gahow/aituner
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
5359463652498cf5353edfe31679b73e466d40df
aituner/runs/opprof-phase5/phase5/metrics.json
Gahow Wang d5b276180d Add OpProf campaign: protocols, results, patches, run evidence (P0-P6) 
Workload-conditioned operator profiling on patched vLLM 0.24.0 +
Qwen3-30B-A3B/H20. H1b PASS (irregular patterns carry +23-45pp R64
raggedness, 8-45% token-efficiency loss vs rectangular controls);
mechanism decomposition kills the padding narrative and finds the
arrival-uniformization artifact (-12.9%); cross-version churn surface
shows TP2/MNS64 -29.4% across vLLM 0.20->0.24 while the argmax held.
Raw Layer-1 JSONL streams (507 MB) stay on disk, git-ignored; footer
sidecars and metrics are tracked.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-13 11:06:10 +08:00

14179 lines
554 KiB
JSON
Raw Blame History

{
"amendment_A_P5_1": {
"reason": "Rate-following throughput drift tracks arrival shape and is not a cold-start stationarity test.",
"recorded_drift_range": [
1.90551033490161,
2.1650001817983497
],
"retained_failed_run_evidence": [
{
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
144
],
[
"PIECEWISE",
216
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2510807) INFO 07-12 12:13:38 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2510807) INFO 07-12 12:13:38 [monitor.py:53] torch.compile took 5.74 s in total",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:00<00:04, 9.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 9.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 9.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:02, 9.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 8.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 8.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 8.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:03, 11.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 11.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:02, 11.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:01, 11.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:02<00:01, 11.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 11.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:03<00:00, 11.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 199,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
336
],
[
"PIECEWISE",
384
]
],
"warmup_long_completions": 12,
"warmup_long_min_tokens": 9015
},
"run": "A1-r2-C00",
"superseded_drift_evidence": {
"bin_seconds": 5.0,
"mean_scheduled_token_throughput": 5512.066666666667,
"normalized_drift": 1.90551033490161,
"normalized_drift_limit": 0.1,
"passed": false,
"reason": "A-P3-6 stabilization criterion not met",
"scheduled_token_throughput": [
7154.0,
9230.4,
151.8
],
"scheduled_tokens": [
35770,
46152,
759
],
"slope_tokens_per_second_squared": -700.22,
"step_counts": [
251,
166,
255
],
"step_indices_continuous": true,
"window_seconds": [
45.0,
60.0
]
}
},
{
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
96
],
[
"PIECEWISE",
128
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2510806) INFO 07-12 12:13:37 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783858417.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2510806) INFO 07-12 12:13:37 [monitor.py:53] torch.compile took 5.65 s in total",
"phase": "pre-ready",
"timestamp_s": 1783858417.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 10.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 9.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 10.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:02, 9.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:02, 9.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:01, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 7.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 11.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 12.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:02, 11.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:01, 11.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:02<00:01, 11.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 11.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:03<00:00, 11.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 12.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 195,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 28,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 16,
"warmup_long_min_tokens": 9015
},
"run": "A3-r1-C00",
"superseded_drift_evidence": {
"bin_seconds": 5.0,
"mean_scheduled_token_throughput": 4346.733333333334,
"normalized_drift": 0.13230625297158005,
"normalized_drift_limit": 0.1,
"passed": false,
"reason": "A-P3-6 stabilization criterion not met",
"scheduled_token_throughput": [
3667.8,
5321.2,
4051.2
],
"scheduled_tokens": [
18339,
26606,
20256
],
"slope_tokens_per_second_squared": 38.33999999999996,
"step_counts": [
350,
472,
465
],
"step_indices_continuous": true,
"window_seconds": [
45.0,
60.0
]
}
},
{
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
168
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2510814) INFO 07-12 12:13:38 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2510814) INFO 07-12 12:13:38 [monitor.py:53] torch.compile took 5.61 s in total",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:00<00:04, 9.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:02, 9.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:02, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:01, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:04<00:00, 8.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 8.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 8.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 11.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:01, 11.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:02<00:01, 11.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 11.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:03<00:00, 11.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 195,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"run": "A4-r1-A4",
"superseded_drift_evidence": {
"bin_seconds": 5.0,
"mean_scheduled_token_throughput": 3667.066666666666,
"normalized_drift": 2.164918372541177,
"normalized_drift_limit": 0.1,
"passed": false,
"reason": "A-P3-6 stabilization criterion not met",
"scheduled_token_throughput": [
5385.4,
5523.0,
92.8
],
"scheduled_tokens": [
26927,
27615,
464
],
"slope_tokens_per_second_squared": -529.2599999999999,
"step_counts": [
379,
202,
188
],
"step_indices_continuous": true,
"window_seconds": [
45.0,
60.0
]
}
},
{
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
168
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2510805) INFO 07-12 12:13:38 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2510805) INFO 07-12 12:13:38 [monitor.py:53] torch.compile took 5.60 s in total",
"phase": "pre-ready",
"timestamp_s": 1783858418.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:00<00:04, 10.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 9.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 9.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:02, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:01, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 9.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 9.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:04<00:00, 8.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 8.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 8.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 8.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:03, 12.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 12.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 12.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 12.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:00<00:03, 12.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 12.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:02, 12.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 12.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 12.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 12.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:01<00:02, 11.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:01, 11.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:02<00:01, 11.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 11.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:03<00:00, 11.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 12.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 196,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"run": "base-r2-C00",
"superseded_drift_evidence": {
"bin_seconds": 5.0,
"mean_scheduled_token_throughput": 3667.066666666666,
"normalized_drift": 2.1650001817983497,
"normalized_drift_limit": 0.1,
"passed": false,
"reason": "A-P3-6 stabilization criterion not met",
"scheduled_token_throughput": [
5385.4,
5523.2,
92.6
],
"scheduled_tokens": [
26927,
27616,
463
],
"slope_tokens_per_second_squared": -529.28,
"step_counts": [
380,
201,
187
],
"step_indices_continuous": true,
"window_seconds": [
45.0,
60.0
]
}
}
],
"uniform_A3_drift": 0.13230625297158005
},
"analysis_seed": 20260716,
"bootstrap_resamples": 100000,
"bridge": {
"ci95": [
-0.32979921703120474,
0.35485858777390844
],
"ci_contains_zero": true,
"point_delta": 0.007040972466240358,
"relative_abs_delta": 0.002688303981677958,
"reuse_passed": true,
"within_3pct": true
},
"config_tier_A2": {
"A2_padding_fraction": 0.02398892811002051,
"base_padding_fraction": 0.1374049032789276,
"delta_E": 0.048375663195566077,
"delta_E_ci95": [
-0.31536063261129355,
0.4114458465370726
],
"padding_reduction": 0.8254143226510356,
"relative_E_delta": 0.0160516410603897
},
"control_sources": {
"P03": "P3-reused",
"P04": "P3-reused"
},
"dominance": {
"A1": {
"per_control": {
"P03": false,
"P04": "NOT_EVALUABLE"
},
"verdict": "NOT EVALUABLE"
},
"A2": {
"per_control": {
"P03": "NOT_EVALUABLE",
"P04": "NOT_EVALUABLE"
},
"verdict": "NOT EVALUABLE"
},
"A3": {
"per_control": {
"P03": "NOT_EVALUABLE",
"P04": "NOT_EVALUABLE"
},
"verdict": "NOT EVALUABLE"
},
"A4": {
"per_control": {
"P03": "NOT_EVALUABLE",
"P04": "NOT_EVALUABLE"
},
"verdict": "NOT EVALUABLE"
}
},
"efficiency": {
"A1": {
"ci95": [
2.8016586581562,
3.3493073318732156
],
"point": 3.0789989299185936,
"runs": [
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11599,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
144
],
[
"PIECEWISE",
216
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2551420) INFO 07-12 12:58:43 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783861123.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2551420) INFO 07-12 12:58:43 [monitor.py:53] torch.compile took 5.72 s in total",
"phase": "pre-ready",
