diff --git a/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png b/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png new file mode 100644 index 0000000..afe8d37 Binary files /dev/null and b/docs/assets/simulator-fidelity/qwen30-prefill-ranking.png differ diff --git a/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py b/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py new file mode 100644 index 0000000..a7260bf --- /dev/null +++ b/runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py @@ -0,0 +1,462 @@ +#!/usr/bin/env python3 +"""Compare the frozen Frontier surface with conservative real capacities.""" + +from __future__ import annotations + +import argparse +import csv +import hashlib +import json +import math +import re +from collections import defaultdict +from datetime import datetime +from pathlib import Path +from typing import Any + + +RUN_PATTERN = re.compile(r"qwen30-prefill-real-tp(?P\d+)-mns(?P\d+)-") + + +def parse_args() -> argparse.Namespace: + parser = argparse.ArgumentParser() + parser.add_argument("--fleet-artifacts", type=Path, required=True) + parser.add_argument( + "--simulator-manifest", type=Path, action="append", required=True + ) + parser.add_argument("--output-root", type=Path, required=True) + return parser.parse_args() + + +def sha256(path: Path) -> str: + digest = hashlib.sha256() + with path.open("rb") as handle: + for chunk in iter(lambda: handle.read(1024 * 1024), b""): + digest.update(chunk) + return digest.hexdigest() + + +def load_json(path: Path) -> dict[str, Any]: + return json.loads(path.read_text()) + + +def sign(value: float) -> int: + return (value > 0) - (value < 0) + + +def kendall_tau_b(real: list[float], simulated: list[float]) -> dict[str, Any]: + if len(real) != len(simulated): + raise ValueError("ranking vectors have different lengths") + concordant = discordant = real_only_ties = simulated_only_ties = both_ties = 0 + for left in range(len(real)): + for right in range(left + 1, len(real)): + real_sign = sign(real[left] - real[right]) + sim_sign = sign(simulated[left] - simulated[right]) + if real_sign == 0 and sim_sign == 0: + both_ties += 1 + elif real_sign == 0: + real_only_ties += 1 + elif sim_sign == 0: + simulated_only_ties += 1 + elif real_sign == sim_sign: + concordant += 1 + else: + discordant += 1 + denominator = math.sqrt( + (concordant + discordant + real_only_ties) + * (concordant + discordant + simulated_only_ties) + ) + tau = (concordant - discordant) / denominator if denominator else None + return { + "kendall_tau_b": tau, + "concordant": concordant, + "discordant": discordant, + "real_only_ties": real_only_ties, + "simulator_only_ties": simulated_only_ties, + "both_ties": both_ties, + } + + +def result_files_by_anchor(root: Path) -> dict[tuple[str, str], Path]: + selected: dict[tuple[str, str], Path] = {} + digests: dict[tuple[str, str], str] = {} + for path in root.glob("artifacts/**/round*/results/r*.json"): + if path.name.startswith("warmup_"): + continue + round_name = path.parent.parent.name + key = (round_name, path.name) + digest = sha256(path) + if key in digests and digests[key] != digest: + raise RuntimeError( + f"conflicting duplicate real anchor {key} under {root}" + ) + digests[key] = digest + if key not in selected or len(path.parts) < len(selected[key].parts): + selected[key] = path + return selected + + +def find_real_runs(root: Path) -> dict[str, Path]: + candidates: dict[str, list[Path]] = defaultdict(list) + for path in root.iterdir(): + if not path.is_dir(): + continue + match = RUN_PATTERN.search(path.name) + if not match: + continue + name = f"tp{int(match.group('tp'))}_mns{int(match.group('mns'))}" + candidates[name].append(path) + selected: dict[str, Path] = {} + for name, paths in candidates.items(): + complete = [ + path + for path in paths + if (path / "remote_run" / "exit_code").is_file() + and (path / "remote_run" / "exit_code").read_text().strip() == "0" + ] + measured_counts = { + path: len(result_files_by_anchor(path)) + for path in complete + } + if not measured_counts: + raise RuntimeError(f"no successful run for {name}") + maximum = max(measured_counts.values()) + richest = [path for path, count in measured_counts.items() if count == maximum] + if len(richest) != 1: + raise RuntimeError(f"ambiguous richest successful run for {name}: {richest}") + selected[name] = richest[0] + if len(selected) != 12: + raise RuntimeError(f"expected 12 real configs, got {sorted(selected)}") + return selected + + +def campaign_resources(root: Path) -> dict[str, Any]: + runs = [] + gpu_hours = 0.0 + for path in sorted(root.iterdir()): + match = RUN_PATTERN.search(path.name) + exit_code = path / "remote_run" / "exit_code" + started_at = path / "remote_run" / "started_at" + finished_at = path / "remote_run" / "finished_at" + if ( + not path.is_dir() + or not match + or not exit_code.is_file() + or exit_code.read_text().strip() != "0" + or not started_at.is_file() + or not finished_at.is_file() + ): + continue + started = datetime.fromisoformat(started_at.read_text().strip()) + finished = datetime.fromisoformat(finished_at.read_text().strip()) + duration_seconds = (finished - started).total_seconds() + tp = int(match.group("tp")) + run_gpu_hours = duration_seconds * tp / 3600.0 + gpu_hours += run_gpu_hours + runs.append( + { + "run": path.name, + "tp": tp, + "duration_seconds": duration_seconds, + "gpu_hours": run_gpu_hours, + } + ) + return { + "successful_fleet_jobs": len(runs), + "gpu_hours": gpu_hours, + "runs": runs, + } + + +def parse_real_config(name: str, run_root: Path) -> dict[str, Any]: + result_files = sorted(result_files_by_anchor(run_root).values()) + if len(result_files) not in {10, 16}: + raise RuntimeError(f"expected 10 base or 16 refined anchors for {name}, got {len(result_files)}") + by_rate: dict[float, list[dict[str, Any]]] = defaultdict(list) + for path in result_files: + payload = load_json(path) + rate = float(payload["workload"]["offered_request_rate"]) + by_rate[rate].append( + { + "path": str(path.resolve()), + "sha256": sha256(path), + "summary": payload["summary"], + } + ) + tp = int(name.split("_")[0][2:]) + base_rates = [4.0, 8.0, 16.0, 32.0, 64.0] + refined_rates = sorted({*base_rates, *(tp * value for value in (5.0, 6.0, 7.0))}) + if tuple(sorted(by_rate)) not in {tuple(base_rates), tuple(refined_rates)}: + raise RuntimeError(f"unexpected rate grid for {name}: {sorted(by_rate)}") + anchors = [] + for rate, rounds in sorted(by_rate.items()): + if len(rounds) != 2: + raise RuntimeError(f"expected two rounds for {name}@{rate}, got {len(rounds)}") + round_feasible = [bool(row["summary"]["slo"]["feasible"]) for row in rounds] + anchors.append( + { + "rate": rate, + "rounds": rounds, + "conservative_feasible": all(round_feasible), + "round_feasible": round_feasible, + "round_ttft_p95_ms": [ + float(row["summary"]["ttft_p95_ms"]) for row in rounds + ], + } + ) + feasible = [row["rate"] for row in anchors if row["conservative_feasible"]] + capacity = max(feasible, default=0.0) + return { + "name": name, + "tp": tp, + "mns": int(name.split("_mns")[1]), + "anchors": anchors, + "capacity": capacity, + "capacity_per_gpu": capacity / tp, + "source_run": str(run_root.resolve()), + } + + +def parse_simulator(manifests: list[Path]) -> tuple[dict[str, Any], list[dict[str, str]]]: + configs: dict[str, Any] = {} + sources = [] + for path in manifests: + payload = load_json(path) + if payload["status"] not in {"complete", "partial_not_decision_bearing"}: + raise RuntimeError(f"simulator manifest has invalid status: {path}") + sources.append({"path": str(path.resolve()), "sha256": sha256(path)}) + result_by_name = { + result["config"]["name"]: result for result in payload["config_results"] + } + for capacity in payload["capacity"]: + name = capacity["config"]["name"] + result = result_by_name.get(name) + if result is None: + raise RuntimeError(f"missing simulator config result {name}") + entry = configs.setdefault( + name, + { + "name": name, + "tp": int(capacity["config"]["tp"]), + "mns": int(capacity["config"]["mns"]), + "anchor_by_rate": {}, + }, + ) + for load in result["loads"]: + rate = float(load["offered_request_rate"]) + if rate in entry["anchor_by_rate"]: + raise RuntimeError(f"duplicate simulator anchor {name}@{rate}") + entry["anchor_by_rate"][rate] = { + "rate": rate, + "feasible": bool(load["score"]["feasible"]), + "pass_rate": float(load["score"]["pass_rate"]), + "ttft_p95_ms": float(load["score"]["ttft_p95_ms"]), + } + if len(configs) != 12: + raise RuntimeError(f"expected 12 simulator configs, got {sorted(configs)}") + for entry in configs.values(): + entry["anchors"] = [ + entry["anchor_by_rate"][rate] for rate in sorted(entry["anchor_by_rate"]) + ] + del entry["anchor_by_rate"] + feasible = [row["rate"] for row in entry["anchors"] if row["feasible"]] + entry["capacity"] = max(feasible, default=0.0) + entry["capacity_per_gpu"] = entry["capacity"] / entry["tp"] + return configs, sources + + +def compare(real: dict[str, Any], simulated: dict[str, Any]) -> dict[str, Any]: + names = sorted(real, key=lambda name: (real[name]["tp"], real[name]["mns"])) + if set(names) != set(simulated): + raise RuntimeError("real and simulator config sets differ") + real_scores = [real[name]["capacity_per_gpu"] for name in names] + sim_scores = [simulated[name]["capacity_per_gpu"] for name in names] + real_best = max(real_scores) + sim_best = max(sim_scores) + real_top = [name for name in names if real[name]["capacity_per_gpu"] == real_best] + sim_top = [ + name for name in names if simulated[name]["capacity_per_gpu"] == sim_best + ] + worst_sim_choice = min(real[name]["capacity_per_gpu"] for name in sim_top) + best_sim_choice = max(real[name]["capacity_per_gpu"] for name in sim_top) + tau = kendall_tau_b(real_scores, sim_scores) + + pairwise = {"all": {"comparable": 0, "correct": 0}, "within_tp": {}} + for left in range(len(names)): + for right in range(left + 1, len(names)): + real_sign = sign(real_scores[left] - real_scores[right]) + sim_sign = sign(sim_scores[left] - sim_scores[right]) + if real_sign: + pairwise["all"]["comparable"] += 1 + pairwise["all"]["correct"] += int(real_sign == sim_sign) + if real[names[left]]["tp"] == real[names[right]]["tp"] and real_sign: + key = f"tp{real[names[left]]['tp']}" + bucket = pairwise["within_tp"].setdefault( + key, {"comparable": 0, "correct": 0} + ) + bucket["comparable"] += 1 + bucket["correct"] += int(real_sign == sim_sign) + for bucket in [pairwise["all"], *pairwise["within_tp"].values()]: + bucket["accuracy"] = ( + bucket["correct"] / bucket["comparable"] + if bucket["comparable"] + else None + ) + + confusion = {"real_pass_sim_pass": 0, "real_pass_sim_fail": 0, + "real_fail_sim_pass": 0, "real_fail_sim_fail": 0} + for name in names: + real_anchors = {row["rate"]: row for row in real[name]["anchors"]} + sim_anchors = {row["rate"]: row for row in simulated[name]["anchors"]} + if set(real_anchors) != set(sim_anchors): + raise RuntimeError(f"anchor grids differ for {name}") + for rate in real_anchors: + real_pass = real_anchors[rate]["conservative_feasible"] + sim_pass = sim_anchors[rate]["feasible"] + key = f"real_{'pass' if real_pass else 'fail'}_sim_{'pass' if sim_pass else 'fail'}" + confusion[key] += 1 + return { + "config_order": names, + "real_top_set": real_top, + "simulator_top_set": sim_top, + "top_set_exact_match": real_top == sim_top, + "top_set_overlap": sorted(set(real_top) & set(sim_top)), + "top1_regret_best": (real_best - best_sim_choice) / real_best, + "top1_regret_worst": (real_best - worst_sim_choice) / real_best, + "real_best_capacity_per_gpu": real_best, + "simulator_best_capacity_per_gpu": sim_best, + "kendall": tau, + "pairwise_non_tied": pairwise, + "anchor_confusion": confusion, + } + + +def write_csv(path: Path, rows: list[dict[str, Any]]) -> None: + with path.open("w", newline="") as handle: + writer = csv.DictWriter( + handle, fieldnames=list(rows[0]), lineterminator="\n" + ) + writer.writeheader() + writer.writerows(rows) + + +def plot(path: Path, rows: list[dict[str, Any]], metrics: dict[str, Any]) -> None: + import matplotlib.pyplot as plt + import numpy as np + + labels = [f"TP{row['tp']}\nMNS{row['mns']}" for row in rows] + x = np.arange(len(rows)) + width = 0.36 + figure, axes = plt.subplots( + 1, 2, figsize=(13.5, 5.0), gridspec_kw={"width_ratios": [3.3, 1.0]} + ) + axis = axes[0] + axis.bar(x - width / 2, [row["real"] for row in rows], width, label="Real vLLM") + axis.bar( + x + width / 2, + [row["simulator"] for row in rows], + width, + label="Frontier profile-only", + ) + axis.set_xticks(x, labels, fontsize=8) + axis.set_ylabel("Max tested SLO-feasible request rate / GPU") + axis.set_title("Qwen3-30B-A3B prefill-only: config ranking") + axis.grid(axis="y", alpha=0.25) + axis.legend(frameon=False, ncols=2) + for separator in (3.5, 7.5): + axis.axvline(separator, color="0.75", linewidth=0.8) + tau = metrics["kendall"]["kendall_tau_b"] + annotation = f"worst regret={metrics['top1_regret_worst'] * 100:.1f}%" + annotation += ( + f"\nKendall τ-b={tau:.3f}" + if tau is not None + else "\nKendall τ-b=undefined" + ) + axis.text( + 0.01, + 0.98, + annotation, + transform=axis.transAxes, + va="top", + fontsize=9, + bbox={"facecolor": "white", "edgecolor": "0.8", "alpha": 0.9}, + ) + + confusion = metrics["anchor_confusion"] + matrix = np.array( + [ + [confusion["real_pass_sim_pass"], confusion["real_pass_sim_fail"]], + [confusion["real_fail_sim_pass"], confusion["real_fail_sim_fail"]], + ] + ) + image = axes[1].imshow(matrix, cmap="Blues", vmin=0) + axes[1].set_xticks([0, 1], ["Sim pass", "Sim fail"]) + axes[1].set_yticks([0, 1], ["Real pass", "Real fail"]) + axes[1].set_title(f"{int(matrix.sum())} anchor decisions") + for row in range(2): + for column in range(2): + axes[1].text(column, row, int(matrix[row, column]), ha="center", va="center") + figure.colorbar(image, ax=axes[1], fraction=0.047, pad=0.04) + figure.suptitle("ISL=2048, OSL=1, TTFT≤1256 ms, 95% pass gate; two real rounds") + figure.tight_layout() + figure.savefig(path, dpi=180, bbox_inches="tight") + plt.close(figure) + + +def main() -> None: + args = parse_args() + real_runs = find_real_runs(args.fleet_artifacts.resolve()) + real = {name: parse_real_config(name, path) for name, path in real_runs.items()} + simulated, simulator_sources = parse_simulator( + [path.resolve() for path in args.simulator_manifest] + ) + metrics = compare(real, simulated) + rows = [ + { + "config": name, + "tp": real[name]["tp"], + "mns": real[name]["mns"], + "real": real[name]["capacity_per_gpu"], + "simulator": simulated[name]["capacity_per_gpu"], + } + for name in metrics["config_order"] + ] + resources = campaign_resources(args.fleet_artifacts.resolve()) + resources["fresh_server_anchors"] = sum( + len(config["anchors"]) * 2 for config in real.values() + ) + resources["measured_requests"] = resources["fresh_server_anchors"] * 64 + resources["warmup_requests"] = sum( + min(32, max(4, math.ceil(anchor["rate"] * 2.0))) * 2 + for config in real.values() + for anchor in config["anchors"] + ) + args.output_root.mkdir(parents=True, exist_ok=True) + payload = { + "schema": "qwen30-prefill-fidelity-comparison-v1", + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "contract": { + "model": "Qwen3-30B-A3B", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": False, + "ttft_slo_ms": 1256.0, + "target_pass_rate": 0.95, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + }, + "metrics": metrics, + "real_campaign_resources": resources, + "real": real, + "simulator": simulated, + "simulator_sources": simulator_sources, + } + (args.output_root / "comparison.json").write_text( + json.dumps(payload, indent=2, sort_keys=True) + "\n" + ) + write_csv(args.output_root / "capacity.csv", rows) + plot(args.output_root / "qwen30-prefill-ranking.png", rows, metrics) + print(json.dumps(metrics, indent=2, sort_keys=True)) + + +if __name__ == "__main__": + main() diff --git a/runs/frontier-phase-factorial-v0/experiment-card.md b/runs/frontier-phase-factorial-v0/experiment-card.md index f291ab7..b2623c6 100644 --- a/runs/frontier-phase-factorial-v0/experiment-card.md +++ b/runs/frontier-phase-factorial-v0/experiment-card.md @@ -1,6 +1,6 @@ # 实验 EXP-SIMFID-PHASE-FACTORIAL：prefill-only 是否是 simulator ranking 的容易区间？ -> **状态：** 已批准，运行中（用户于 2026-07-17 明确要求先完成 235B mixed 与 30B prefill-only） +> **状态：** 已完成（2026-07-17） ## Claim 与决策 @@ -17,6 +17,7 @@ - **30B system context：** community vLLM 0.20.0+cu129，BF16 weights/activation/KV，TP∈{1,2,4}，MNS∈{8,16,32,64}，MBT=8192，chunked prefill on，prefix off；real 保留 runtime 默认 CUDA graph，Frontier profile-only 不做 E2E calibration。 - **30B workload：** fixed ISL=2048、OSL=1，64 个不同 token-chain prompts，uniform open-loop QPS；fresh server per `(config, rate, round)`，target-rate warmup 与 measured requests 分离。 - **30B SLO：** TTFT≤1256 ms，至少 61/64 requests 通过；primary score 为最大共同 tested feasible req/s / 实际 TP GPUs。 +- **Boundary refinement rule：** base grid `{4,8,16,32,64}` 先定位每个 TP 的 pass→fail 区间；若除以 TP 后的离散容量产生无法区分的 top tie，则在查看最终 ranking 前追加共同 per-GPU lattice `5/6/7 req/s/GPU`，即 TP1 测 `{5,6,7}`、TP2 测 `{10,12,14}`、TP4 测 `{20,24,28}`。refinement 不替换或删除 base anchors。 - **235B baseline：** 已冻结 `ISL=2048, OSL=128`、8 configs、68 个 fresh-server anchors；primary sensitivity TTFT≤1256 ms、TPOT≤150 ms。 - **Baselines：** real community vLLM；Frontier same-stack profile-only；historical Qwen30 mixed profile-only；historical frozen per-TP calibration 只作为 upper bound，不参与本 case 拟合。 - **Metrics：** top set、worst tie-break regret、Kendall τ-b、exact/non-tied pair direction、anchor confusion、absolute capacity、TTFT p50/p95、real trial variance与GPU-hour。 @@ -35,7 +36,7 @@ | Selective benchmarking | PASS for initial screen | 同时报已有 235B mixed success和内部 pairwise failure；后续 expansion 由预注册 verdict 触发 | | Simplified workload | NEEDS EVIDENCE | fixed-shape 只用于 phase isolation，不外推 trace-faithful mixed | | Calibration=evaluation | PASS | 新 case 不用 serving E2E 数据拟合 scale | -| Missing significance | NEEDS EVIDENCE until run | boundary anchors做独立 fresh-server repeat，保留 disagreement | +| Missing significance | PASS for ranking screen | 96 个 config-rate cells 均做两个独立 fresh-server rounds；两轮都 pass 才算 feasible | | Relative-only result | PASS by design | 同时报 req/s/GPU、TTFT distribution、rank/regret | ## 复现信息 @@ -43,12 +44,13 @@ - **Code：** AITuner branch `codex/fidelity-prefix-pilot-20260714`；Frontier upstream `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + frozen known patches。 - **Environment：** 只使用 dash0 8×H20；Qwen30 venv `/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1`；model `/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B`。 - **产物路径：** local/remote `runs/frontier-phase-factorial-v0/`；raw GPU artifacts 由 fleet harvest，condensed JSON/CSV 进入结果目录。 -- **已知 deviation：** 235B 为 FP8/vLLM0.10.2/FlashInfer eager，30B 为 BF16/vLLM0.20/FA3/default CUDA graph；因此跨模型只检验 hypothesis consistency，causal phase claim 最终仍需 same-model phase pair。 +- **已知 deviation：** 235B 为 FP8/vLLM0.10.2/FlashInfer eager，30B 为 BF16/vLLM0.20/FA3/default CUDA graph；因此跨模型只检验 hypothesis consistency，causal phase claim 最终仍需 same-model phase pair。初始 continuous fleet monitor 未在 fresh-server 间隙保留 controller-level GPU reservation，产生重叠 launch；该 attempt 整体移动到 `invalid-overlap-*`，不进入统计。后续一次试图在“其余 4 卡”上并行 refinement 时，探针再次命中 TP4 fresh-server 空窗，把 4 个 TP1 jobs 放到了同一 GPU set。这四个 jobs 尚未产生 measured result；但当时 TP4/MNS64 已产生的单个 anchor 也按污染处理。两者都整体移到 `invalid-overlap-20260716T1750Z`，TP4/MNS64 从空目录重跑。此后的 barrier waves 不在运行中追加 job；但 Wave 3 的 harvest monitor 在未及时返回 launch 状态时已发射 MNS16/32，紧接的 monitor retry 又在它们的启动空窗发射 MNS64。三者都未产生 measured result，整体移到 `invalid-overlap-20260716T1836Z`。最终 TP4 waves 使用只含本波 jobs 的独立 queue state，不再依赖 pending-job 探针调度。 ## 结果 -- **观察事实：** 235B fixed-shape mixed 已完成：real/sim TP4 top set exact match，worst regret=0、τ-b=0.8944；但 20 个 real non-tie pairs 只保持 16 个，10/34 anchors false-infeasible，TP8 MNS×MBT interaction 被漏掉。30B prefill-only 待运行。 -- **异常：** 无。 -- **Interpretation 与剩余 alternatives：** 强版本“mixed 必然失败”已被 235B top-set result 削弱；仍可能存在 phase-dependent error magnitude，由 topology margin 掩盖。 -- **Claim update：** unchanged，等待 30B prefill-only。 -- **下一步：** freeze 30B simulator surface → guided real anchors → joint verdict；只有判定需要时扩展 decode-heavy 235B 或 trace-shaped 30B prefill。 +- **观察事实：** 235B fixed-shape mixed 中 real/sim 的四个 TP4 top configs 完全一致，worst regret=0、τ-b=0.8944；但 20 个 real non-tie pairs 只保持 16 个，10/34 anchors false-infeasible，TP8 MNS×MBT interaction 被漏掉。30B prefill-only 中真机 capacity/GPU 为 TP1=7、TP2=7、TP4=8，Frontier 为 TP1=8、TP2=8、TP4=6；real top set 是四个 TP4 configs，simulator top set 是全部八个 TP1/TP2 configs，无交集。worst regret=12.5%、τ-b=-1.0，32 个 real non-tie pairs 中 0 个同序。96 个 anchor labels 中有 8 个 false-feasible 和 8 个 false-infeasible。 +- **实验成本：** 接受 24 个 fleet jobs、192 个 fresh-server anchors、12,288 个 measured requests 和 4,512 个 warmups，消耗 12.0744 H20-GPU-hours。 +- **异常与排除：** fleet controller 在 fresh-server 空窗期没有保留 GPU reservation，产生了三批重叠 launch。污染 attempt 不进入 accepted artifact root，未产生 measured result 的重叠 jobs 也不被计数；同一波中已产生的 TP4/MNS64 单 anchor 同样按污染丢弃，从空目录重跑。最终 TP4 refinement 使用彼此独立的 queue states。analyzer 按 `(round, filename)` 去重相同 harvest copy，如果 hash 冲突则直接报错。 +- **Interpretation 与剩余 alternatives：** `H-phase` 的强形式（prefill-only 是 fidelity 充分条件）被否证；`H-margin` 与数据更一致。235B 的真机 TP4/TP8 最优 margin 为 2×，足以掩盖内部 residual；30B 的 8-vs-7 margin 被 TP-dependent saturation residual 穿过。但这是跨 stack comparison，不能把差异因果归结为 model size。 +- **Claim update：** “prefill-only 容易，decode/mixed 困难”的强假设被否证。新的可证伪命题是：config-ranking fidelity 取决于 scheduler-state-conditioned action residual 是否大于 real decision margin。 +- **下一步：** 不继续扩展跨模型 phase cases。在 Qwen30 prefill-only 上依次做 measured-collective injection、batch-composition-conditioned pure-prefill attention/step profile、TP1@8/TP2@16/TP4@32 的 scheduler batch/queue/per-step trace 对齐，最后再测 routing/graph。 diff --git a/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml b/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml new file mode 100644 index 0000000..5553f93 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-base-rerun.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-base-rerun" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-exclusive.toml b/runs/frontier-phase-factorial-v0/fleet-exclusive.toml new file mode 100644 index 0000000..7de87b2 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-exclusive.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-exclusive" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml new file mode 100644 index 0000000..a7e6466 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave3.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine-tp4-wave3" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml new file mode 100644 index 0000000..3cba17e --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine-tp4-wave4.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine-tp4-wave4" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/fleet-refine.toml b/runs/frontier-phase-factorial-v0/fleet-refine.toml new file mode 100644 index 0000000..786d090 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/fleet-refine.toml @@ -0,0 +1,24 @@ +version = 1 + +[paths] +state_dir = "runs/frontier-phase-factorial-v0/fleet-state-refine" +artifacts_dir = "runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive" + +[ssh] +connect_timeout_sec = 10 + +[scheduler] +gpu_free_memory_mb = 95000 +gpu_free_utilization_pct = 5 +prefer_pack = true + +[sync] +mode = "scp" +local_path = "runs/frontier-phase-factorial-v0/remote-sync-marker" + +[[hosts]] +name = "dash0" +ssh_alias = "dash0" +enabled = true +sync_remote_path = "/home/admin/cpfs/wjh/aituner/phase-factorial-sync-marker" +fleet_root = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0" diff --git a/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml b/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml new file mode 100644 index 0000000..2eb2f2d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_base_rerun.toml @@ -0,0 +1,21 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "4 8 16 32 64" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_full.toml b/runs/frontier-phase-factorial-v0/jobs_full.toml index c67f255..05d7621 100644 --- a/runs/frontier-phase-factorial-v0/jobs_full.toml +++ b/runs/frontier-phase-factorial-v0/jobs_full.toml @@ -1,7 +1,7 @@ version = 1 [[jobs]] -name = "qwen30-prefill-real-tp1-mns8-20260717-v1" +name = "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive" gpus = 1 gpu_model = "H20" hosts = ["dash0"] @@ -21,7 +21,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp1-mns16-20260717-v1" +name = "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive" gpus = 1 gpu_model = "H20" hosts = ["dash0"] @@ -41,7 +41,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp1-mns32-20260717-v1" +name = "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive" gpus = 1 gpu_model = "H20" hosts = ["dash0"] @@ -61,7 +61,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp1-mns64-20260717-v1" +name = "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive" gpus = 1 gpu_model = "H20" hosts = ["dash0"] @@ -81,7 +81,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp2-mns8-20260717-v1" +name = "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive" gpus = 2 gpu_model = "H20" hosts = ["dash0"] @@ -101,7 +101,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp2-mns16-20260717-v1" +name = "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive" gpus = 2 gpu_model = "H20" hosts = ["dash0"] @@ -121,7 +121,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp2-mns32-20260717-v1" +name = "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive" gpus = 2 gpu_model = "H20" hosts = ["dash0"] @@ -141,7 +141,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp2-mns64-20260717-v1" +name = "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive" gpus = 2 gpu_model = "H20" hosts = ["dash0"] @@ -161,7 +161,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp4-mns8-20260717-v1" +name = "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive" gpus = 4 gpu_model = "H20" hosts = ["dash0"] @@ -181,7 +181,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp4-mns16-20260717-v1" +name = "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive" gpus = 4 gpu_model = "H20" hosts = ["dash0"] @@ -201,7 +201,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp4-mns32-20260717-v1" +name = "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive" gpus = 4 gpu_model = "H20" hosts = ["dash0"] @@ -221,7 +221,7 @@ VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" [[jobs]] -name = "qwen30-prefill-real-tp4-mns64-20260717-v1" +name = "qwen30-prefill-real-tp4-mns64-20260717-v2-exclusive" gpus = 4 gpu_model = "H20" hosts = ["dash0"] diff --git a/runs/frontier-phase-factorial-v0/jobs_refine.toml b/runs/frontier-phase-factorial-v0/jobs_refine.toml new file mode 100644 index 0000000..55fd01d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine.toml @@ -0,0 +1,240 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns8-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "8" +RATES = "5 6 7" +SERVER_PORT = "8720" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" +[[jobs]] +name = "qwen30-prefill-real-tp1-mns16-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "16" +RATES = "5 6 7" +SERVER_PORT = "8721" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns32-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "32" +RATES = "5 6 7" +SERVER_PORT = "8722" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp1-mns64-20260717-v3-refine" +gpus = 1 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp1-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "1" +MNS = "64" +RATES = "5 6 7" +SERVER_PORT = "8723" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp1-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns8-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "8" +RATES = "10 12 14" +SERVER_PORT = "8724" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns16-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "16" +RATES = "10 12 14" +SERVER_PORT = "8725" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns32-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "32" +RATES = "10 12 14" +SERVER_PORT = "8726" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp2-mns64-20260717-v3-refine" +gpus = 2 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp2-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "2" +MNS = "64" +RATES = "10 12 14" +SERVER_PORT = "8727" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp2-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns8-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns8-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "8" +RATES = "20 24 28" +SERVER_PORT = "8728" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns8-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns16-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "16" +RATES = "20 24 28" +SERVER_PORT = "8729" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns32-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "32" +RATES = "20 24 28" +SERVER_PORT = "8730" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v3-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "20 24 28" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml new file mode 100644 index 0000000..b830690 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave3.toml @@ -0,0 +1,41 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns16-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns16-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "16" +RATES = "20 24 28" +SERVER_PORT = "8729" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns16-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns32-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns32-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "32" +RATES = "20 24 28" +SERVER_PORT = "8730" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns32-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml new file mode 100644 index 0000000..b69872d --- /dev/null +++ b/runs/frontier-phase-factorial-v0/jobs_refine_tp4_wave4.toml @@ -0,0 +1,21 @@ +version = 1 + +[[jobs]] +name = "qwen30-prefill-real-tp4-mns64-20260717-v4-refine" +gpus = 4 +gpu_model = "H20" +hosts = ["dash0"] +command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-phase-factorial-v0 && timeout --signal=TERM --kill-after=30s 7200 bash run_qwen30_prefill_real_config.sh" +artifacts = ["artifacts/real-tp4-mns64-v1"] + +[jobs.env] +HOME = "/tmp/wjh" +XDG_CACHE_HOME = "/tmp/wjh/.cache" +VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm" +TP = "4" +MNS = "64" +RATES = "20 24 28" +SERVER_PORT = "8731" +OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/gpu-fleet-phase-factorial-v0/artifacts/real-tp4-mns64-v1" +VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1" +MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B" diff --git a/runs/frontier-phase-factorial-v0/results/final/capacity.csv b/runs/frontier-phase-factorial-v0/results/final/capacity.csv new file mode 100644 index 0000000..de7a968 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/results/final/capacity.csv @@ -0,0 +1,13 @@ +config,tp,mns,real,simulator +tp1_mns8,1,8,7.0,8.0 +tp1_mns16,1,16,7.0,8.0 +tp1_mns32,1,32,7.0,8.0 +tp1_mns64,1,64,7.0,8.0 +tp2_mns8,2,8,7.0,8.0 +tp2_mns16,2,16,7.0,8.0 +tp2_mns32,2,32,7.0,8.0 +tp2_mns64,2,64,7.0,8.0 +tp4_mns8,4,8,8.0,6.0 +tp4_mns16,4,16,8.0,6.0 +tp4_mns32,4,32,8.0,6.0 +tp4_mns64,4,64,8.0,6.0 diff --git a/runs/frontier-phase-factorial-v0/results/final/comparison.json b/runs/frontier-phase-factorial-v0/results/final/comparison.json new file mode 100644 index 0000000..db46764 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/results/final/comparison.json @@ -0,0 +1,5860 @@ +{ + "contract": { + "input_tokens": 2048, + "model": "Qwen3-30B-A3B", + "output_tokens": 1, + "prefix_caching": false, + "real_anchor_merge": "both_fresh_server_rounds_must_pass", + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "metrics": { + "anchor_confusion": { + "real_fail_sim_fail": 24, + "real_fail_sim_pass": 8, + "real_pass_sim_fail": 8, + "real_pass_sim_pass": 56 + }, + "config_order": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64", + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "kendall": { + "both_ties": 34, + "concordant": 0, + "discordant": 32, + "kendall_tau_b": -1.0, + "real_only_ties": 0, + "simulator_only_ties": 0 + }, + "pairwise_non_tied": { + "all": { + "accuracy": 0.0, + "comparable": 32, + "correct": 0 + }, + "within_tp": {} + }, + "real_best_capacity_per_gpu": 8.0, + "real_top_set": [ + "tp4_mns8", + "tp4_mns16", + "tp4_mns32", + "tp4_mns64" + ], + "simulator_best_capacity_per_gpu": 8.0, + "simulator_top_set": [ + "tp1_mns8", + "tp1_mns16", + "tp1_mns32", + "tp1_mns64", + "tp2_mns8", + "tp2_mns16", + "tp2_mns32", + "tp2_mns64" + ], + "top1_regret_best": 0.125, + "top1_regret_worst": 0.125, + "top_set_exact_match": false, + "top_set_overlap": [] + }, + "objective": "maximum_tested_slo_feasible_offered_request_rate_per_gpu", + "real": { + "tp1_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.15452707000077, + 152.64549700077623 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r4p00.json", + "sha256": "d82e386ec3effdf57c148171311928298405da0e4f896584bf05db2485bb6bd4", + "summary": { + "admission_lag_max_ms": 0.11588609777390957, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.6234939750284, + "ttft_p50_ms": 151.05855697765946, + "ttft_p95_ms": 153.15452707000077 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r4p00.json", + "sha256": "1565be89131e3c8a9c352c6f29694ef2793734b0623a2db1cbea7e0b0e8a2bf6", + "summary": { + "admission_lag_max_ms": 0.11451996397227049, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.18561100866646, + "ttft_p50_ms": 151.18564700242132, + "ttft_p95_ms": 152.64549700077623 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.8308589477092, + 152.43642206769437 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r5p00.json", + "sha256": "e4447d94a326c541412b40157cfe283a6350680e3285b5d10fed6b8de5c0cd93", + "summary": { + "admission_lag_max_ms": 0.17109804321080446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.35138496104628, + "ttft_p50_ms": 150.91424097772688, + "ttft_p95_ms": 152.8308589477092 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r5p00.json", + "sha256": "572c17e309c6c9f23c490979e7acb885475f142bc530ecfee88916b048259ea2", + "summary": { + "admission_lag_max_ms": 0.1321269664913416, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.11930498946458, + "ttft_p50_ms": 150.50161199178547, + "ttft_p95_ms": 152.43642206769437 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.56597595382482, + 152.16859406791627 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r6p00.json", + "sha256": "68c8b39154d46f1ceda17f47e8cbd84a8c8a9f088c5e82d8f0ac25a7180dc14c", + "summary": { + "admission_lag_max_ms": 0.11641765013337135, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.50876995641738, + "ttft_p50_ms": 150.41159105021507, + "ttft_p95_ms": 153.56597595382482 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r6p00.json", + "sha256": "a970a2cdf7034026a56af3ea4da59856fd588da35d66ef4f9e15481a892edcc9", + "summary": { + "admission_lag_max_ms": 0.11551892384886742, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 168.65782206878066, + "ttft_p50_ms": 150.03317291848361, + "ttft_p95_ms": 152.16859406791627 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 300.90478900820017, + 293.0445579113439 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r7p00.json", + "sha256": "1aa74c18adbff53b36ae8a6f07b5b0865f6b68fb3e7862a70b36f67c18f0bba0", + "summary": { + "admission_lag_max_ms": 0.10505830869078636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.15459000505507, + "ttft_p50_ms": 263.44375393819064, + "ttft_p95_ms": 300.90478900820017 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r7p00.json", + "sha256": "e14bc9dba35d453862839eeed77262a1ed9b8394404ca90c60ec8cb26a1aa73b", + "summary": { + "admission_lag_max_ms": 0.10963028762489557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 295.1545650139451, + "ttft_p50_ms": 243.97043196950108, + "ttft_p95_ms": 293.0445579113439 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1323.4397460473701, + 1315.5151740647852 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r8p00.json", + "sha256": "9aeee7453ffe5098d203e02b84f5009ed347c478b736e89297bd3695dd81d477", + "summary": { + "admission_lag_max_ms": 0.1251589274033904, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1477.699184906669, + "ttft_p50_ms": 776.2576789828017, + "ttft_p95_ms": 1323.4397460473701 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r8p00.json", + "sha256": "7630f4901cf4ec4ea955738b119a1c32aba53f9ca1edb8819556fc60ae1fda09", + "summary": { + "admission_lag_max_ms": 0.11316593736410141, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1404.566578916274, + "ttft_p50_ms": 761.7035870207474, + "ttft_p95_ms": 1315.5151740647852 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4892.253074911423, + 4893.758205929771 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r16p00.json", + "sha256": "d3a2a85cf06a4f080edc9a3809f05493ac55b0aa1d890806ffbc551c8eca14f8", + "summary": { + "admission_lag_max_ms": 0.39830803871154785, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.215195022523, + "ttft_p50_ms": 2742.7029330283403, + "ttft_p95_ms": 4892.253074911423 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r16p00.json", + "sha256": "67110284c38a9a60228f00b1d2b0de9477d6dec660f1c91eb4f39e1b79e272a5", + "summary": { + "admission_lag_max_ms": 0.11306500528007746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.818040987477, + "ttft_p50_ms": 2742.512951954268, + "ttft_p95_ms": 4893.758205929771 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6650.756464921869, + 6751.845097984187 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r32p00.json", + "sha256": "affc62b27382400497bb82e9da31ed9891b60155318fc4f250dca0704ccc5e02", + "summary": { + "admission_lag_max_ms": 0.11010200250893831, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6920.621327939443, + "ttft_p50_ms": 3650.839229929261, + "ttft_p95_ms": 6650.756464921869 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r32p00.json", + "sha256": "df0c8d703b023692814be41f47da4db4ce1b7eb0c3423a742c10884b6447583f", + "summary": { + "admission_lag_max_ms": 0.1288279891014099, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6897.979240980931, + "ttft_p50_ms": 3730.34580796957, + "ttft_p95_ms": 6751.845097984187 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7732.2855240199715, + 7744.056030060165 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round1/results/r64p00.json", + "sha256": "fbfad78129c843c85905c3f94f278f468a0a019c0cdf0b62140d0fea8f0aaffc", + "summary": { + "admission_lag_max_ms": 0.11090096086263657, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7836.245508980937, + "ttft_p50_ms": 4328.587644966319, + "ttft_p95_ms": 7732.2855240199715 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z/artifacts/artifacts/real-tp1-mns16-v1/round2/results/r64p00.json", + "sha256": "0f005c642327708e1ceb058f71f806b4d46ea666f45695605552e8f234521608", + "summary": { + "admission_lag_max_ms": 0.10862003546208143, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7847.871358972043, + "ttft_p50_ms": 4332.9262170009315, + "ttft_p95_ms": 7744.056030060165 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp1_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.44141807872802, + 152.2580359596759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r4p00.json", + "sha256": "211fa276e4a99a061126882085b9850549b453520ac3414dd2d332c89e52b431", + "summary": { + "admission_lag_max_ms": 0.09491504170000553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 169.56262104213238, + "ttft_p50_ms": 150.8563900133595, + "ttft_p95_ms": 153.44141807872802 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r4p00.json", + "sha256": "17428902a5b20e9084f20792963f0620a1af06cdc1afb76ee99fd15dd905b89e", + "summary": { + "admission_lag_max_ms": 0.12033001985400915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.99700601119548, + "ttft_p50_ms": 150.5822929320857, + "ttft_p95_ms": 152.2580359596759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.05012692604214, + 152.00478909537196 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r5p00.json", + "sha256": "b0eb19ff7ab8a26e34e91f213a983e58689c63071285e27b3719bda0a765ff32", + "summary": { + "admission_lag_max_ms": 0.11265103239566088, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.48999004345387, + "ttft_p50_ms": 150.21017810795456, + "ttft_p95_ms": 152.05012692604214 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r5p00.json", + "sha256": "3b857dff42dd16ebee95356ceacd6b58ff7bd0a9b91f3d6208d489b84985697f", + "summary": { + "admission_lag_max_ms": 0.11149293277412653, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.33813393954188, + "ttft_p50_ms": 150.45341395307332, + "ttft_p95_ms": 152.00478909537196 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.74227899499238, + 152.82704099081457 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r6p00.json", + "sha256": "7615fa664a99f02f9f6d4f1fb0f2d6cba11842772c086634b939bc08cd0f2f4d", + "summary": { + "admission_lag_max_ms": 0.1096969936043024, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 154.2669020127505, + "ttft_p50_ms": 150.6725640501827, + "ttft_p95_ms": 152.74227899499238 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r6p00.json", + "sha256": "345e3cd3f8cd4d7c7a5e9343f937dd9b1b5928500133830f3046b743da6df64e", + "summary": { + "admission_lag_max_ms": 0.12192933354526758, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.12304301187396, + "ttft_p50_ms": 150.68741200957447, + "ttft_p95_ms": 152.82704099081457 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 302.5522669777274, + 294.5978350471705 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r7p00.json", + "sha256": "3f54a3e3f2279a5325a173ff714d600fb1a2cdfcfba1e8e5c980f7b21b99be59", + "summary": { + "admission_lag_max_ms": 0.11122168507426977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 304.24335307907313, + "ttft_p50_ms": 262.4604320153594, + "ttft_p95_ms": 302.5522669777274 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r7p00.json", + "sha256": "0832169cdf1bef49c5006464c6cbcd50f4567e0ab90bec98d722f64f7b476763", + "summary": { + "admission_lag_max_ms": 0.1059951027855277, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 297.1431679325178, + "ttft_p50_ms": 249.51713997870684, + "ttft_p95_ms": 294.5978350471705 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1321.9368209829554, + 1318.7565059633926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r8p00.json", + "sha256": "a0f5900ac9d2c47dcc7d7ac08d3f3d23e1d6f529fdd0200989afa819dd3df055", + "summary": { + "admission_lag_max_ms": 0.17414195463061333, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.90625, + "passed": 58, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1475.8616799954325, + "ttft_p50_ms": 775.7647309917957, + "ttft_p95_ms": 1321.9368209829554 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r8p00.json", + "sha256": "c0fc3c85095ddacc66849e6f58be0851abab6d4022ee4eb7120c8cb8ea9e5d2c", + "summary": { + "admission_lag_max_ms": 2.749277977272868, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1407.0930189918727, + "ttft_p50_ms": 763.8335369993001, + "ttft_p95_ms": 1318.7565059633926 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4893.3289990527555, + 4893.17379205022 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r16p00.json", + "sha256": "4e9ccaf5384f622b1da7c58d3de803c86099dd9deae91e363aabccc14aeabc05", + "summary": { + "admission_lag_max_ms": 0.11893699411302805, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4972.85822511185, + "ttft_p50_ms": 2742.8449960425496, + "ttft_p95_ms": 4893.3289990527555 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r16p00.json", + "sha256": "c207e415186aed1148e16085145bcdfd189314e7b92f74ea72fb2e4972ef4eae", + "summary": { + "admission_lag_max_ms": 0.09541702456772327, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.2590809213, + "ttft_p50_ms": 2740.232800017111, + "ttft_p95_ms": 4893.17379205022 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6653.632597066462, + 6647.155530983582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r32p00.json", + "sha256": "780f563a2739d892761905031f82ca8bd0ac4f4eb485c454390dcb2bcdfc1b81", + "summary": { + "admission_lag_max_ms": 0.105265062302351, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6922.717701992951, + "ttft_p50_ms": 3652.452490059659, + "ttft_p95_ms": 6653.632597066462 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r32p00.json", + "sha256": "2a5eb8fb37f03e759de149d24f9b23c812a26afe72ffaaffa030688904634e43", + "summary": { + "admission_lag_max_ms": 