --- a/src/aituner/spec.py +++ b/src/aituner/spec.py @@ -411,8 +411,10 @@ class TraceSpec: synthetic_prompt_cap = data.get("synthetic_prompt_cap_tokens") completion_tokens_override = data.get("completion_tokens_override") request_mode = str(data.get("request_mode") or "chat").strip().lower() - if request_mode not in {"chat", "decode_only"}: - raise SpecError("trace.request_mode must be one of: chat, decode_only.") + if request_mode not in {"chat", "decode_only", "raw_completion"}: + raise SpecError( + "trace.request_mode must be one of: chat, decode_only, raw_completion." + ) if completion_tokens_override is not None: completion_tokens_override = _require_int( completion_tokens_override, diff --git a/src/aituner/trace.py b/src/aituner/trace.py index cabb3a5..9e39323 100644 --- a/src/aituner/trace.py +++ b/src/aituner/trace.py @@ -40,6 +40,7 @@ class TraceRequest: body: dict[str, Any] prompt_tokens_hint: int | None completion_tokens_hint: int | None + api_path: str = "/v1/chat/completions" metadata: dict[str, Any] = field(default_factory=dict) @@ -186,26 +187,41 @@ def load_trace_requests(study: StudySpec, *, study_spec_path: Path) -> tuple[Win prompt_tokens_hint = _coerce_prompt_tokens(row) if not _matches_input_length_filter(study, prompt_tokens_hint=prompt_tokens_hint): continue - try: - messages = _coerce_messages(row) - except TraceError: - capped_prompt_tokens = prompt_tokens_hint or 0 - if study.trace.synthetic_prompt_cap_tokens is not None: - capped_prompt_tokens = min( - capped_prompt_tokens, study.trace.synthetic_prompt_cap_tokens + api_path = "/v1/chat/completions" + if study.trace.request_mode == "raw_completion": + prompt = row.get("prompt") + if not isinstance(prompt, str) or not prompt: + raise TraceError( + f"trace row {idx} is missing prompt required by raw_completion" ) - messages = [ - { - "role": "user", - "content": _synthetic_prompt_from_tokens(capped_prompt_tokens), - } - ] - body: dict[str, Any] = { - "model": study.model.served_model_name, - "messages": messages, - "stream": True, - "stream_options": {"include_usage": True}, - } + body: dict[str, Any] = { + "model": study.model.served_model_name, + "prompt": prompt, + "stream": True, + "stream_options": {"include_usage": True}, + } + api_path = "/v1/completions" + else: + try: + messages = _coerce_messages(row) + except TraceError: + capped_prompt_tokens = prompt_tokens_hint or 0 + if study.trace.synthetic_prompt_cap_tokens is not None: + capped_prompt_tokens = min( + capped_prompt_tokens, study.trace.synthetic_prompt_cap_tokens + ) + messages = [ + { + "role": "user", + "content": _synthetic_prompt_from_tokens(capped_prompt_tokens), + } + ] + body = { + "model": study.model.served_model_name, + "messages": messages, + "stream": True, + "stream_options": {"include_usage": True}, + } completion_tokens = ( study.trace.completion_tokens_override if study.trace.completion_tokens_override is not None @@ -225,6 +241,7 @@ def load_trace_requests(study: StudySpec, *, study_spec_path: Path) -> tuple[Win body=body, prompt_tokens_hint=prompt_tokens_hint, completion_tokens_hint=completion_tokens, + api_path=api_path, metadata={ "hash_ids": row.get("hash_ids") if isinstance(row.get("hash_ids"), list) else None, "turn": row.get("turn"), diff --git a/src/aituner/worker.py b/src/aituner/worker.py index d8db0b1..d8f0ca6 100644 --- a/src/aituner/worker.py +++ b/src/aituner/worker.py @@ -107,7 +107,12 @@ def _run_one_request( timeout_s: float, ) -> RequestOutcome: try: - metrics = stream_chat_completion(base_url=base_url, body=request.body, timeout_s=timeout_s) + metrics = stream_chat_completion( + base_url=base_url, + body=request.body, + timeout_s=timeout_s, + api_path=request.api_path, + ) expected_completion_tokens = request.completion_tokens_hint actual_completion_tokens = metrics.completion_tokens completion_tokens_source = getattr(metrics, "completion_tokens_source", "") @@ -127,6 +132,7 @@ def _run_one_request( f"actual={actual_completion_tokens}" ), completion_tokens_source=completion_tokens_source, + completed_mono_ns=time.monotonic_ns(), ) if actual_completion_tokens != expected_completion_tokens: return RequestOutcome( @@ -142,6 +148,7 @@ def _run_one_request( f"actual={actual_completion_tokens}" ), completion_tokens_source=completion_tokens_source, + completed_mono_ns=time.monotonic_ns(), ) return RequestOutcome( request_id=request.row_id, @@ -151,6 +158,7 @@ def _run_one_request( prompt_tokens=request.prompt_tokens_hint, completion_tokens=actual_completion_tokens or request.completion_tokens_hint, completion_tokens_source=completion_tokens_source, + completed_mono_ns=time.monotonic_ns(), ) except HttpClientError as exc: return RequestOutcome( @@ -161,6 +169,7 @@ def _run_one_request( prompt_tokens=request.prompt_tokens_hint, completion_tokens=request.completion_tokens_hint, error=str(exc), + completed_mono_ns=time.monotonic_ns(), ) diff --git a/tests/test_core_flow.py b/tests/test_core_flow.py index e665c81..1e9d0a6 100644 --- a/tests/test_core_flow.py +++ b/tests/test_core_flow.py @@ -5586,6 +5586,42 @@ class CoreFlowTests(unittest.TestCase): self.assertEqual(requests[2].body["min_tokens"], 1) self.assertEqual(requests[2].body["max_tokens"], 1) + def test_raw_completion_mode_preserves_trace_prompt_and_endpoint(self) -> None: + with tempfile.TemporaryDirectory() as tmp: + tmp_path = Path(tmp) + study_path = _write_study_assets( + tmp_path, + trace_overrides={"request_mode": "raw_completion"}, + ) + trace_path = tmp_path / "trace_windows" / "traces" / "chat_w1.jsonl" + rows = [json.loads(line) for line in trace_path.read_text().splitlines()] + for idx, row in enumerate(rows): + row["prompt"] = f"<|im_start|>user\nraw prompt {idx}<|im_end|>" + trace_path.write_text( + "".join(json.dumps(row) + "\n" for row in rows), + encoding="utf-8", + ) + + study = load_study_spec(study_path) + _, requests = load_trace_requests(study, study_spec_path=study_path) + + self.assertEqual(study.trace.request_mode, "raw_completion") + self.assertEqual(requests[0].api_path, "/v1/completions") + self.assertEqual(requests[0].body["prompt"], rows[0]["prompt"]) + self.assertNotIn("messages", requests[0].body) + + def test_raw_completion_mode_requires_prompt(self) -> None: + with tempfile.TemporaryDirectory() as tmp: + study_path = _write_study_assets( + Path(tmp), + trace_overrides={"request_mode": "raw_completion"}, + ) + study = load_study_spec(study_path) + with self.assertRaisesRegex( + ValueError, "missing prompt required by raw_completion" + ): + load_trace_requests(study, study_spec_path=study_path) + def test_run_one_request_fails_fixed_length_completion_mismatch(self) -> None: request = TraceRequest( row_id="r1", @@ -5604,15 +5640,17 @@ class CoreFlowTests(unittest.TestCase): completion_tokens=1, ), ): - outcome = _run_one_request( - request, - base_url="http://127.0.0.1:8000", - timeout_s=1.0, - ) + with mock.patch("aituner.worker.time.monotonic_ns", return_value=123456789): + outcome = _run_one_request( + request, + base_url="http://127.0.0.1:8000", + timeout_s=1.0, + ) self.assertFalse(outcome.success) self.assertEqual(outcome.error, "completion_tokens_mismatch expected=2 actual=1") self.assertEqual(outcome.completion_tokens, 1) + self.assertEqual(outcome.completed_mono_ns, 123456789) def test_build_prompt_mentions_completion_tokens_override(self) -> None: with tempfile.TemporaryDirectory() as tmp: @@ -8813,6 +8851,35 @@ class CoreFlowTests(unittest.TestCase): self.assertIsNone(metrics.completion_tokens) self.assertEqual(metrics.completion_tokens_source, "none") + def test_stream_chat_completion_reads_raw_completion_text(self) -> None: + class FakeResponse: + def __enter__(self): + return self + + def __exit__(self, exc_type, exc, traceback): + return False + + def __iter__(self): + return iter( + [ + b'data: {"choices": [{"text": "x"}]}\n', + b'data: {"choices": [], "usage": {"completion_tokens": 1}}\n', + b"data: [DONE]\n", + ] + ) + + with mock.patch("aituner.http_client._urlopen", return_value=FakeResponse()): + metrics = stream_chat_completion( + base_url="http://127.0.0.1:8000", + body={"model": "m", "prompt": "raw"}, + timeout_s=1.0, + api_path="/v1/completions", + ) + + self.assertIsNotNone(metrics.ttft_ms) + self.assertEqual(metrics.completion_tokens, 1) + self.assertEqual(metrics.streamed_chunk_count, 1) + def test_stream_chat_completion_marks_same_instant_multitoken_tpot_unmeasurable(self) -> None: class FakeResponse: def __enter__(self):