"timestamp_s": 1783861123.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 10.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 10.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 9.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:02, 9.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 2%|\u258f | 1/51 [00:00<00:05, 8.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:05, 8.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 10.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 10.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 10.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 10.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:02, 10.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 10.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 10.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:01, 10.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 11.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 203,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
336
],
[
"PIECEWISE",
384
]
],
"warmup_long_completions": 12,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 1.0420550710898726,
"drain_quarantined": false,
"drain_seconds": 5.944268725987058,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.014135554431996595,
"prefix_hits": 15936,
"prefix_present_steps": 11599,
"prefix_queries": 1127370,
"pure_decode_bucket_tokens": 27662,
"pure_decode_padding_fraction": 0.07700093991757646,
"pure_decode_padding_tokens": 2130,
"pure_decode_steps": 11344,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16
],
"pure_decode_support_coverage": 0.955394922425952,
"run_dir": "runs/phase5/primary/A1-r1-C00/moderate",
"run_id": "A1-r1-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "0d66448dc676e8ecee475fd57cf9fa2fa4f07b7959d28ddb0564c241454b5dad",
"token_efficiency_per_ms": 3.083111225991122,
"waiting_block_cv": 2.471768598095917,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11561,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
144
],
[
"PIECEWISE",
216
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2529912) INFO 07-12 12:36:40 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2529912) INFO 07-12 12:36:40 [monitor.py:53] torch.compile took 6.03 s in total",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:05, 8.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:05, 8.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:05, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 7.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 6.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 6.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 11.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 10.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 10.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 11.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 203,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
336
],
[
"PIECEWISE",
384
]
],
"warmup_long_completions": 12,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 1.0454725886338505,
"drain_quarantined": false,
"drain_seconds": 6.003959469002439,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.014135554431996595,
"prefix_hits": 15936,
"prefix_present_steps": 11561,
"prefix_queries": 1127370,
"pure_decode_bucket_tokens": 27649,
"pure_decode_padding_fraction": 0.07613295236717423,
"pure_decode_padding_tokens": 2105,
"pure_decode_steps": 11307,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16
],
"pure_decode_support_coverage": 0.9556911647651897,
"run_dir": "runs/phase5/primary/A1-r2-C00/moderate",
"run_id": "A1-r2-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "319839e156b8d6c131daa2e653b36288c0342eefe4501a6f5645ba833f40c763",
"token_efficiency_per_ms": 3.0706567733209598,
"waiting_block_cv": 2.4485483622648805,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11605,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
144
],
[
"PIECEWISE",
216
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2537991) INFO 07-12 12:45:13 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2537991) INFO 07-12 12:45:13 [monitor.py:53] torch.compile took 5.69 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:00<00:04, 9.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 10.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:02, 10.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:02, 9.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:02, 9.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:01, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 9.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:03<00:01, 9.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:04<00:00, 7.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 12.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:03, 12.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 12.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 12.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 12.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:00<00:03, 12.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 12.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:02, 12.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 12.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 12.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 12.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:01<00:02, 12.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 12.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:01, 12.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 12.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 12.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:02<00:01, 12.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:02<00:01, 12.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 12.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 12.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 12.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 12.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:03<00:00, 12.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:03<00:00, 12.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 12.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 12.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 193,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"PIECEWISE",
80
],
[
"PIECEWISE",
336
],
[
"PIECEWISE",
384
]
],
"warmup_long_completions": 12,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 1.0381467494614438,
"drain_quarantined": false,
"drain_seconds": 5.94826945499517,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.014135554431996595,
"prefix_hits": 15936,
"prefix_present_steps": 11605,
"prefix_queries": 1127370,
"pure_decode_bucket_tokens": 27660,
"pure_decode_padding_fraction": 0.07693420101229212,
"pure_decode_padding_tokens": 2128,
"pure_decode_steps": 11350,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13,
14,
15,
16
],
"pure_decode_support_coverage": 0.9555066079295155,
"run_dir": "runs/phase5/primary/A1-r3-C00/moderate",
"run_id": "A1-r3-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "eddb675dd5114deb6aea66c5c2dcfdacdc97ee4926d0b87a89226e29f84e2ed3",
"token_efficiency_per_ms": 3.083262835538282,
"waiting_block_cv": 2.506236198086574,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
}
]
},
"A2": {
"ci95": [
2.8028115114668224,
3.31427584217418
],
"point": 3.0621275290369585,
"runs": [
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11501,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
160
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2537990) INFO 07-12 12:45:13 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/589f04d9d986d8f9ad09371c",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2537990) INFO 07-12 12:45:13 [monitor.py:53] torch.compile took 5.76 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258e | 2/55 [00:00<00:04, 11.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 7%|\u258b | 4/55 [00:00<00:04, 11.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 11%|\u2588 | 6/55 [00:00<00:04, 10.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 15%|\u2588\u258d | 8/55 [00:00<00:04, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 10/55 [00:00<00:04, 9.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:04, 10.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:04, 10.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:03, 10.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:03, 10.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:01<00:03, 10.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:02<00:03, 10.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:03, 10.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:03, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:02<00:02, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 29/55 [00:02<00:02, 9.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:02<00:02, 9.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 56%|\u2588\u2588\u2588\u2588\u2588\u258b | 31/55 [00:03<00:02, 9.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:03<00:02, 9.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 60%|\u2588\u2588\u2588\u2588\u2588\u2588 | 33/55 [00:03<00:02, 9.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 64%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 35/55 [00:03<00:02, 9.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:01, 9.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 37/55 [00:03<00:01, 9.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:03<00:01, 9.70it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 39/55 [00:03<00:01, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:04<00:01, 9.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 41/55 [00:04<00:01, 9.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:04<00:01, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 43/55 [00:04<00:01, 9.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:04<00:01, 7.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 45/55 [00:04<00:01, 7.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:04<00:01, 7.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 85%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 47/55 [00:04<00:01, 7.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:05<00:00, 7.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 89%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 49/55 [00:06<00:02, 2.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:06<00:01, 2.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 93%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e| 51/55 [00:07<00:02, 1.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:07<00:01, 2.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b| 53/55 [00:08<00:01, 1.