0.10375294368714094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6917.111590038985, + "ttft_p50_ms": 3649.0042679943144, + "ttft_p95_ms": 6647.155530983582 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7747.191094094887, + 7839.257381972857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round1/results/r64p00.json", + "sha256": "b94a39a91842ab39d545c473fc4f872ec79eb1888943c3e03ab03c18961a9105", + "summary": { + "admission_lag_max_ms": 0.6591080455109477, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7851.370444986969, + "ttft_p50_ms": 4335.578468977474, + "ttft_p95_ms": 7747.191094094887 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z/artifacts/artifacts/real-tp1-mns32-v1/round2/results/r64p00.json", + "sha256": "69ab47305c4bb14f3b4ed5bc0b18a37b8a5e7c1cdbee9232e6d2b9616ed10f84", + "summary": { + "admission_lag_max_ms": 0.10753795504570007, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.125, + "passed": 8, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7885.349316056818, + "ttft_p50_ms": 4009.1016669757664, + "ttft_p95_ms": 7839.257381972857 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp1_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 156.76817402709275, + 152.9719429090619 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r4p00.json", + "sha256": "df798dd7a1cde3cfaf698286f44bbac2b0730a2c806f125268bdf7226e2bb47c", + "summary": { + "admission_lag_max_ms": 0.1147780567407608, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 162.9874319769442, + "ttft_p50_ms": 152.67050196416676, + "ttft_p95_ms": 156.76817402709275 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r4p00.json", + "sha256": "df80057224d69ce2a79f937aa45c2093d66d131765f3c7bfcc2289a9aaff3d6a", + "summary": { + "admission_lag_max_ms": 0.2131829969584942, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 153.917470946908, + "ttft_p50_ms": 150.96506499685347, + "ttft_p95_ms": 152.9719429090619 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.4973110537976, + 155.23677493911237 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r5p00.json", + "sha256": "e614ffaf879859407fc6420847f0140a81bda9a94fcc438fb6782c9019bf5378", + "summary": { + "admission_lag_max_ms": 0.1283780438825488, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.08639602921903, + "ttft_p50_ms": 152.1085740532726, + "ttft_p95_ms": 153.4973110537976 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r5p00.json", + "sha256": "13e825eb5ced3d0cb2b5dd68d9243fdf1cee8779b90e9aad77be04053be22be2", + "summary": { + "admission_lag_max_ms": 0.1683390000835061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.6691420264542, + "ttft_p50_ms": 152.42570801638067, + "ttft_p95_ms": 155.23677493911237 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 152.53963600844145, + 152.85827894695103 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r6p00.json", + "sha256": "ac530f98f57ec452538ebe2311291d0a7c83bbd5554e82f53e78a2311802e411", + "summary": { + "admission_lag_max_ms": 0.13490929268300533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 157.18225098680705, + "ttft_p50_ms": 150.29372205026448, + "ttft_p95_ms": 152.53963600844145 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r6p00.json", + "sha256": "2f1445ee0890b0972252d10180ef6f00cbad45ee73bd174f21b370cdf950b654", + "summary": { + "admission_lag_max_ms": 0.16514293383806944, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 158.88926689513028, + "ttft_p50_ms": 150.2534201135859, + "ttft_p95_ms": 152.85827894695103 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 346.96662798523903, + 295.1050230767578 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r7p00.json", + "sha256": "c4765794d442daf7eeb64a2f587792701ebe7659c7702a3bbbfc8900801fff0e", + "summary": { + "admission_lag_max_ms": 0.16206549480557442, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 354.881492909044, + "ttft_p50_ms": 310.15722593292594, + "ttft_p95_ms": 346.96662798523903 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r7p00.json", + "sha256": "30addb463839e23a7356a5cf6d4060ec1eebfa2b9564fa9500be5bcb7aecb537", + "summary": { + "admission_lag_max_ms": 0.10446517262607813, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 296.6241450048983, + "ttft_p50_ms": 246.06937903445214, + "ttft_p95_ms": 295.1050230767578 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1376.107804942876, + 1321.1636180058122 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r8p00.json", + "sha256": "89e2975f9c7319e477789ea77015541d5235efa61d90fb8bd5b0420a58f7f186", + "summary": { + "admission_lag_max_ms": 0.1416490413248539, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.890625, + "passed": 57, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1501.330335973762, + "ttft_p50_ms": 802.2130889585242, + "ttft_p95_ms": 1376.107804942876 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r8p00.json", + "sha256": "9ca6935f1320033c3ec4b72a35c5d9909ccd7b0fff2e4b2c92c6a26f52887df1", + "summary": { + "admission_lag_max_ms": 0.12422900181263685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.921875, + "passed": 59, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1410.047525074333, + "ttft_p50_ms": 766.8001820566133, + "ttft_p95_ms": 1321.1636180058122 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4950.155022088438, + 4892.750842031091 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r16p00.json", + "sha256": "4fce2749435bf2e8c77999e11be0a788cd85f459a647760f3db34f1dbc0aee85", + "summary": { + "admission_lag_max_ms": 0.10520406067371368, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.171875, + "passed": 11, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 5031.207721913233, + "ttft_p50_ms": 2775.8131299633533, + "ttft_p95_ms": 4950.155022088438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r16p00.json", + "sha256": "33ded0192d8986dbcf0a7f5ab6329a440994243d4e528d6c1b278bb6b92720fd", + "summary": { + "admission_lag_max_ms": 0.14005706179887056, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4971.664474112913, + "ttft_p50_ms": 2743.070787983015, + "ttft_p95_ms": 4892.750842031091 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6750.775643973611, + 6754.7571890754625 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r32p00.json", + "sha256": "44030c4d3374cecebe67b886b428ba8014c87d941c00a178d6049bc57c809f46", + "summary": { + "admission_lag_max_ms": 0.7358170114457607, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6899.517056066543, + "ttft_p50_ms": 3728.6347220651805, + "ttft_p95_ms": 6750.775643973611 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r32p00.json", + "sha256": "f397bad36552c6df5bf5bcee702cfa968ddf3e20b20e9976554ddfbe838ef3b2", + "summary": { + "admission_lag_max_ms": 0.7476879982277751, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6901.606322033331, + "ttft_p50_ms": 3729.3128239689395, + "ttft_p95_ms": 6754.7571890754625 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7800.616211956367, + 7748.703246936202 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round1/results/r64p00.json", + "sha256": "61eadc2af60386d6972d69fe7b083874e705864162073d54d1bde2bc0e6de7e7", + "summary": { + "admission_lag_max_ms": 0.11387700214982033, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7905.77321103774, + "ttft_p50_ms": 4366.223473916762, + "ttft_p95_ms": 7800.616211956367 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z/artifacts/artifacts/real-tp1-mns64-v1/round2/results/r64p00.json", + "sha256": "db02e9a814154a9259cbe3fefdf15700923458f7e1208596e861b69b6a6ddd41", + "summary": { + "admission_lag_max_ms": 0.14900392852723598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7852.748095989227, + "ttft_p50_ms": 4337.09563605953, + "ttft_p95_ms": 7748.703246936202 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp1_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.07429495826364, + 153.62840006127954 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r4p00.json", + "sha256": "23d5ab491b70c02f51c2f06dec0407edc0627819a33b9bcef1349d974a4e20fe", + "summary": { + "admission_lag_max_ms": 0.14362391084432602, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 171.46877304185182, + "ttft_p50_ms": 151.62539994344115, + "ttft_p95_ms": 154.07429495826364 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r4p00.json", + "sha256": "563aa230e0f5e420cf453dfd9b829d7df493919192d638cbe76f85eb77b6d2e7", + "summary": { + "admission_lag_max_ms": 0.11323392391204834, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.62100906390697, + "ttft_p50_ms": 151.69239102397114, + "ttft_p95_ms": 153.62840006127954 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 5.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 154.20477103907615, + 153.81942794192582 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r5p00.json", + "sha256": "6f2a04a5c11eed5826e2a6e1ce6e6f8d98504b31d0fec30d9440df2cdb62e56d", + "summary": { + "admission_lag_max_ms": 0.10794296395033598, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.37187200970948, + "ttft_p50_ms": 152.59731002151966, + "ttft_p95_ms": 154.20477103907615 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r5p00.json", + "sha256": "27a0bfa65a66e5cc77e10a5cf39a846f0fd1dd266c10926dae9db7a80c581d6e", + "summary": { + "admission_lag_max_ms": 0.1204309519380331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 155.40366806089878, + "ttft_p50_ms": 152.0398510619998, + "ttft_p95_ms": 153.81942794192582 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 6.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 153.00951700191945, + 153.5524509381503 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r6p00.json", + "sha256": "4562fae05d510cd50f2c9d7479916d056c8ac4d04851d2388b98f8410192c23a", + "summary": { + "admission_lag_max_ms": 0.10347133502364159, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 156.3970319693908, + "ttft_p50_ms": 150.414744974114, + "ttft_p95_ms": 153.00951700191945 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r6p00.json", + "sha256": "bd13283e2807150396b278f46ff630ae9820895e0c054a9efcfd8fc213947d47", + "summary": { + "admission_lag_max_ms": 0.13600767124444246, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 159.63539096992463, + "ttft_p50_ms": 151.83425601571798, + "ttft_p95_ms": 153.5524509381503 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 7.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 336.8800369789824, + 329.65395506471395 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r7p00.json", + "sha256": "fd5afe52238d6ddb72bdf7f7b142fddbeb3f62e160e4a1b3a639b308717964e6", + "summary": { + "admission_lag_max_ms": 0.14346837997436523, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 341.1952300230041, + "ttft_p50_ms": 294.81438896618783, + "ttft_p95_ms": 336.8800369789824 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r7p00.json", + "sha256": "b0fdc1ef617313261c490010adeb2f4ee93bbb053f1db10734ed5a63fd37fa76", + "summary": { + "admission_lag_max_ms": 0.11047185398638248, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 334.06115300022066, + "ttft_p50_ms": 281.1476809438318, + "ttft_p95_ms": 329.65395506471395 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 8.0, + "round_feasible": [ + false, + true + ], + "round_ttft_p95_ms": [ + 1289.2397560644895, + 1254.356418037787 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r8p00.json", + "sha256": "8877ed44b90f421afd22ec110fe6337581c67cd973b44378bcb7528fc8624de3", + "summary": { + "admission_lag_max_ms": 0.09899691212922335, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.9375, + "passed": 60, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1454.3529660440981, + "ttft_p50_ms": 813.0663179326802, + "ttft_p95_ms": 1289.2397560644895 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r8p00.json", + "sha256": "7ab24f5ab33bf166589af7e2f5c2a274ceef6a64f0029a96ec1879b42332123d", + "summary": { + "admission_lag_max_ms": 0.11694896966218948, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 0.953125, + "passed": 61, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1347.0055809011683, + "ttft_p50_ms": 744.0344559727237, + "ttft_p95_ms": 1254.356418037787 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4903.538785991259, + 4900.28300601989 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r16p00.json", + "sha256": "363d9b43164b5451bc3dd668caa0ce141d15114087cf444e7fd79d12b7563fd6", + "summary": { + "admission_lag_max_ms": 0.11420203372836113, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4982.48829995282, + "ttft_p50_ms": 2749.9175920384005, + "ttft_p95_ms": 4903.538785991259 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r16p00.json", + "sha256": "8aec70dfd6b69f4903bccc02c5d5474165ed07d9dd010c860ff9b447429da950", + "summary": { + "admission_lag_max_ms": 0.19414699636399746, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.1875, + "passed": 12, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4979.007771937177, + "ttft_p50_ms": 2749.115635990165, + "ttft_p95_ms": 4900.28300601989 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 6760.187636013143, + 6534.733187989332 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r32p00.json", + "sha256": "7684ac0d2d6f851218b88fb37db3a5b54d877a6f1ead3bb604a6cb4d9a61cf3f", + "summary": { + "admission_lag_max_ms": 0.43534801807254553, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.15625, + "passed": 10, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6907.940509961918, + "ttft_p50_ms": 3735.540736000985, + "ttft_p95_ms": 6760.187636013143 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r32p00.json", + "sha256": "947974a9ae04e170b8c1aad32ddbda8c1000f3902a9b572c6338a7d0d4f23146", + "summary": { + "admission_lag_max_ms": 1.221106038428843, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.140625, + "passed": 9, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 6798.48151106853, + "ttft_p50_ms": 3595.78589303419, + "ttft_p95_ms": 6534.733187989332 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 7754.704045015387, + 7756.428897031583 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round1/results/r64p00.json", + "sha256": "abfc2a158087f0ad75b560dd420604a9f81359ffdf39f0ce8e2ea25b451e6904", + "summary": { + "admission_lag_max_ms": 0.13654399663209915, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7857.916672946885, + "ttft_p50_ms": 4341.527819051407, + "ttft_p95_ms": 7754.704045015387 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z/artifacts/artifacts/real-tp1-mns8-v1/round2/results/r64p00.json", + "sha256": "192c0c850a52d2b64780bc6321cbfd10faddcc0ff10741ddad283b7d789b9def", + "summary": { + "admission_lag_max_ms": 0.11000398080796003, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.109375, + "passed": 7, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 7859.458995983005, + "ttft_p50_ms": 4341.894763987511, + "ttft_p95_ms": 7756.428897031583 + } + } + ] + } + ], + "capacity": 7.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp1_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 99.1211449727416, + 101.08128399588168 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r4p00.json", + "sha256": "f149882da390114f2a2a140a3bd2ca6f15b6f7761591ea7b49b5a961c9d18df5", + "summary": { + "admission_lag_max_ms": 0.10085199028253555, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.28417506348342, + "ttft_p50_ms": 95.90165293775499, + "ttft_p95_ms": 99.1211449727416 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r4p00.json", + "sha256": "154dfaf2729001a8ef5b7430290f4ca906df1d4758e867d5b350489410f36d5c", + "summary": { + "admission_lag_max_ms": 0.11863000690937042, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 105.76263803523034, + "ttft_p50_ms": 96.27402503974736, + "ttft_p95_ms": 101.08128399588168 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.1850419882685, + 97.03529300168157 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r8p00.json", + "sha256": "9338d5cb994b0c12a6420f030ae2a9335a04986b4a81e31f4781ac838aeef2d4", + "summary": { + "admission_lag_max_ms": 0.12352492194622755, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.26434096693993, + "ttft_p50_ms": 95.07436107378453, + "ttft_p95_ms": 97.1850419882685 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r8p00.json", + "sha256": "a35b67970948976482ee835e595ee5631fa921607276d5ea7928460514f0d06b", + "summary": { + "admission_lag_max_ms": 0.11170096695423126, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.72453604172915, + "ttft_p50_ms": 94.03157397173345, + "ttft_p95_ms": 97.03529300168157 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.10817302018404, + 94.37024802900851 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r10p00.json", + "sha256": "f67e0f4f6b41861fcbcbee04b4efbf0dee46e273239ec94ef22d650ba6afa22a", + "summary": { + "admission_lag_max_ms": 0.10012194979935884, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.40214702207595, + "ttft_p50_ms": 94.17163301259279, + "ttft_p95_ms": 96.10817302018404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r10p00.json", + "sha256": "5a64b52f0298e9ecda2fddb8230cc27ab4b9c1db01575f68b5331816b324d1d7", + "summary": { + "admission_lag_max_ms": 0.10465399827808142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.07159405015409, + "ttft_p50_ms": 93.24954205658287, + "ttft_p95_ms": 94.37024802900851 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 447.0475659472868, + 454.8899739747867 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r12p00.json", + "sha256": "c6e27b9078281a54feeb1fab09616385a70de819791ce14f46b001bf540a80ab", + "summary": { + "admission_lag_max_ms": 0.12051803059875965, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 455.56937297806144, + "ttft_p50_ms": 280.09015799034387, + "ttft_p95_ms": 447.0475659472868 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r12p00.json", + "sha256": "861f9c4186c33e9ee6365d19c53e98c10615c5ee529a84e69f84dc292f53ac4f", + "summary": { + "admission_lag_max_ms": 0.11322635691612959, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 457.4526828946546, + "ttft_p50_ms": 290.1579710887745, + "ttft_p95_ms": 454.8899739747867 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1000.5082850111648, + 1004.6314439969137 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r14p00.json", + "sha256": "2dcaf1087edd1f12fbfe0618a6ab636b22484756a5729919bc0e51de153f8f4a", + "summary": { + "admission_lag_max_ms": 0.1881921198219061, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1089.2393708927557, + "ttft_p50_ms": 611.2175739835948, + "ttft_p95_ms": 1000.5082850111648 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r14p00.json", + "sha256": "16937b31f72298c7a50ec3c7b9c1b9c798f305d83469b3d3c058efc4d53135da", + "summary": { + "admission_lag_max_ms": 0.10659568943083286, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1096.6680150013417, + "ttft_p50_ms": 617.3943240428343, + "ttft_p95_ms": 1004.6314439969137 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1458.4544260287657, + 1471.3564389385283 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r16p00.json", + "sha256": "5576a7fd06977f51f1c43bb73c3c474cc21a93cd288b422ec3d09db671b3afe7", + "summary": { + "admission_lag_max_ms": 0.135264010168612, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1583.2410319708288, + "ttft_p50_ms": 904.5663280412555, + "ttft_p95_ms": 1458.4544260287657 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r16p00.json", + "sha256": "2ccf16426046c412829442f72932268d5b27ce051a4312d135d32b240b48ec7c", + "summary": { + "admission_lag_max_ms": 0.11741300113499165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.78125, + "passed": 50, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1537.9131329245865, + "ttft_p50_ms": 896.1611289996654, + "ttft_p95_ms": 1471.3564389385283 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3348.027400090359, + 3348.575382027775 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r32p00.json", + "sha256": "d258760d15f1fdf2adefca72f7a3505c0b839036799dcb074450f7f526f76718", + "summary": { + "admission_lag_max_ms": 0.18610397819429636, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3442.3487790627405, + "ttft_p50_ms": 1801.6119640087709, + "ttft_p95_ms": 3348.027400090359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r32p00.json", + "sha256": "edf5cccc1400639a94b5430897a04f20d2214bb1ffee7f4bfa62514150179a6a", + "summary": { + "admission_lag_max_ms": 0.1144439447671175, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3443.0311269825324, + "ttft_p50_ms": 1801.435066969134, + "ttft_p95_ms": 3348.575382027775 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4286.911493982188, + 4287.898648995906 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round1/results/r64p00.json", + "sha256": "2a6303efb826637c3c6448d6f766f7de4fc4d7b63db163a48f72c20df68394f4", + "summary": { + "admission_lag_max_ms": 0.10997895151376724, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.311317006126, + "ttft_p50_ms": 2396.4726959820837, + "ttft_p95_ms": 4286.911493982188 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z/artifacts/artifacts/real-tp2-mns16-v1/round2/results/r64p00.json", + "sha256": "6fafe3bf206583b8689be1b2b57d6b92cbd1ba54f7fb230ca05f6c68b3ab9757", + "summary": { + "admission_lag_max_ms": 0.12669397983700037, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4340.538130956702, + "ttft_p50_ms": 2394.112486974336, + "ttft_p95_ms": 4287.898648995906 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 16, + "name": "tp2_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.2925309324637, + 96.13083000294864 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r4p00.json", + "sha256": "812f8a108a9390fb759b46e5bd41e3ddad66bc144d47bf4024d834e42b47c47e", + "summary": { + "admission_lag_max_ms": 0.11386198457330465, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.76773096807301, + "ttft_p50_ms": 95.72782705072314, + "ttft_p95_ms": 97.2925309324637 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r4p00.json", + "sha256": "70f922eb7651d46d81d7d394b8453c5b778f921069cf48fb2ae2677afa94a842", + "summary": { + "admission_lag_max_ms": 0.30313804745674133, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 129.06646099872887, + "ttft_p50_ms": 94.53632694203407, + "ttft_p95_ms": 96.13083000294864 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 95.42673197574914, + 98.94769196398556 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r8p00.json", + "sha256": "24097d38b0ffb45f6cdbd93f0aad04dce3b7da41ff45e948eb3a794902b6547f", + "summary": { + "admission_lag_max_ms": 0.11571904178708792, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.04934400785714, + "ttft_p50_ms": 94.34181300457567, + "ttft_p95_ms": 95.42673197574914 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r8p00.json", + "sha256": "7251f076d34ba9484b08fa3fc4b898d2ba8d1199cb1eaeaaffc19930f8c889b9", + "summary": { + "admission_lag_max_ms": 0.1177559606730938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.93720292206854, + "ttft_p50_ms": 94.77824496570975, + "ttft_p95_ms": 98.94769196398556 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.28185909241438, + 95.9081610199064 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r10p00.json", + "sha256": "85796457715a25c15fc909f6def32776fbd49956eb7890df1c4abcef8c5afe78", + "summary": { + "admission_lag_max_ms": 0.14786410611122847, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.46025596279651, + "ttft_p50_ms": 93.54631893802434, + "ttft_p95_ms": 96.28185909241438 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r10p00.json", + "sha256": "0bf4d4d006bf0b5bd8b989ceeeb4bacb0e3ee1b8ea66f30b3deb120a14871cf0", + "summary": { + "admission_lag_max_ms": 0.11162098962813616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.54603894241154, + "ttft_p50_ms": 93.77408598084003, + "ttft_p95_ms": 95.9081610199064 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 439.0145930228755, + 429.9719400005415 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r12p00.json", + "sha256": "68de6e36bd7c5624cab107fe5cb21bb817711691e545439a958c926a27e3fe29", + "summary": { + "admission_lag_max_ms": 0.10475562885403633, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 445.90794993564487, + "ttft_p50_ms": 270.3119309153408, + "ttft_p95_ms": 439.0145930228755 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r12p00.json", + "sha256": "d8d3a44a07b46f1673817ece2948fd97c30f41bd451d93a2fa02edf074536c63", + "summary": { + "admission_lag_max_ms": 0.4323029424995184, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 434.6071509644389, + "ttft_p50_ms": 265.2625450864434, + "ttft_p95_ms": 429.9719400005415 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1014.2896320903674, + 999.8802930349484 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r14p00.json", + "sha256": "10196c2f9a9b69c2e47f64d06bf5ee6863af4cd2b612301622f0e304949a942a", + "summary": { + "admission_lag_max_ms": 0.09005970787256956, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1105.1907669752836, + "ttft_p50_ms": 621.6445650206879, + "ttft_p95_ms": 1014.2896320903674 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r14p00.json", + "sha256": "c10cab85588a3563bed36fc4480462635d34d240938beaa0e5034f13186578a6", + "summary": { + "admission_lag_max_ms": 0.11327012907713652, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1090.6395060010254, + "ttft_p50_ms": 623.5586369875818, + "ttft_p95_ms": 999.8802930349484 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1488.3801550604403, + 1484.6198010491207 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r16p00.json", + "sha256": "ef5046c49c32178855bed6a0488ea21e8da22f786ddaabc503bea2d1025bb2f3", + "summary": { + "admission_lag_max_ms": 0.7892940193414688, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1613.616122980602, + "ttft_p50_ms": 905.2404230460525, + "ttft_p95_ms": 1488.3801550604403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r16p00.json", + "sha256": "de02c1d49414cffe7e20023abf4d502a4efc6d2e01fcb3e49e928081aaef77b6", + "summary": { + "admission_lag_max_ms": 0.10548799764364958, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1550.9842830942944, + "ttft_p50_ms": 904.9503420246765, + "ttft_p95_ms": 1484.6198010491207 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3391.811708919704, + 3362.9186518955976 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r32p00.json", + "sha256": "876c5444d17c4ce81eca907ae17d5cf7062ffa67fa266b491580e52f712614ec", + "summary": { + "admission_lag_max_ms": 0.12557301670312881, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3486.074637970887, + "ttft_p50_ms": 1840.204154024832, + "ttft_p95_ms": 3391.811708919704 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r32p00.json", + "sha256": "02b4cd93ce1f9816e822022fb608ba8d4c89265f673f6d27c8fa3d4f1523ffe0", + "summary": { + "admission_lag_max_ms": 0.11014204937964678, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3456.965696066618, + "ttft_p50_ms": 1812.343467026949, + "ttft_p95_ms": 3362.9186518955976 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4306.351027917117, + 4302.602957002819 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round1/results/r64p00.json", + "sha256": "13ba282f179b2d6ce13a0aa2fd04eb29f82e579b7995e3406cb0afd69a2d5ee2", + "summary": { + "admission_lag_max_ms": 0.3526840591803193, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4359.505696920678, + "ttft_p50_ms": 2411.1130589153618, + "ttft_p95_ms": 4306.351027917117 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z/artifacts/artifacts/real-tp2-mns32-v1/round2/results/r64p00.json", + "sha256": "767622d62f0757ef2d8c7651ea470bfa2e62edad0ed4e75f66f84a0b722e35da", + "summary": { + "admission_lag_max_ms": 0.10626309085637331, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4355.409953044727, + "ttft_p50_ms": 2409.5974090741947, + "ttft_p95_ms": 4302.602957002819 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 32, + "name": "tp2_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 96.54812794178724, + 95.08875198662281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r4p00.json", + "sha256": "6e6a45d0c1010094c3395969815fd8cf9c35b919db202d772fb514744f981ebc", + "summary": { + "admission_lag_max_ms": 0.11306197848170996, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.27749610505998, + "ttft_p50_ms": 93.01548008807003, + "ttft_p95_ms": 96.54812794178724 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r4p00.json", + "sha256": "58aa71d694704cfea15c37cb29420dbac1dc505045c29a731e8be904846cca5f", + "summary": { + "admission_lag_max_ms": 0.11853000614792109, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 98.74943597242236, + "ttft_p50_ms": 93.57558900956064, + "ttft_p95_ms": 95.08875198662281 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.24941800534725, + 96.21043398510665 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r8p00.json", + "sha256": "311350af236cb0306aa635f05f8999b66bf5080bad7cc1f6703fffe0be31aee6", + "summary": { + "admission_lag_max_ms": 0.19771000370383263, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52211803849787, + "ttft_p50_ms": 92.4453770276159, + "ttft_p95_ms": 94.24941800534725 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r8p00.json", + "sha256": "60fe989a36a526e76f6f5c98e217b9feaf081bc680b16b9d53650bf085047c06", + "summary": { + "admission_lag_max_ms": 0.3933720290660858, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 102.97230398282409, + "ttft_p50_ms": 93.26327103190124, + "ttft_p95_ms": 96.21043398510665 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.98661107197404, + 92.9473100695759 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r10p00.json", + "sha256": "d0a04444a0e41bf76f71ac6ea3815bfbe76f7a51f8ddf669976bc3a4896dd0c1", + "summary": { + "admission_lag_max_ms": 0.10270101483911276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 103.49742788821459, + "ttft_p50_ms": 