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:08<00:00, 2.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:08<00:00, 2.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:08<00:00, 6.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258e | 2/55 [00:00<00:04, 11.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 7%|\u258b | 4/55 [00:00<00:04, 11.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 11%|\u2588 | 6/55 [00:00<00:04, 11.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 15%|\u2588\u258d | 8/55 [00:00<00:04, 11.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 18%|\u2588\u258a | 10/55 [00:00<00:03, 11.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:03, 11.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:03, 11.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:03, 11.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:03, 11.70it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:01<00:03, 11.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:01<00:02, 11.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:02, 11.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:02, 11.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:02<00:02, 11.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:02<00:02, 11.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:02<00:01, 11.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 62%|\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 34/55 [00:02<00:01, 11.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:01, 11.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:03<00:01, 11.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:03<00:01, 11.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:03<00:01, 11.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:03<00:00, 11.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:03<00:00, 11.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:04<00:00, 11.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:04<00:00, 11.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:04<00:00, 11.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:04<00:00, 11.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:04<00:00, 11.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 197,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
1,
2,
3,
4,
5,
6,
7,
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
32
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8867549548299586,
"drain_quarantined": false,
"drain_seconds": 0.6978098230028991,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11501,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 26808,
"pure_decode_padding_fraction": 0.024321098179647865,
"pure_decode_padding_tokens": 652,
"pure_decode_steps": 11250,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9886222222222222,
"run_dir": "runs/phase5/primary/A2-r1-A2/moderate",
"run_id": "A2-r1-A2-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "0321bffb2fbfa4daac3a9f744cf0b9de121644a92de9b8382a03515ace953542",
"token_efficiency_per_ms": 3.059593049598507,
"waiting_block_cv": 6.202752383943476,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11528,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
160
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2551427) INFO 07-12 12:58:42 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/589f04d9d986d8f9ad09371c",
"phase": "pre-ready",
"timestamp_s": 1783861122.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2551427) INFO 07-12 12:58:42 [monitor.py:53] torch.compile took 5.62 s in total",
"phase": "pre-ready",
"timestamp_s": 1783861122.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258e | 2/55 [00:00<00:05, 10.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 7%|\u258b | 4/55 [00:00<00:04, 10.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 11%|\u2588 | 6/55 [00:00<00:04, 10.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 15%|\u2588\u258d | 8/55 [00:00<00:04, 9.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258b | 9/55 [00:00<00:05, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2588 | 11/55 [00:01<00:04, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:04, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 13/55 [00:01<00:04, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:04, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 15/55 [00:01<00:04, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:04, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588 | 17/55 [00:01<00:04, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:03, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258d | 19/55 [00:01<00:03, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:02<00:03, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 38%|\u2588\u2588\u2588\u258a | 21/55 [00:02<00:03, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:02<00:03, 9.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 42%|\u2588\u2588\u2588\u2588\u258f | 23/55 [00:02<00:03, 9.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:03, 9.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 25/55 [00:02<00:03, 8.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:03, 8.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 27/55 [00:02<00:03, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:02<00:02, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 29/55 [00:03<00:02, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:03<00:02, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 56%|\u2588\u2588\u2588\u2588\u2588\u258b | 31/55 [00:03<00:02, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:03<00:02, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 60%|\u2588\u2588\u2588\u2588\u2588\u2588 | 33/55 [00:03<00:02, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 62%|\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 34/55 [00:03<00:02, 9.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 64%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 35/55 [00:03<00:02, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:02, 9.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 37/55 [00:03<00:01, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:04<00:01, 9.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 39/55 [00:04<00:01, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:04<00:01, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 41/55 [00:04<00:01, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:04<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 43/55 [00:04<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:04<00:01, 8.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 45/55 [00:04<00:01, 8.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:05<00:01, 7.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 85%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 47/55 [00:05<00:01, 7.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:05<00:00, 7.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 89%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 49/55 [00:05<00:00, 6.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:05<00:00, 7.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 93%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e| 51/55 [00:05<00:00, 7.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:05<00:00, 7.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b| 53/55 [00:05<00:00, 7.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:06<00:00, 7.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:06<00:00, 7.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:06<00:00, 8.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258e | 2/55 [00:00<00:04, 10.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 7%|\u258b | 4/55 [00:00<00:04, 10.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 11%|\u2588 | 6/55 [00:00<00:04, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 15%|\u2588\u258d | 8/55 [00:00<00:04, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 18%|\u2588\u258a | 10/55 [00:00<00:04, 11.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:03, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:03, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:03, 10.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:03, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:01<00:03, 10.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:02<00:03, 10.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:02, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:02, 11.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:02<00:02, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:02<00:02, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:02<00:02, 11.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 62%|\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 34/55 [00:03<00:01, 10.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:01, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:03<00:01, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:03<00:01, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:03<00:01, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:04<00:01, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:04<00:00, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:04<00:00, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:04<00:00, 10.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 160,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:04<00:00, 11.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 161,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:04<00:00, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 162,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:04<00:00, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 207,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
1,
2,
3,
4,
5,
6,
7,
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
32
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8857349310142049,
"drain_quarantined": false,
"drain_seconds": 0.6761808950104751,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11528,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 26799,
"pure_decode_padding_fraction": 0.02376954363968805,
"pure_decode_padding_tokens": 637,
"pure_decode_steps": 11277,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9889154917087878,
"run_dir": "runs/phase5/primary/A2-r2-A2/moderate",
"run_id": "A2-r2-A2-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "bcdd34b8bcdb4e1d9b9a8d40df608b5bee0c9b5d87d83d20bb4403427ed92d78",
"token_efficiency_per_ms": 3.064480599960168,
"waiting_block_cv": 6.191388338087139,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11531,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
160
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2544922) INFO 07-12 12:52:10 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/589f04d9d986d8f9ad09371c",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2544922) INFO 07-12 12:52:10 [monitor.py:53] torch.compile took 5.79 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/55 [00:00<00:05, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 5%|\u258c | 3/55 [00:00<00:05, 9.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 7%|\u258b | 4/55 [00:00<00:05, 9.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 9%|\u2589 | 5/55 [00:00<00:05, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 11%|\u2588 | 6/55 [00:00<00:05, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 15%|\u2588\u258d | 8/55 [00:00<00:04, 9.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258b | 9/55 [00:00<00:05, 8.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2588 | 11/55 [00:01<00:04, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:04, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 13/55 [00:01<00:04, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:04, 9.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 15/55 [00:01<00:04, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:04, 9.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588 | 17/55 [00:01<00:04, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:04, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258d | 19/55 [00:02<00:03, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:02<00:03, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 38%|\u2588\u2588\u2588\u258a | 21/55 [00:02<00:03, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:02<00:03, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 42%|\u2588\u2588\u2588\u2588\u258f | 23/55 [00:02<00:03, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:03, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 25/55 [00:02<00:03, 8.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:03, 8.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 27/55 [00:02<00:03, 8.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:03<00:03, 8.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 29/55 [00:03<00:02, 8.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:03<00:02, 8.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 56%|\u2588\u2588\u2588\u2588\u2588\u258b | 31/55 [00:03<00:02, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:03<00:02, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 60%|\u2588\u2588\u2588\u2588\u2588\u2588 | 33/55 [00:03<00:02, 8.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 62%|\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 34/55 [00:03<00:02, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 64%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 35/55 [00:03<00:02, 9.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:02, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 37/55 [00:04<00:01, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:04<00:01, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 39/55 [00:04<00:01, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:04<00:01, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 41/55 [00:04<00:01, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:04<00:01, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 43/55 [00:04<00:01, 9.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:04<00:01, 8.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 45/55 [00:04<00:01, 8.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:05<00:01, 7.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 85%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 47/55 [00:05<00:01, 7.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:05<00:00, 7.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 89%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 49/55 [00:05<00:00, 6.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:05<00:00, 7.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 93%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e| 51/55 [00:05<00:00, 7.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:06<00:00, 7.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b| 53/55 [00:06<00:00, 6.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:06<00:00, 7.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:06<00:00, 7.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:06<00:00, 8.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/55 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258e | 2/55 [00:00<00:04, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 7%|\u258b | 4/55 [00:00<00:04, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 11%|\u2588 | 6/55 [00:00<00:04, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 15%|\u2588\u258d | 8/55 [00:00<00:04, 11.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 18%|\u2588\u258a | 10/55 [00:00<00:04, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 22%|\u2588\u2588\u258f | 12/55 [00:01<00:03, 11.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 25%|\u2588\u2588\u258c | 14/55 [00:01<00:03, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 29%|\u2588\u2588\u2589 | 16/55 [00:01<00:03, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 33%|\u2588\u2588\u2588\u258e | 18/55 [00:01<00:03, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 36%|\u2588\u2588\u2588\u258b | 20/55 [00:01<00:03, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 40%|\u2588\u2588\u2588\u2588 | 22/55 [00:01<00:02, 11.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 44%|\u2588\u2588\u2588\u2588\u258e | 24/55 [00:02<00:02, 11.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 26/55 [00:02<00:02, 11.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 28/55 [00:02<00:02, 11.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 30/55 [00:02<00:02, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 58%|\u2588\u2588\u2588\u2588\u2588\u258a | 32/55 [00:02<00:02, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 62%|\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 34/55 [00:03<00:01, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258c | 36/55 [00:03<00:01, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u2589 | 38/55 [00:03<00:01, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 40/55 [00:03<00:01, 10.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 42/55 [00:03<00:01, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 44/55 [00:03<00:01, 10.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 46/55 [00:04<00:00, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 160,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 87%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 48/55 [00:04<00:00, 11.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 161,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 91%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 50/55 [00:04<00:00, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 162,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 95%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 52/55 [00:04<00:00, 11.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 163,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 54/55 [00:04<00:00, 11.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 164,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 55/55 [00:04<00:00, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 209,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
1,
2,
3,
4,
5,
6,
7,
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
3
],
[
"FULL",
4
],
[
"FULL",
5
],
[
"FULL",
6
],
[
"FULL",
7
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
32
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8872251794933065,
"drain_quarantined": false,
"drain_seconds": 0.6785303539945744,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11531,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 26805,
"pure_decode_padding_fraction": 0.02387614251072561,
"pure_decode_padding_tokens": 640,
"pure_decode_steps": 11280,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9888297872340426,
"run_dir": "runs/phase5/primary/A2-r3-A2/moderate",
"run_id": "A2-r3-A2-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "1338d1dbd5c37888bc971c711e970a28163c72df7d51a912ab2d66a20adae1e1",
"token_efficiency_per_ms": 3.062312844779378,
"waiting_block_cv": 6.1913044154881565,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
}
]
},
"A3": {
"ci95": [
2.4518517510711058,
2.7926839826962455
],
"point": 2.626154180729837,
"runs": [
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 18304,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
96
],
[
"PIECEWISE",
128
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2529900) INFO 07-12 12:36:40 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2529900) INFO 07-12 12:36:40 [monitor.py:53] torch.compile took 5.92 s in total",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 9.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 9.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 8.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 10.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:04, 10.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 10.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:02, 11.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 11.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 204,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 28,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 16,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 0.6604105665135865,
"drain_quarantined": false,
"drain_seconds": 1.1182762120151892,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4708333333333333,
"prefix_hit_ratio": 0.017412117941776724,
"prefix_hits": 17328,
"prefix_present_steps": 18304,
"prefix_queries": 995169,
"pure_decode_bucket_tokens": 27196,
"pure_decode_padding_fraction": 0.06162670981026622,
"pure_decode_padding_tokens": 1676,
"pure_decode_steps": 17998,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7
],
"pure_decode_support_coverage": 1.0,
"run_dir": "runs/phase5/primary/A3-r1-C00/moderate",
"run_id": "A3-r1-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "5681d6fa26cdadd9d5438cad56f56ec232aba359ddf08605c06b0c77adb73b62",
"token_efficiency_per_ms": 2.6235339995000753,
"waiting_block_cv": 6.855654600401045,
"warmup_completions": 28,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 18400,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
96
],
[
"PIECEWISE",
128