94.55623896792531, + "ttft_p95_ms": 97.98661107197404 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r10p00.json", + "sha256": "f330f26f45f6039c48150a0b6b146ca3dbb9ea8b2dff086a9efd5938ab981eb7", + "summary": { + "admission_lag_max_ms": 0.11239596642553806, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.4133670097217, + "ttft_p50_ms": 91.2688790122047, + "ttft_p95_ms": 92.9473100695759 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 260.68170997314155, + 319.04888805001974 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r12p00.json", + "sha256": "fed5eacf18018402b0c2675b8c5595f92693a9bddec30f354d989098f9fe4965", + "summary": { + "admission_lag_max_ms": 0.11784560047090054, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 345.19017208367586, + "ttft_p50_ms": 211.41935500781983, + "ttft_p95_ms": 260.68170997314155 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r12p00.json", + "sha256": "f2be27cbeff3e5c1a95050418f2dc3c712a4648cada8cdb3d36382cfffc9593a", + "summary": { + "admission_lag_max_ms": 0.12930459342896938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 326.86293800361454, + "ttft_p50_ms": 229.8808820778504, + "ttft_p95_ms": 319.04888805001974 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 895.7287209341303, + 912.1733580250293 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r14p00.json", + "sha256": "c2e157808e0308cc443f3abdf5114898cef66c435688d24321f80bef2d966208", + "summary": { + "admission_lag_max_ms": 0.14902697876095772, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 962.3544370988384, + "ttft_p50_ms": 576.2168710352853, + "ttft_p95_ms": 895.7287209341303 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r14p00.json", + "sha256": "749ef7d31bf8782d42719b75ee1a2e4a443f097510af721d02cf53df191bc66e", + "summary": { + "admission_lag_max_ms": 0.42896345257759094, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 986.0872740391642, + "ttft_p50_ms": 569.7114520007744, + "ttft_p95_ms": 912.1733580250293 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1364.6171110449359, + 1366.6857269126922 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r16p00.json", + "sha256": "7545ea9210348019de88ede5d2cba7527913534274deee034fa6811c173a6e72", + "summary": { + "admission_lag_max_ms": 0.12612901628017426, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1435.8308439841494, + "ttft_p50_ms": 848.9949750946835, + "ttft_p95_ms": 1364.6171110449359 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r16p00.json", + "sha256": "d75e27137f6a7811126d8c957bb5a081944fc1bb7643852cdd785111773f6010", + "summary": { + "admission_lag_max_ms": 0.1258660340681672, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.859375, + "passed": 55, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1436.2916270038113, + "ttft_p50_ms": 846.2129919789732, + "ttft_p95_ms": 1366.6857269126922 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3243.9196659252048, + 3214.524411014281 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r32p00.json", + "sha256": "366ebddc167338ed096f373de5667740721422c13d2db1f887ed09186afafd19", + "summary": { + "admission_lag_max_ms": 0.09669095743447542, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3337.068362045102, + "ttft_p50_ms": 1751.3202880509198, + "ttft_p95_ms": 3243.9196659252048 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r32p00.json", + "sha256": "01b8f5081e36a64d5776285c8c5f423b53b1ad684c4b8f95a8caa5d997c6bf09", + "summary": { + "admission_lag_max_ms": 0.8310690755024552, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.328125, + "passed": 21, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3277.297410997562, + "ttft_p50_ms": 1790.5721209244803, + "ttft_p95_ms": 3214.524411014281 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4319.725521025248, + 4367.031275993213 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round1/results/r64p00.json", + "sha256": "92202709fd6bbd558ff6f1986df213848254110c4274447a7bd5879153290d46", + "summary": { + "admission_lag_max_ms": 0.10392803233116865, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4360.145599930547, + "ttft_p50_ms": 2403.5812759539112, + "ttft_p95_ms": 4319.725521025248 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z/artifacts/artifacts/real-tp2-mns64-v1/round2/results/r64p00.json", + "sha256": "9660d15d31bc70c5550ed29477f563313d4d81baf0e561126aebd3b7d097cfa3", + "summary": { + "admission_lag_max_ms": 0.17276196740567684, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.25, + "passed": 16, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4414.442001027055, + "ttft_p50_ms": 2260.2788450894877, + "ttft_p95_ms": 4367.031275993213 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 64, + "name": "tp2_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.17482794076204, + 98.03147299680859 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r4p00.json", + "sha256": "0e315f6d9e0087ad553e874b0c16c2e6de79eb6eca22eb118dc5deca52e6efe3", + "summary": { + "admission_lag_max_ms": 0.11708191595971584, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.92840604577214, + "ttft_p50_ms": 95.13642196543515, + "ttft_p95_ms": 97.17482794076204 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r4p00.json", + "sha256": "baa394a24d2a3db40a3a7ae6dae0c5ce49ff05499a1e17eaa0bca4b479ae33bf", + "summary": { + "admission_lag_max_ms": 0.10946101974695921, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 100.52333807107061, + "ttft_p50_ms": 93.19468098692596, + "ttft_p95_ms": 98.03147299680859 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 97.67980605829507, + 93.80627400241792 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r8p00.json", + "sha256": "303df27c6d4604d5af65f9056acacb7c4373fb6eb7edb6e5d42689eeaa384aaa", + "summary": { + "admission_lag_max_ms": 0.11111306957900524, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 113.2329449756071, + "ttft_p50_ms": 93.05417106952518, + "ttft_p95_ms": 97.67980605829507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r8p00.json", + "sha256": "cdbb1bca17b24b67e18ef3dd4997494d6bd2a887b3b5d1264fbf411e688c53bc", + "summary": { + "admission_lag_max_ms": 0.12463307939469814, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 101.12245299387723, + "ttft_p50_ms": 92.26030297577381, + "ttft_p95_ms": 93.80627400241792 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 10.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 94.45723297540098, + 93.57301203999668 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r10p00.json", + "sha256": "fe17ef2417319307d651bc10b01a618234d61bab797ad87a7332cab20d78ecb2", + "summary": { + "admission_lag_max_ms": 0.3267280990257859, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 99.28593400400132, + "ttft_p50_ms": 92.22474799025804, + "ttft_p95_ms": 94.45723297540098 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r10p00.json", + "sha256": "550f693708a34624a82d77afd4da0ab4bc01e09f42242a3a79e4532897681a51", + "summary": { + "admission_lag_max_ms": 0.11238199658691883, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.77756198309362, + "ttft_p50_ms": 92.07128698471934, + "ttft_p95_ms": 93.57301203999668 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 12.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 320.49866404850036, + 336.9630330707878 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r12p00.json", + "sha256": "edab8169453151de732923c076b8f7dca4082ee47ccd2597e65e7ba68516bc2e", + "summary": { + "admission_lag_max_ms": 0.09914662223309278, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 327.930542989634, + "ttft_p50_ms": 231.4880370395258, + "ttft_p95_ms": 320.49866404850036 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r12p00.json", + "sha256": "a6dde4363d362ca10254409e3dc734d0f0e640e63156426f3f4fd7115db49e16", + "summary": { + "admission_lag_max_ms": 0.17016706988215446, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 428.6657409975305, + "ttft_p50_ms": 237.09688894450665, + "ttft_p95_ms": 336.9630330707878 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 14.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 911.7195709841326, + 910.7652489328757 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r14p00.json", + "sha256": "7eba81ee5c85c457a056f9aef28abd94d375fdaa116929cfe0a0c282a0738bd2", + "summary": { + "admission_lag_max_ms": 0.13163150288164616, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 999.2953530745581, + "ttft_p50_ms": 580.023527960293, + "ttft_p95_ms": 911.7195709841326 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r14p00.json", + "sha256": "6a113d3d876ff01a9c1101e8b75efa2fc71f96683d6bb430e26d5976426b5d23", + "summary": { + "admission_lag_max_ms": 0.11634768452495337, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 988.8870939612389, + "ttft_p50_ms": 578.9327350212261, + "ttft_p95_ms": 910.7652489328757 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 16.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 1389.8832038976252, + 1365.0342189939693 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r16p00.json", + "sha256": "d677251ee1a941c52469052a239c2e13be0a837285069aa8182a16e196445e48", + "summary": { + "admission_lag_max_ms": 0.11070503387600183, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1460.5093500576913, + "ttft_p50_ms": 864.3530959961936, + "ttft_p95_ms": 1389.8832038976252 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r16p00.json", + "sha256": "f5177f53aa0f87a4e7291f470978d86ad8bc46bd1a8362b79f48cd128646dc10", + "summary": { + "admission_lag_max_ms": 0.3672370221465826, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.828125, + "passed": 53, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1491.050804965198, + "ttft_p50_ms": 841.7225609300658, + "ttft_p95_ms": 1365.0342189939693 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 32.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 3258.010295103304, + 3258.664960041642 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r32p00.json", + "sha256": "02a6ac2b9c2b9ab97854d6fcc13b64bbb36a079e4afdae84231e4cb187fac4ef", + "summary": { + "admission_lag_max_ms": 0.1296499976888299, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.278529899195, + "ttft_p50_ms": 1755.0864539807662, + "ttft_p95_ms": 3258.010295103304 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r32p00.json", + "sha256": "9ca2f44fc7e3cb3d38ada548af858903b31bbee5cd50f1723c29255ebd4224e9", + "summary": { + "admission_lag_max_ms": 1.1343059595674276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.3125, + "passed": 20, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 3352.7101600775495, + "ttft_p50_ms": 1755.3453800501302, + "ttft_p95_ms": 3258.664960041642 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 4190.255103982054, + 4189.1050330596045 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round1/results/r64p00.json", + "sha256": "53908d3a254e19691839831b4337014b09591ced9189c2a9b29e0d92e80fcb56", + "summary": { + "admission_lag_max_ms": 0.11591101065278053, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4242.029857938178, + "ttft_p50_ms": 2340.5529640149325, + "ttft_p95_ms": 4190.255103982054 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z/artifacts/artifacts/real-tp2-mns8-v1/round2/results/r64p00.json", + "sha256": "1c8c85c873aa96c01dc0200f848474a53c1173ef72f3828e8c23aeae70e92b11", + "summary": { + "admission_lag_max_ms": 0.284243025816977, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.234375, + "passed": 15, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 4240.770195960067, + "ttft_p50_ms": 2339.3583690049127, + "ttft_p95_ms": 4189.1050330596045 + } + } + ] + } + ], + "capacity": 14.0, + "capacity_per_gpu": 7.0, + "mns": 8, + "name": "tp2_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 74.66826005838811, + 66.5705680148676 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r4p00.json", + "sha256": "fb354fc8a097caa21844d4ea5515b98b03459f400816e5e5590a6cda81cf4815", + "summary": { + "admission_lag_max_ms": 0.1035250024870038, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 93.38432096410543, + "ttft_p50_ms": 65.75745099689811, + "ttft_p95_ms": 74.66826005838811 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r4p00.json", + "sha256": "ac8e8412f79a8a32ebf085445a94aa548dd2d646fce0b9f56b4378081dcc8960", + "summary": { + "admission_lag_max_ms": 0.12108997907489538, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.89856600761414, + "ttft_p50_ms": 62.386377016082406, + "ttft_p95_ms": 66.5705680148676 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.44799195136875, + 72.29184894822538 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r8p00.json", + "sha256": "05ebbd68bbf24e9c8b25684474411c7db122b92701cc1617f9d786cf19592751", + "summary": { + "admission_lag_max_ms": 0.1825150102376938, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.92095802165568, + "ttft_p50_ms": 60.598687967285514, + "ttft_p95_ms": 70.44799195136875 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r8p00.json", + "sha256": "d24f7daaf41073cc8aa030b74c06203783684c3e556770366883c37b826406f2", + "summary": { + "admission_lag_max_ms": 0.10628998279571533, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 75.13270596973598, + "ttft_p50_ms": 60.161596979014575, + "ttft_p95_ms": 72.29184894822538 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.02151494007558, + 107.58775402791798 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r16p00.json", + "sha256": "a29b2128983af630e1da8f970cbda802849036bc522eb506821bf2fe1a34172f", + "summary": { + "admission_lag_max_ms": 0.10468997061252594, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.93136999942362, + "ttft_p50_ms": 59.5597909996286, + "ttft_p95_ms": 66.02151494007558 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r16p00.json", + "sha256": "275ae35651cbe2200d16c9b06a68568efda527564065255a10dee6fe93346010", + "summary": { + "admission_lag_max_ms": 0.3144120564684272, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 119.78317995090038, + "ttft_p50_ms": 59.12985489703715, + "ttft_p95_ms": 107.58775402791798 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 193.8455159543082, + 187.52167106140405 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r20p00.json", + "sha256": "4cb485f30b81c004bab379e6522d94b811c0bdc0be6d8e42c9d8ae842d9600dc", + "summary": { + "admission_lag_max_ms": 0.10407401714473963, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 197.7908629924059, + "ttft_p50_ms": 142.73560303263366, + "ttft_p95_ms": 193.8455159543082 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r20p00.json", + "sha256": "63aa9dc03a08b781c5d6e8805f2b61698c75e71673608ecd38bd1132c62e6987", + "summary": { + "admission_lag_max_ms": 0.09386602323502302, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 190.29632199089974, + "ttft_p50_ms": 136.1723990412429, + "ttft_p95_ms": 187.52167106140405 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 