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2544934) INFO 07-12 12:52:10 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2544934) INFO 07-12 12:52:10 [monitor.py:53] torch.compile took 5.86 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:05, 8.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:05, 8.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:05, 8.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:05, 8.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:01<00:04, 8.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 8.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 8.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 8.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 8.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 8.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 8.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 8.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:05<00:00, 8.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 10.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 11.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 10.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 10.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 160,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 10.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 161,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 206,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 28,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 16,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 0.6449486465588222,
"drain_quarantined": false,
"drain_seconds": 1.1097721990081482,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4708333333333333,
"prefix_hit_ratio": 0.017412117941776724,
"prefix_hits": 17328,
"prefix_present_steps": 18400,
"prefix_queries": 995169,
"pure_decode_bucket_tokens": 27142,
"pure_decode_padding_fraction": 0.059686095350379484,
"pure_decode_padding_tokens": 1620,
"pure_decode_steps": 18091,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7
],
"pure_decode_support_coverage": 1.0,
"run_dir": "runs/phase5/primary/A3-r2-C00/moderate",
"run_id": "A3-r2-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "acfaefdff2db8d474e014e05d7824558b776ec754092dab488e01d28775f3367",
"token_efficiency_per_ms": 2.628373282268568,
"waiting_block_cv": 6.855654600401044,
"warmup_completions": 28,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 18364,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
96
],
[
"PIECEWISE",
128
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
240
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
368
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2537992) INFO 07-12 12:45:13 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2537992) INFO 07-12 12:45:13 [monitor.py:53] torch.compile took 5.98 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 10.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 10.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:04, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:04, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 9.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 8.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 8.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 8.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 8.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 11.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:03, 11.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 11.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 11.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 11.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:00, 11.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 11.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 11.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 200,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 28,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"PIECEWISE",
272
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 16,
"warmup_long_min_tokens": 9015
},
"decode_B_block_cv": 0.6600548630249607,
"drain_quarantined": false,
"drain_seconds": 1.0868711060029455,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4708333333333333,
"prefix_hit_ratio": 0.017412117941776724,
"prefix_hits": 17328,
"prefix_present_steps": 18364,
"prefix_queries": 995169,
"pure_decode_bucket_tokens": 27198,
"pure_decode_padding_fraction": 0.06136480623575263,
"pure_decode_padding_tokens": 1669,
"pure_decode_steps": 18057,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7
],
"pure_decode_support_coverage": 1.0,
"run_dir": "runs/phase5/primary/A3-r3-C00/moderate",
"run_id": "A3-r3-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "fa64970b9b5e3ba633073bcdde02002e4dd463dc2ac7834c7f76192fb8e75834",
"token_efficiency_per_ms": 2.6265598098178837,
"waiting_block_cv": 6.855654600401045,
"warmup_completions": 28,
"warmup_gate_branch": "A-P5-1-cold-start"
}
]
},
"A4": {
"ci95": [
2.760493550434248,
3.274105133500723
],
"point": 3.0209586215373587,
"runs": [
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11279,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2529899) INFO 07-12 12:36:40 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2529899) INFO 07-12 12:36:40 [monitor.py:53] torch.compile took 5.91 s in total",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 9.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 9.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.53it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 9.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 8.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:03, 8.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:03, 8.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 8.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 8.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:02, 8.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 8.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:04<00:01, 8.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 8.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 8.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 8.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 8.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:05<00:00, 7.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:05<00:00, 8.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 7.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:06<00:00, 6.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:06<00:00, 8.47it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 2%|\u258f | 1/51 [00:00<00:05, 9.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 6%|\u258c | 3/51 [00:00<00:04, 9.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 10%|\u2589 | 5/51 [00:00<00:04, 10.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 14%|\u2588\u258e | 7/51 [00:00<00:04, 10.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 18%|\u2588\u258a | 9/51 [00:00<00:03, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:03, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 11.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:01<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:02, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 11.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:01, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:02<00:01, 11.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 10.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:03<00:01, 10.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:03<00:00, 10.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:03<00:00, 10.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:04<00:00, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:04<00:00, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 204,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8687639860916586,
"drain_quarantined": false,
"drain_seconds": 0.6976415269891731,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11279,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30210,
"pure_decode_padding_fraction": 0.13518702416418404,
"pure_decode_padding_tokens": 4084,
"pure_decode_steps": 11035,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9866787494336203,
"run_dir": "runs/phase5/primary/A4-r1-A4/moderate",
"run_id": "A4-r1-A4-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "b167195bcdd3cdcf62883b6f3cb6f03a31358fb47f35e778badc0668e87e1301",
"token_efficiency_per_ms": 3.0182905257093,
"waiting_block_cv": 6.31543696870254,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11298,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
168
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2544936) INFO 07-12 12:52:10 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2544936) INFO 07-12 12:52:10 [monitor.py:53] torch.compile took 5.65 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 10.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 9.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 10.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 11.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 11.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 10.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 10.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 10.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.44it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 204,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8691990645892419,
"drain_quarantined": false,
"drain_seconds": 0.7184635890007485,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11298,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30204,
"pure_decode_padding_fraction": 0.13544563633955767,
"pure_decode_padding_tokens": 4091,
"pure_decode_steps": 11053,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9865194969691486,
"run_dir": "runs/phase5/primary/A4-r2-A4/moderate",
"run_id": "A4-r2-A4-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "982b3429d41fd3fec33ac9afdbbfe32faa49e9eaa706c95a51ccab574cec6fb8",
"token_efficiency_per_ms": 3.0207115888946916,
"waiting_block_cv": 6.32197662519898,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11336,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
168
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2551429) INFO 07-12 12:58:42 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783861122.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2551429) INFO 07-12 12:58:42 [monitor.py:53] torch.compile took 5.67 s in total",
"phase": "pre-ready",
"timestamp_s": 1783861122.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 9.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 9.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.52it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 8.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:03, 7.59it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 7.45it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:03, 7.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.99it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 9.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.30it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 9.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 8.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 11.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 10.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 10.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 10.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 10.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 10.54it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:03, 10.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:02, 10.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 10.43it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 10.62it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 10.74it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 10.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 10.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 10.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 202,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
32
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8688877121485955,
"drain_quarantined": false,
"drain_seconds": 0.7056757730024401,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11336,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30163,
"pure_decode_padding_fraction": 0.13403839140669033,
"pure_decode_padding_tokens": 4043,
"pure_decode_steps": 11091,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9874673158416735,
"run_dir": "runs/phase5/primary/A4-r3-A4/moderate",
"run_id": "A4-r3-A4-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "b1f305bfe2b70523c10b79f011ff04aed84d5b8ba943aa63ec15fd3d4783f310",
"token_efficiency_per_ms": 3.023878963072675,
"waiting_block_cv": 6.3162574579358255,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
}
]
},
"base": {
"ci95": [
2.75446150338882,
3.267963655381781
],
"point": 3.0137518658413924,
"runs": [
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11178,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2537999) INFO 07-12 12:45:13 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2537999) INFO 07-12 12:45:13 [monitor.py:53] torch.compile took 5.82 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860313.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:04, 10.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 8%|\u258a | 4/51 [00:00<00:04, 10.16it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.01it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.68it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 8.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 8.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 8.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 8.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 8.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.37it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.49it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.70it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 5.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 6.39it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.61it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.63it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 2%|\u258f | 1/51 [00:00<00:05, 9.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 6%|\u258c | 3/51 [00:00<00:04, 10.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 10%|\u2589 | 5/51 [00:00<00:04, 10.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 14%|\u2588\u258e | 7/51 [00:00<00:04, 10.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 18%|\u2588\u258a | 9/51 [00:00<00:04, 10.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:03, 10.24it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 10.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 10.50it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 10.72it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:02, 10.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:01<00:02, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:02, 10.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:02, 10.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 10.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:02<00:02, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:02<00:01, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:01, 10.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:03<00:01, 10.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:03<00:01, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:03<00:00, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 10.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 10.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:04<00:00, 10.70it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:04<00:00, 10.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 204,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8690218691128104,
"drain_quarantined": false,
"drain_seconds": 0.7464378089935053,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11178,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30310,
"pure_decode_padding_fraction": 0.13873309138898055,
"pure_decode_padding_tokens": 4205,
"pure_decode_steps": 10934,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9849094567404426,
"run_dir": "runs/phase5/primary/base-r1-C00/moderate",
"run_id": "base-r1-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "7343f40c8a2fc814d2ed834bdef903162bac4404f29b98cd9d7d9a9c35ada749",
"token_efficiency_per_ms": 3.0116380077286706,
"waiting_block_cv": 5.413884524461967,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11265,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2529914) INFO 07-12 12:36:40 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2529914) INFO 07-12 12:36:40 [monitor.py:53] torch.compile took 6.01 s in total",
"phase": "pre-ready",
"timestamp_s": 1783859800.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 8.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 4%|\u258d | 2/51 [00:00<00:05, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:05, 9.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 9.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 14%|\u2588\u258e | 7/51 [00:00<00:04, 9.75it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 8.42it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 20%|\u2588\u2589 | 10/51 [00:01<00:04, 8.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:04, 9.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.34it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:02<00:03, 9.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 8.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 8.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:03, 8.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.41it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 9.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 9.06it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.94it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 8.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 9.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 9.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 9.14it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 9.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:00, 9.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.60it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 7.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.38it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.11it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 6.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 130,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 6.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 6.69it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.66it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.17it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 11.22it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.29it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.33it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.35it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.40it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.36it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 11.23it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 11.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 11.21it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 11.15it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:03<00:00, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.92it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 158,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 159,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 160,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 11.19it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 205,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8690061140995009,
"drain_quarantined": false,
"drain_seconds": 0.7273033229866996,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11265,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30197,
"pure_decode_padding_fraction": 0.13534457065271385,
"pure_decode_padding_tokens": 4087,
"pure_decode_steps": 11021,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9864803556846021,
"run_dir": "runs/phase5/primary/base-r2-C00/moderate",
"run_id": "base-r2-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "824f1ed4af5e9458f188bf7fd69259b1a122659915401d9e275620c1c864a3d7",
"token_efficiency_per_ms": 3.0164059164480457,
"waiting_block_cv": 6.315997149746135,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
},
{
"clean_duration_s": 240.0,
"clean_failed": 0,
"clean_steps": 11232,
"client_invariants": {
"clean_duration": true,
"clean_failures_zero": true,
"client_sanity": true,
"drain_re_adjudicated": true,
"failed_records_accounted": true,
"manifest_no_wrap": true,
"profile_count": true,
"warmup_cold_start_gate": true
},
"cold_start_gate": {
"amendment": "A-P5-1",
"clean_capture_events": 0,
"clean_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
16
],
[
"PIECEWISE",
24
],
[
"PIECEWISE",
48
],
[
"PIECEWISE",
72
],
[
"PIECEWISE",
104
],
[
"PIECEWISE",
200
],
[
"PIECEWISE",
224
],
[
"PIECEWISE",
248
],
[
"PIECEWISE",
320
],
[
"PIECEWISE",
352
],
[
"PIECEWISE",
384
],
[
"PIECEWISE",
400
],
[
"PIECEWISE",
416
],
[
"PIECEWISE",
432
],
[
"PIECEWISE",
448
],
[
"PIECEWISE",
480
]
],
"compile_capture_events": [
{
"line": 66,