568.2198900030926, + 573.4539540717378 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r24p00.json", + "sha256": "5e4d5b44850090d07fe1d3abc44eddea3bfdf2b682f59b12a45a866621e38f98", + "summary": { + "admission_lag_max_ms": 0.7511250441893935, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 613.2891150191426, + "ttft_p50_ms": 375.27791305910796, + "ttft_p95_ms": 568.2198900030926 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r24p00.json", + "sha256": "239e7789aed2e9b2bcf1d127215a6b122071eb856be2db64ccc184fda153aa48", + "summary": { + "admission_lag_max_ms": 0.11180061846971512, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 614.9205249967054, + "ttft_p50_ms": 375.5133739905432, + "ttft_p95_ms": 573.4539540717378 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 860.0242879474536, + 886.1482660286129 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r28p00.json", + "sha256": "8b249b52779e5b3cd9028aaee30ee73ef34b0b0a64dcd75b9aad57266c29fbc2", + "summary": { + "admission_lag_max_ms": 0.11425174307078123, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.8854410191998, + "ttft_p50_ms": 534.7980660153553, + "ttft_p95_ms": 860.0242879474536 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r28p00.json", + "sha256": "c9b28c8f3187542a343ecf1067783569ec2abfa64631d6827a94ba6876aa7b63", + "summary": { + "admission_lag_max_ms": 0.312068616040051, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 922.4315020255744, + "ttft_p50_ms": 549.7413279954344, + "ttft_p95_ms": 886.1482660286129 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1147.1082780044526, + 1137.236590962857 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r32p00.json", + "sha256": "d27ff9f940a6cd7fba339d9a176de39cf4b5bd164dcc500f2af36bcb07cdef30", + "summary": { + "admission_lag_max_ms": 0.16146607231348753, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.626160996966, + "ttft_p50_ms": 696.4708579471335, + "ttft_p95_ms": 1147.1082780044526 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r32p00.json", + "sha256": "093044e87fffdaa222e5bad3807cbf4900aa64891d5b4ba27f187c7a411252d7", + "summary": { + "admission_lag_max_ms": 0.10430300608277321, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.045956928283, + "ttft_p50_ms": 684.3201500596479, + "ttft_p95_ms": 1137.236590962857 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2064.763988018967, + 2039.6780440350994 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round1/results/r64p00.json", + "sha256": "fad7bdd80d5ca603790a4e5d19dadd4f494540b106ce583e3e8c1c05d2374a58", + "summary": { + "admission_lag_max_ms": 6.394341005943716, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.53125, + "passed": 34, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2144.80376499705, + "ttft_p50_ms": 1158.5425049997866, + "ttft_p95_ms": 2064.763988018967 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z/artifacts/artifacts/real-tp4-mns16-v1/round2/results/r64p00.json", + "sha256": "24fdb945706dbbdd29916e0f08e0eff807334db212fded94c729da2e1b46efd3", + "summary": { + "admission_lag_max_ms": 0.10294502135366201, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.4275270039216, + "ttft_p50_ms": 1131.1213460285217, + "ttft_p95_ms": 2039.6780440350994 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp4_mns16", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.80304794665426, + 64.59337309934199 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r4p00.json", + "sha256": "2e82a30f02c2702668e59661930431cf82323746a111d42bf38b7fdc0d083b3d", + "summary": { + "admission_lag_max_ms": 0.1332589890807867, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.74050498381257, + "ttft_p50_ms": 64.10286005120724, + "ttft_p95_ms": 67.80304794665426 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r4p00.json", + "sha256": "8294196c7dc52765a6820f5e1d8439c480490daed9505263f707810f73c95783", + "summary": { + "admission_lag_max_ms": 0.1283179735764861, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.6796320155263, + "ttft_p50_ms": 60.962425894103944, + "ttft_p95_ms": 64.59337309934199 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 70.46496798284352, + 68.22343391831964 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r8p00.json", + "sha256": "38bbfeab47e5eb6296fcccf20f04ef25dc24db658be92300c492734bbfe333cf", + "summary": { + "admission_lag_max_ms": 0.10555307380855083, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 87.7126749837771, + "ttft_p50_ms": 62.65173899009824, + "ttft_p95_ms": 70.46496798284352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r8p00.json", + "sha256": "716f90177d6f203717c80a8e9b28697e53141f204af470493c299bc73a1ea3ee", + "summary": { + "admission_lag_max_ms": 0.24518708232790232, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 80.40577603969723, + "ttft_p50_ms": 60.29119901359081, + "ttft_p95_ms": 68.22343391831964 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 89.53388908412308, + 76.93439698778093 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r16p00.json", + "sha256": "063016efe822c31c2425e7d963eda4c054f3f7c0341df0a3cd73ab86a942e873", + "summary": { + "admission_lag_max_ms": 0.2996840048581362, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 97.41822991054505, + "ttft_p50_ms": 60.97789097111672, + "ttft_p95_ms": 89.53388908412308 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r16p00.json", + "sha256": "eecc6bdeff90a8bf195dfc5387c6bc2f4db20ed7c9aeb9b4ebe050efccc6051a", + "summary": { + "admission_lag_max_ms": 0.36267400719225407, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 79.42048599943519, + "ttft_p50_ms": 60.29245804529637, + "ttft_p95_ms": 76.93439698778093 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 206.2494510319084, + 201.13301498349756 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r20p00.json", + "sha256": "22a62729358fea7c38ca166a1a49f1235e98c7f07ecfe4948322110cda4b7178", + "summary": { + "admission_lag_max_ms": 0.11367700062692165, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 255.14698098413646, + "ttft_p50_ms": 155.46583198010921, + "ttft_p95_ms": 206.2494510319084 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r20p00.json", + "sha256": "be103a759100bd2ab6dd31ee4ce41b6f0da40d232a60ebe137e3d933194a487c", + "summary": { + "admission_lag_max_ms": 0.17853605095297098, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 209.22932599205524, + "ttft_p50_ms": 155.12292610947043, + "ttft_p95_ms": 201.13301498349756 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 580.5047769099474, + 577.3971510352567 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r24p00.json", + "sha256": "890a733a97f399deb4c5efcca661d9a442ed76080cd5d8b6a2c09368aef701b7", + "summary": { + "admission_lag_max_ms": 0.1084543764591217, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 621.5670789824799, + "ttft_p50_ms": 374.6346529806033, + "ttft_p95_ms": 580.5047769099474 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r24p00.json", + "sha256": "87e8543652b1743b1b9b8c3de0fc680401b0bf9a33d61fa6132120c944b2a6eb", + "summary": { + "admission_lag_max_ms": 0.11801638174802065, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 623.2662099646404, + "ttft_p50_ms": 382.11427396163344, + "ttft_p95_ms": 577.3971510352567 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 873.0683700414374, + 889.4930009264499 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r28p00.json", + "sha256": "f877781730a5437b4cbdb76948915f1c871574b19a6df27076d21b65bb18e678", + "summary": { + "admission_lag_max_ms": 0.6807943573221564, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 943.9258739585057, + "ttft_p50_ms": 542.9852310335264, + "ttft_p95_ms": 873.0683700414374 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r28p00.json", + "sha256": "82151dc539b35302489c17976cfe89b8f4c64944298954607fad8f5c9cb9e7d8", + "summary": { + "admission_lag_max_ms": 6.750226486474276, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 936.5667289821431, + "ttft_p50_ms": 546.0105559322983, + "ttft_p95_ms": 889.4930009264499 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1135.3467140579596, + 1146.743991994299 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r32p00.json", + "sha256": "082cc4c47b2eec6e49f0edd14cff5a51b880bd998d98da8b9830bbf7615483ef", + "summary": { + "admission_lag_max_ms": 0.10443700011819601, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1199.8225760180503, + "ttft_p50_ms": 680.4794340860099, + "ttft_p95_ms": 1135.3467140579596 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r32p00.json", + "sha256": "517d3cbc1104b5ea1bdb0e381c033cf978181a93449b6498d30b52cd3dd3e770", + "summary": { + "admission_lag_max_ms": 0.4415180301293731, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1210.0602959981188, + "ttft_p50_ms": 693.0603379150853, + "ttft_p95_ms": 1146.743991994299 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2084.9693010095507, + 2041.9401369290426 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round1/results/r64p00.json", + "sha256": "7e11771ed598826fb3c5594c7e05d0989e01030be2d8b36fa21813aff462496d", + "summary": { + "admission_lag_max_ms": 0.10768393985927105, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.128676981665, + "ttft_p50_ms": 1113.0280290963128, + "ttft_p95_ms": 2084.9693010095507 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z/artifacts/artifacts/real-tp4-mns32-v1/round2/results/r64p00.json", + "sha256": "cf550423832a92cba942b7e1dc14a42000e8c9507403de8cd49aa56af9db6706", + "summary": { + "admission_lag_max_ms": 0.11262902989983559, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2122.8132860269397, + "ttft_p50_ms": 1131.2310450011864, + "ttft_p95_ms": 2041.9401369290426 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp4_mns32", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 66.89354998525232, + 83.23234610725194 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r4p00.json", + "sha256": "266cba4088f8a9483ceb41fdc02d04a2181cb7ffb6b52372b6b0687e62f2863e", + "summary": { + "admission_lag_max_ms": 0.10568101424723864, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 72.09669100120664, + "ttft_p50_ms": 62.014322029426694, + "ttft_p95_ms": 66.89354998525232 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r4p00.json", + "sha256": "26f3926c40504b96a37cb7b9dae1af5040c085c1a730e20feaf1d3e9ddbf8e54", + "summary": { + "admission_lag_max_ms": 0.108166947029531, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 96.42452490516007, + "ttft_p50_ms": 61.261708033271134, + "ttft_p95_ms": 83.23234610725194 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 67.26283801253885, + 67.34374002553523 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r8p00.json", + "sha256": "e84e6ba2c8822e012aceea7cc6f376ea1b6265fe4a3bb68b7badfc36b8813545", + "summary": { + "admission_lag_max_ms": 0.15694997273385525, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54619695153087, + "ttft_p50_ms": 59.63437003083527, + "ttft_p95_ms": 67.26283801253885 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r8p00.json", + "sha256": "dae50073c9f79919181e8dcf274ca8929ba41c512f64363e40187de68839b87e", + "summary": { + "admission_lag_max_ms": 0.09962497279047966, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.00161597877741, + "ttft_p50_ms": 60.2562150452286, + "ttft_p95_ms": 67.34374002553523 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 71.37343694921583, + 64.9534990079701 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r16p00.json", + "sha256": "776c9956f65acf7c9b6c8df517a4f8c4d462289893a19e625a2a0ad7dcbf5506", + "summary": { + "admission_lag_max_ms": 0.1066719414666295, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 82.25136005785316, + "ttft_p50_ms": 58.52357891853899, + "ttft_p95_ms": 71.37343694921583 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r16p00.json", + "sha256": "90057204a5f6343ee4ee9cdd08323d87d83660d5bf8f02480882eb479d2e587e", + "summary": { + "admission_lag_max_ms": 0.10102998930960894, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 65.3582609957084, + "ttft_p50_ms": 58.98014397826046, + "ttft_p95_ms": 64.9534990079701 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 192.98715889453888, + 258.1990590551868 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r20p00.json", + "sha256": "6ce029388b206d0663acd41f4da1d67a6562389e1c84cf1826ce77c05d176d9a", + "summary": { + "admission_lag_max_ms": 0.12691703159362078, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 198.67739104665816, + "ttft_p50_ms": 142.75032398290932, + "ttft_p95_ms": 192.98715889453888 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r20p00.json", + "sha256": "0dbfb3f012574988ddebfb0af1532c735ac62d6d88863d24cd820aed50a92a59", + "summary": { + "admission_lag_max_ms": 0.3961039474233985, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 259.55891504418105, + "ttft_p50_ms": 162.6282810466364, + "ttft_p95_ms": 258.1990590551868 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 583.3089610096067, + 571.3486529421061 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r24p00.json", + "sha256": "0774ba2f285d81447c63343b32551d6fc33369b54010629381efa9a37df23a9c", + "summary": { + "admission_lag_max_ms": 0.09822729043662548, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 628.714332007803, + "ttft_p50_ms": 366.9840869260952, + "ttft_p95_ms": 583.3089610096067 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r24p00.json", + "sha256": "1986a05c366ebe73af0f89d2cd0e4efd3e577c800cf1e80997cd26ea0aaaec09", + "summary": { + "admission_lag_max_ms": 0.11414836626499891, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 617.9165309295058, + "ttft_p50_ms": 376.65714998729527, + "ttft_p95_ms": 571.3486529421061 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 869.708820944652, + 866.4831441128626 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r28p00.json", + "sha256": "ab987729fc833a59817eabaa18c320b0d0a2ecd402fd02bdb534b6c773c029d8", + "summary": { + "admission_lag_max_ms": 0.578614417463541, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 941.3277570856735, + "ttft_p50_ms": 540.4732850147411, + "ttft_p95_ms": 869.708820944652 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r28p00.json", + "sha256": "38f9c82c9ee30927cc0ba857e5e3fce2e79b8a24de03efaec77ce27258633a7e", + "summary": { + "admission_lag_max_ms": 0.12532551772892475, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 938.2033230504021, + "ttft_p50_ms": 541.139594046399, + "ttft_p95_ms": 866.4831441128626 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1132.6635719742626, + 1144.196804962121 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r32p00.json", + "sha256": "358f5b8e1b72bff5752c076f2966b966d98f6f5767ce52d5a255b7f68f061e54", + "summary": { + "admission_lag_max_ms": 0.32093306072056293, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1197.7118930080906, + "ttft_p50_ms": 678.4568330040202, + "ttft_p95_ms": 1132.6635719742626 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r32p00.json", + "sha256": "3f6dbb760582fdd7470422d8edf6b62949d158ad0d437f04015441a0a4509003", + "summary": { + "admission_lag_max_ms": 1.1304179206490517, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1193.1529619032517, + "ttft_p50_ms": 692.5535890040919, + "ttft_p95_ms": 1144.196804962121 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2042.27216495201, + 2043.943466967903 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round1/results/r64p00.json", + "sha256": "a24eae7db6eb4cd105ed903917988e8fe6eef5d4cc2549d4e05e9a03dfafb107", + "summary": { + "admission_lag_max_ms": 0.134881935082376, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2132.23939598538, + "ttft_p50_ms": 1137.3268909519538, + "ttft_p95_ms": 2042.27216495201 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z/artifacts/artifacts/real-tp4-mns64-v1/round2/results/r64p00.json", + "sha256": "e85efaada0b64bb9dd686f59282171ccb256a1d8483c6f037589a68e3e1a104a", + "summary": { + "admission_lag_max_ms": 0.11892011389136314, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2125.3299129894003, + "ttft_p50_ms": 1134.724210947752, + "ttft_p95_ms": 2043.943466967903 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp4_mns64", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "conservative_feasible": true, + "rate": 4.