"message_prefix": "(EngineCore pid=2544921) INFO 07-12 12:52:10 [decorators.py:311] Directly load AOT compilation from path /home/admin/cpfs/wjh/.cache/vllm/torch_compile_cache/torch_aot_compile/7a0273078f7c7f7f0453a4a0",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 67,
"message_prefix": "(EngineCore pid=2544921) INFO 07-12 12:52:10 [monitor.py:53] torch.compile took 5.74 s in total",
"phase": "pre-ready",
"timestamp_s": 1783860730.0
},
{
"line": 81,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 82,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 2%|\u258f | 1/51 [00:00<00:05, 9.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 83,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 6%|\u258c | 3/51 [00:00<00:04, 10.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 84,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 10%|\u2589 | 5/51 [00:00<00:04, 10.00it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 85,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 12%|\u2588\u258f | 6/51 [00:00<00:04, 9.93it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 86,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 16%|\u2588\u258c | 8/51 [00:00<00:04, 9.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 87,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 18%|\u2588\u258a | 9/51 [00:00<00:04, 9.27it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 88,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 22%|\u2588\u2588\u258f | 11/51 [00:01<00:04, 9.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 89,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:04, 9.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 90,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 25%|\u2588\u2588\u258c | 13/51 [00:01<00:03, 9.51it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 91,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 9.48it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 92,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 29%|\u2588\u2588\u2589 | 15/51 [00:01<00:03, 9.26it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 93,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 9.31it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 94,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 33%|\u2588\u2588\u2588\u258e | 17/51 [00:01<00:03, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 95,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:03, 9.25it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 96,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 37%|\u2588\u2588\u2588\u258b | 19/51 [00:01<00:03, 9.32it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 97,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:02<00:03, 9.20it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 98,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 41%|\u2588\u2588\u2588\u2588 | 21/51 [00:02<00:03, 9.12it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 99,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:02<00:03, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 100,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 45%|\u2588\u2588\u2588\u2588\u258c | 23/51 [00:02<00:03, 9.18it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 101,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 9.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 102,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 49%|\u2588\u2588\u2588\u2588\u2589 | 25/51 [00:02<00:03, 8.55it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 103,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 8.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 104,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 53%|\u2588\u2588\u2588\u2588\u2588\u258e | 27/51 [00:02<00:02, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 105,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:03<00:02, 8.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 106,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 57%|\u2588\u2588\u2588\u2588\u2588\u258b | 29/51 [00:03<00:02, 8.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 107,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:03<00:02, 8.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 108,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 61%|\u2588\u2588\u2588\u2588\u2588\u2588 | 31/51 [00:03<00:02, 8.83it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 109,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:03<00:02, 8.90it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 110,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 65%|\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 33/51 [00:03<00:02, 8.78it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 111,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 8.89it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 112,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 69%|\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 35/51 [00:03<00:01, 8.88it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 113,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 8.91it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 114,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 73%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 37/51 [00:04<00:01, 8.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 115,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:04<00:01, 8.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 116,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 76%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 39/51 [00:04<00:01, 8.71it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 117,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:04<00:01, 8.58it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 118,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 80%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 41/51 [00:04<00:01, 8.65it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 119,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:04<00:01, 8.67it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 120,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 84%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 43/51 [00:04<00:00, 8.76it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 121,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 8.28it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 122,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 88%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 45/51 [00:04<00:00, 8.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 123,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:05<00:00, 8.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 124,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 92%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f| 47/51 [00:05<00:00, 7.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 125,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:05<00:00, 7.64it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 126,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 96%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258c| 49/51 [00:05<00:00, 7.56it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 127,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:05<00:00, 7.57it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 128,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 7.46it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 129,
"message_prefix": "Capturing CUDA graphs (mixed prefill-decode, PIECEWISE): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:05<00:00, 8.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 131,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 0%| | 0/51 [00:00<?, ?it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 132,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 4%|\u258d | 2/51 [00:00<00:04, 10.84it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 133,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 8%|\u258a | 4/51 [00:00<00:04, 11.07it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 134,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 12%|\u2588\u258f | 6/51 [00:00<00:04, 11.05it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 135,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 16%|\u2588\u258c | 8/51 [00:00<00:03, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 136,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 20%|\u2588\u2589 | 10/51 [00:00<00:03, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 137,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 24%|\u2588\u2588\u258e | 12/51 [00:01<00:03, 11.09it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 138,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 27%|\u2588\u2588\u258b | 14/51 [00:01<00:03, 11.13it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 139,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 31%|\u2588\u2588\u2588\u258f | 16/51 [00:01<00:03, 11.10it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 140,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 35%|\u2588\u2588\u2588\u258c | 18/51 [00:01<00:02, 11.08it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 141,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 39%|\u2588\u2588\u2588\u2589 | 20/51 [00:01<00:02, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 142,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 43%|\u2588\u2588\u2588\u2588\u258e | 22/51 [00:01<00:02, 11.04it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 143,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 47%|\u2588\u2588\u2588\u2588\u258b | 24/51 [00:02<00:02, 11.02it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 144,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 51%|\u2588\u2588\u2588\u2588\u2588 | 26/51 [00:02<00:02, 10.95it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 145,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 55%|\u2588\u2588\u2588\u2588\u2588\u258d | 28/51 [00:02<00:02, 10.96it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 146,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 59%|\u2588\u2588\u2588\u2588\u2588\u2589 | 30/51 [00:02<00:01, 10.98it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 147,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 63%|\u2588\u2588\u2588\u2588\u2588\u2588\u258e | 32/51 [00:02<00:01, 10.86it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 148,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 67%|\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 34/51 [00:03<00:01, 10.87it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 149,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 71%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 36/51 [00:03<00:01, 10.77it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 150,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 75%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d | 38/51 [00:03<00:01, 10.82it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 151,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 78%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a | 40/51 [00:03<00:01, 10.85it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 152,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 82%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258f | 42/51 [00:03<00:00, 10.80it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 153,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 86%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258b | 44/51 [00:04<00:00, 10.81it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 154,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 90%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588 | 46/51 [00:04<00:00, 10.73it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 155,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 94%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258d| 48/51 [00:04<00:00, 10.79it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 156,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 98%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u258a| 50/51 [00:04<00:00, 11.03it/s]",
"phase": "pre-ready",
"timestamp_s": null
},
{
"line": 157,
"message_prefix": "Capturing CUDA graphs (decode, FULL): 100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 51/51 [00:04<00:00, 10.97it/s]",
"phase": "pre-ready",
"timestamp_s": null
}
],
"event_gate_passed": true,
"invariants": {
"clean_descriptors_covered": true,
"events_preclean": true,
"startup_modes_complete": true,
"startup_sizes_present": true,
"warmup_completions_ge_16": true,
"warmup_long_completion": true,
"zero_clean_capture_events": true
},
"passed": true,
"server_ready_line": 202,
"startup_capture_modes": [
"FULL",
"PIECEWISE"
],
"startup_capture_sizes": [
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,
272,
288,
304,
320,
336,
352,
368,
384,
400,
416,
432,
448,
464,
480,
496,
512
],
"uncovered_clean_descriptors": [],
"warmup_completions": 25,
"warmup_descriptors": [
[
"FULL",
1
],
[
"FULL",
2
],
[
"FULL",
4
],
[
"FULL",
8
],
[
"FULL",
16
],
[
"PIECEWISE",
40
],
[
"PIECEWISE",
400
]
],
"warmup_long_completions": 14,
"warmup_long_min_tokens": 9915
},
"decode_B_block_cv": 0.8739119293684442,
"drain_quarantined": false,
"drain_seconds": 0.7502137859992217,
"layer1_invariants": {
"cudagraph_identity": true,
"footer_balanced": true,
"footer_written_matches": true,
"schema_1": true,
"sidecar_agrees": true,
"sidecar_final": true,
"steps_unique_contiguous": true,
"token_composition": true,
"zero_drops": true
},
"offered_rps": 0.4875,
"prefix_hit_ratio": 0.017100051594983262,
"prefix_hits": 17632,
"prefix_present_steps": 11232,
"prefix_queries": 1031108,
"pure_decode_bucket_tokens": 30296,
"pure_decode_padding_fraction": 0.13813704779508845,
"pure_decode_padding_tokens": 4185,
"pure_decode_steps": 10989,
"pure_decode_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"pure_decode_support_coverage": 0.9845299845299845,
"run_dir": "runs/phase5/primary/base-r3-C00/moderate",
"run_id": "base-r3-C00-moderate",
"server_invariants": {
"a2_sizes": true,
"chunked_mbt": true,
"drain_shutdown": true,
"tp1": true,
"triton_moe": true
},
"stream_sha256": "f4149c9650466b10af112eb2c69b00741b6a9098b5c875ef966da0876158fee0",
"token_efficiency_per_ms": 3.013215573434887,
"waiting_block_cv": 6.313666777381936,
"warmup_completions": 25,
"warmup_gate_branch": "A-P5-1-cold-start"
}
]
}
},
"gpu": {
"completed_burnins": 3,
"completed_measured_runs_including_background": 16,
"controller_status": "primary_complete",
"hard_cap": 6.0,
"new_h20_hours": 2.438816744287809
},
"holm": {
"adjusted_p": {
"A1": 1.0,
"A2": 1.0,
"A3": 0.05536,
"A4": 1.0
},
"family": [
"A1",
"A2",
"A3",
"A4"
],
"raw_p": {
"A1": 0.73108,
"A2": 0.79022,
"A3": 0.01384,
"A4": 0.9723
}
},
"ledgers": {
"P03": {
"base_E": 3.0137518658413924,
"control_E": 4.735599715364607,
"control_source": "P3-reused",
"denominator_nonpositive_fraction": 0.0,
"diagnostic_share_sum": -0.15493134437905817,
"gap_E": 1.7218478495232148,
"gap_ci95": [
1.4590921308319604,
1.9887718695651007
],
"mechanisms": {
"A1": {
"delta_E": 0.06524706407720116,
"delta_E_ci95": [
-0.309460837248546,
0.44256485170072374
],
"diagnostic_share": 0.03789362927465878,
"diagnostic_share_ci95": [
-0.20110135273946905,
0.23440437178025111
],
"manipulation_passed": true,
"p_holm": 1.0,
"p_two_sided": 0.73108,
"share": 0.03789362927465878,
"share_ci95": [
-0.20110135273946905,
0.23440437178025111
],
"share_status": "REPORTABLE"
},
"A2": {
"delta_E": 0.048375663195566077,
"delta_E_ci95": [
-0.31536063261129355,
0.4114458465370726
],
"diagnostic_share": 0.028095202029007064,
"diagnostic_share_ci95": [
-0.20491850311211388,
0.21696446278610113
],
"manipulation_passed": false,
"p_holm": 1.0,
"p_two_sided": 0.79022,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014manipulation failed"
},
"A3": {
"delta_E": -0.38759768511155546,
"delta_E_ci95": [
-0.6934442672980398,
-0.07814103201161723
],
"diagnostic_share": -0.225105653335678,
"diagnostic_share_ci95": [
-0.46252418395426365,
-0.04030479637326536
],
"manipulation_passed": false,
"p_holm": 0.05536,
"p_two_sided": 0.01384,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014manipulation failed"
},
"A4": {
"delta_E": 0.007206755695966294,
"delta_E_ci95": [
-0.35763676876840356,
0.36930915635169426
],
"diagnostic_share": 0.004185477652953986,
"diagnostic_share_ci95": [
-0.23399192365687496,
0.1948154488970133
],
"manipulation_passed": false,
"p_holm": 1.0,
"p_two_sided": 0.9723,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014manipulation failed"
}
},
"residual_interaction": 1.154931344379058,
"residual_interaction_ci95": [
0.5535562238890501,
1.9397452067536105
],
"residual_status": "DIAGNOSTIC_ONLY\u2014incomplete official share ledger",
"stable_denominator": true,
"status": "EVALUABLE"
},
"P04": {
"base_E": 3.0137518658413924,
"control_E": 3.0547035940055496,
"control_source": "P3-reused",
"denominator_nonpositive_fraction": 0.45353,
"diagnostic_share_sum": -6.5142110993086195,
"gap_E": 0.04095172816415715,
"gap_ci95": [
-0.6490964964724942,
0.7525939493359092
],
"mechanisms": {
"A1": {
"delta_E": 0.06524706407720116,
"delta_E_ci95": [
-0.309460837248546,
0.44256485170072374
],
"diagnostic_share": 1.5932676593196475,
"diagnostic_share_ci95": [
-6.892455369893276,
7.063609245006344
],
"manipulation_passed": true,
"p_holm": 1.0,
"p_two_sided": 0.73108,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014unstable control denominator"
},
"A2": {
"delta_E": 0.048375663195566077,
"delta_E_ci95": [
-0.31536063261129355,
0.4114458465370726
],
"diagnostic_share": 1.1812850242033668,
"diagnostic_share_ci95": [
-6.287075134424673,
6.4923565807582335
],
"manipulation_passed": false,
"p_holm": 1.0,
"p_two_sided": 0.79022,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014unstable control denominator"
},
"A3": {
"delta_E": -0.38759768511155546,
"delta_E_ci95": [
-0.6934442672980398,
-0.07814103201161723
],
"diagnostic_share": -9.464745506168867,
"diagnostic_share_ci95": [
-17.244996602975487,
17.410215608834356
],
"manipulation_passed": false,
"p_holm": 0.05536,
"p_two_sided": 0.01384,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014unstable control denominator"
},
"A4": {
"delta_E": 0.007206755695966294,
"delta_E_ci95": [
-0.35763676876840356,
0.36930915635169426
],
"diagnostic_share": 0.17598172333723341,
"diagnostic_share_ci95": [
-6.258975640244697,
6.49251601129639
],
"manipulation_passed": false,
"p_holm": 1.0,
"p_two_sided": 0.9723,
"share": null,
"share_ci95": null,
"share_status": "N/A\u2014unstable control denominator"
}
},
"residual_interaction": 7.5142110993086195,
"residual_interaction_ci95": [
-21.39654327199287,
22.87671051130793
],
"residual_status": "DIAGNOSTIC_ONLY\u2014incomplete official share ledger",
"stable_denominator": false,
"status": "INCONCLUSIVE\u2014unstable denominator"
}
},
"limitation": "Recorded-arrival P5 bridge ledger anchored to P3 controls; not a literal decomposition of P3's already-uniform P10 gap.",
"manipulations": {
"A1": {
"ablated_R16": 0.4734094392575692,
"base_R16": 0.6417435053185918,
"max_added_delay_seconds": 62.7431113362687,
"passed": true,
"prefix_hit_ratio_delta": -0.0029644971629866675
},
"A2": {
"ablated_padding_fraction": 0.02398892811002051,
"base_padding_fraction": 0.1374049032789276,
"observed_support": [
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
11,
12,
13
],
"padding_reduction": 0.8254143226510356,
"passed": false,
"support_coverage": 0.9887891670550175
},
"A3": {
"ablated_decode_B_cv": 0.6551380253657898,
"ablated_waiting_cv": 6.855654600401045,
"base_decode_B_cv": 0.8706466375269185,
"base_waiting_cv": 6.014516150530013,
"passed": false
},
"A4": {
"passed": false,
"prefix_hits": 52896,
"prefix_queries": 3093324
}
},
"p3_base_E": 2.6191132082635966,
"sanity": {
"declared": {
"bridge_reuse_passed": true,
"control_sources": {
"P03": "P3-reused",
"P04": "P3-reused"
},
"manipulation_failures": [
"A2",
"A3",
"A4"
]
},
"invariants": {
"a_p5_1_cold_start_gates": true,
"bridge_decision_resolved": true,
"clean_duration_240": true,
"clean_failures_zero": true,
"client_invariants": true,
"control_denominators_stable": false,
"drain_quarantine_under_20pct": true,
"gpu_budget_below_6": true,
"layer1_accounting": true,
"manifests_same_ids": true,
"manifests_same_token_sums": true,
"offered_rate_within_5pct": true,
"per_arm_results_not_all_identical": true,
"primary_runs_15": true,
"ratios_finite": true,
"server_invariants": true,
"three_replicates_per_arm": true
},
"numeric": {
"clean_steps": {
"distinct_n": 15,
"finite_n": 15,
"max": 18400.0,
"min": 11178.0,
"missing_n": 0,
"n": 15
},
"diagnostic_share": {
"distinct_n": 8,
"finite_n": 8,
"max": 1.5932676593196475,
"min": -9.464745506168867,
"missing_n": 0,
"n": 8
},
"drain_seconds": {
"distinct_n": 15,
"finite_n": 15,
"max": 6.003959469002439,
"min": 0.6761808950104751,
"missing_n": 0,
"n": 15
},
"offered_rps": {
"distinct_n": 2,
"finite_n": 15,
"max": 0.4875,
"min": 0.4708333333333333,
"missing_n": 0,
"n": 15
},
"primary_E": {
"distinct_n": 15,
"finite_n": 15,
"max": 3.083262835538282,
"min": 2.6235339995000753,
"missing_n": 0,
"n": 15
},
"residual_interaction": {
"distinct_n": 2,
"finite_n": 2,
"max": 7.5142110993086195,
"min": 1.154931344379058,
"missing_n": 0,
"n": 2
},
"share_ci_width": {
"distinct_n": 8,
"finite_n": 8,
"max": 34.65521221180984,
"min": 0.421882965898215,
"missing_n": 0,
"n": 8
}
},
"red_flags": [
"control_denominators_stable"
]
},
"schema": 1,
"status": "INCONCLUSIVE_OR_PARTIAL"
}
Reference in New Issue View Git Blame Copy Permalink