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 82.33969006687403, + 69.57727600820363 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r4p00.json", + "sha256": "e69b89ad2e07e6560ebd870773532b33e4415e9ca1515e585cbe465616678569", + "summary": { + "admission_lag_max_ms": 0.14186999760568142, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 86.54833806212991, + "ttft_p50_ms": 62.749656033702195, + "ttft_p95_ms": 82.33969006687403 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r4p00.json", + "sha256": "2c76dcf112a828a6720e079d29bb54a6a65d37d26abaaafdc6d5580751dbad65", + "summary": { + "admission_lag_max_ms": 0.11478399392217398, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 81.3983449479565, + "ttft_p50_ms": 62.220132909715176, + "ttft_p95_ms": 69.57727600820363 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 8.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 63.71171900536865, + 68.14328604377806 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r8p00.json", + "sha256": "0e5d687dfcdb4b6a0d1d060ceefe8a74faab94a5b241580d45b53c452f893b23", + "summary": { + "admission_lag_max_ms": 0.09818794205784798, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 69.63772000744939, + "ttft_p50_ms": 60.89008797425777, + "ttft_p95_ms": 63.71171900536865 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r8p00.json", + "sha256": "1bf016b603473d46fda4bc6d4daeaee5e1091ff4cda1cab7224e0c5edea512f8", + "summary": { + "admission_lag_max_ms": 0.11399004142731428, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 76.76690397784114, + "ttft_p50_ms": 60.9520060243085, + "ttft_p95_ms": 68.14328604377806 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 16.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 133.86176899075508, + 61.95738597307354 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r16p00.json", + "sha256": "197808b1b5fa8082e603a255f358826b5f8f48707223c7fd63abfe029674b817", + "summary": { + "admission_lag_max_ms": 0.10530801955610514, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 191.68931001331657, + "ttft_p50_ms": 61.92394997924566, + "ttft_p95_ms": 133.86176899075508 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r16p00.json", + "sha256": "53e809470daf9efcdabb254e85e8cc3b45d0ab8593c6c2f3aa3396e60d1656da", + "summary": { + "admission_lag_max_ms": 0.1288399798795581, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 70.40882098954171, + "ttft_p50_ms": 59.417470009066164, + "ttft_p95_ms": 61.95738597307354 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 20.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 201.29013701807708, + 200.0734859611839 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r20p00.json", + "sha256": "05b977824fb301517cdbe2d077b360b6053d57fee1c20acad3537321827d027a", + "summary": { + "admission_lag_max_ms": 0.3848400665447116, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.5107699232176, + "ttft_p50_ms": 150.0862929970026, + "ttft_p95_ms": 201.29013701807708 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r20p00.json", + "sha256": "baf422b7a3c95b0d3d376a31b0a30e075f69216c794ebc2ef51afafe1ec55229", + "summary": { + "admission_lag_max_ms": 0.24754798505455256, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 204.8192109214142, + "ttft_p50_ms": 148.99871195666492, + "ttft_p95_ms": 200.0734859611839 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 24.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 564.2021730309352, + 560.4996709153056 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r24p00.json", + "sha256": "029b2a9695322d536539c3dd7665725f933b21a8bb48fa94c0448c9f7f8bf0eb", + "summary": { + "admission_lag_max_ms": 0.14956865925341845, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 610.4218009859324, + "ttft_p50_ms": 368.3909500250593, + "ttft_p95_ms": 564.2021730309352 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r24p00.json", + "sha256": "94b46583af3a7a4bbb9f7256c4c4a5ce804b3092459abe3984ea0ce67f591f3c", + "summary": { + "admission_lag_max_ms": 0.11110503692179918, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 606.6705230623484, + "ttft_p50_ms": 367.50321893487126, + "ttft_p95_ms": 560.4996709153056 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 28.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 870.7952990662307, + 894.6108610834926 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r28p00.json", + "sha256": "07bfe263107bb3d2acaa356257343f90045e4295c799178bc0a0b5f1c1994284", + "summary": { + "admission_lag_max_ms": 0.6798935355618596, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 942.792383953929, + "ttft_p50_ms": 543.340316042304, + "ttft_p95_ms": 870.7952990662307 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r28p00.json", + "sha256": "861dd55076ef40a78c34824345c5279dff86262b604db120e9e38efa27b5be55", + "summary": { + "admission_lag_max_ms": 0.09912776295095682, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 930.2741150604561, + "ttft_p50_ms": 549.9696529004723, + "ttft_p95_ms": 894.6108610834926 + } + } + ] + }, + { + "conservative_feasible": true, + "rate": 32.0, + "round_feasible": [ + true, + true + ], + "round_ttft_p95_ms": [ + 1136.603599996306, + 1134.0074760373682 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r32p00.json", + "sha256": "332b805322b571087b018ee459dd91c443dfe0b2bb539b5bdfb4de902bba6da2", + "summary": { + "admission_lag_max_ms": 0.19067595712840557, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1201.8779760692269, + "ttft_p50_ms": 683.3873559953645, + "ttft_p95_ms": 1136.603599996306 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r32p00.json", + "sha256": "996a12a5ed962218dfbe6f30be688119030f10f6539647b6e4a9f25f032b22cb", + "summary": { + "admission_lag_max_ms": 0.11958007235080004, + "completed": 64, + "failed": 0, + "slo": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 1198.4891929896548, + "ttft_p50_ms": 682.7383439522237, + "ttft_p95_ms": 1134.0074760373682 + } + } + ] + }, + { + "conservative_feasible": false, + "rate": 64.0, + "round_feasible": [ + false, + false + ], + "round_ttft_p95_ms": [ + 2040.8174369949847, + 2082.3236129945144 + ], + "rounds": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round1/results/r64p00.json", + "sha256": "c5f17f625071e905b5f1bf8045cd233b2fea0fe6ca00044935dee9c663d1526b", + "summary": { + "admission_lag_max_ms": 0.7995429914444685, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.546875, + "passed": 35, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2131.160691031255, + "ttft_p50_ms": 1134.8516750149429, + "ttft_p95_ms": 2040.8174369949847 + } + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z/artifacts/artifacts/real-tp4-mns8-v1/round2/results/r64p00.json", + "sha256": "775db5b941347d9fd5d6df32d6ee037f8168bd47b9eae9a7bdb3dbd8c40aa990", + "summary": { + "admission_lag_max_ms": 0.17620902508497238, + "completed": 64, + "failed": 0, + "slo": { + "feasible": false, + "pass_rate": 0.5625, + "passed": 36, + "ttft_threshold_ms": 1256.0 + }, + "ttft_max_ms": 2129.518053960055, + "ttft_p50_ms": 1112.3740070033818, + "ttft_p95_ms": 2082.3236129945144 + } + } + ] + } + ], + "capacity": 32.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp4_mns8", + "source_run": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive/qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + }, + "real_campaign_resources": { + "fresh_server_anchors": 192, + "gpu_hours": 12.074444444444445, + "measured_requests": 12288, + "runs": [ + { + "duration_seconds": 908.0, + "gpu_hours": 0.25222222222222224, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v2-exclusive-20260716T164730291172Z", + "tp": 1 + }, + { + "duration_seconds": 540.0, + "gpu_hours": 0.15, + "run": "qwen30-prefill-real-tp1-mns16-20260717-v3-refine-20260716T181031481585Z", + "tp": 1 + }, + { + "duration_seconds": 937.0, + "gpu_hours": 0.2602777777777778, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v2-exclusive-20260716T164731376588Z", + "tp": 1 + }, + { + "duration_seconds": 558.0, + "gpu_hours": 0.155, + "run": "qwen30-prefill-real-tp1-mns32-20260717-v3-refine-20260716T181032543571Z", + "tp": 1 + }, + { + "duration_seconds": 972.0, + "gpu_hours": 0.27, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v2-exclusive-20260716T164732482666Z", + "tp": 1 + }, + { + "duration_seconds": 572.0, + "gpu_hours": 0.15888888888888889, + "run": "qwen30-prefill-real-tp1-mns64-20260717-v3-refine-20260716T181033562318Z", + "tp": 1 + }, + { + "duration_seconds": 886.0, + "gpu_hours": 0.2461111111111111, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v2-exclusive-20260716T164729267199Z", + "tp": 1 + }, + { + "duration_seconds": 525.0, + "gpu_hours": 0.14583333333333334, + "run": "qwen30-prefill-real-tp1-mns8-20260717-v3-refine-20260716T181030504792Z", + "tp": 1 + }, + { + "duration_seconds": 983.0, + "gpu_hours": 0.5461111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v2-exclusive-20260716T164734745039Z", + "tp": 2 + }, + { + "duration_seconds": 569.0, + "gpu_hours": 0.3161111111111111, + "run": "qwen30-prefill-real-tp2-mns16-20260717-v3-refine-20260716T181035686437Z", + "tp": 2 + }, + { + "duration_seconds": 1033.0, + "gpu_hours": 0.5738888888888889, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v2-exclusive-20260716T170909930153Z", + "tp": 2 + }, + { + "duration_seconds": 584.0, + "gpu_hours": 0.3244444444444444, + "run": "qwen30-prefill-real-tp2-mns32-20260717-v3-refine-20260716T182308386197Z", + "tp": 2 + }, + { + "duration_seconds": 1063.0, + "gpu_hours": 0.5905555555555555, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v2-exclusive-20260716T170910954319Z", + "tp": 2 + }, + { + "duration_seconds": 597.0, + "gpu_hours": 0.33166666666666667, + "run": "qwen30-prefill-real-tp2-mns64-20260717-v3-refine-20260716T182309391513Z", + "tp": 2 + }, + { + "duration_seconds": 957.0, + "gpu_hours": 0.5316666666666666, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v2-exclusive-20260716T164733581881Z", + "tp": 2 + }, + { + "duration_seconds": 563.0, + "gpu_hours": 0.31277777777777777, + "run": "qwen30-prefill-real-tp2-mns8-20260717-v3-refine-20260716T181034592841Z", + "tp": 2 + }, + { + "duration_seconds": 1008.0, + "gpu_hours": 1.12, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v2-exclusive-20260716T172922998830Z", + "tp": 4 + }, + { + "duration_seconds": 559.0, + "gpu_hours": 0.6211111111111111, + "run": "qwen30-prefill-real-tp4-mns16-20260717-v4-refine-20260716T183829231278Z", + "tp": 4 + }, + { + "duration_seconds": 1024.0, + "gpu_hours": 1.1377777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v2-exclusive-20260716T172923955862Z", + "tp": 4 + }, + { + "duration_seconds": 556.0, + "gpu_hours": 0.6177777777777778, + "run": "qwen30-prefill-real-tp4-mns32-20260717-v4-refine-20260716T183830222875Z", + "tp": 4 + }, + { + "duration_seconds": 959.0, + "gpu_hours": 1.0655555555555556, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v2b-exclusive-20260716T175248149074Z", + "tp": 4 + }, + { + "duration_seconds": 567.0, + "gpu_hours": 0.63, + "run": "qwen30-prefill-real-tp4-mns64-20260717-v4-refine-20260716T184826349294Z", + "tp": 4 + }, + { + "duration_seconds": 995.0, + "gpu_hours": 1.1055555555555556, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v2-exclusive-20260716T170911977980Z", + "tp": 4 + }, + { + "duration_seconds": 550.0, + "gpu_hours": 0.6111111111111112, + "run": "qwen30-prefill-real-tp4-mns8-20260717-v3-refine-20260716T182310414444Z", + "tp": 4 + } + ], + "successful_fleet_jobs": 24, + "warmup_requests": 4512 + }, + "schema": "qwen30-prefill-fidelity-comparison-v1", + "simulator": { + "tp1_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "tp1_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "tp1_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "tp1_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 5.0, + "ttft_p95_ms": 171.58529929215405 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 6.0, + "ttft_p95_ms": 333.9001759332376 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 7.0, + "ttft_p95_ms": 427.64609771465524 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 516.4107605377559 + }, + { + "feasible": false, + "pass_rate": 0.3125, + "rate": 16.0, + "ttft_p95_ms": 3193.8404181099845 + }, + { + "feasible": false, + "pass_rate": 0.203125, + "rate": 32.0, + "ttft_p95_ms": 4962.713638565687 + }, + { + "feasible": false, + "pass_rate": 0.140625, + "rate": 64.0, + "ttft_p95_ms": 5853.338638565687 + } + ], + "capacity": 8.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "tp2_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "tp2_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "tp2_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "tp2_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 122.24062760264154 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 10.0, + "ttft_p95_ms": 297.1650738307572 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 12.0, + "ttft_p95_ms": 384.6764910356253 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 14.0, + "ttft_p95_ms": 386.4490667329015 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 962.8278877316862 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 32.0, + "ttft_p95_ms": 2758.1797930156335 + }, + { + "feasible": false, + "pass_rate": 0.265625, + "rate": 64.0, + "ttft_p95_ms": 3622.636520081023 + } + ], + "capacity": 16.0, + "capacity_per_gpu": 8.0, + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "tp4_mns16": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "tp4_mns32": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "tp4_mns64": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "tp4_mns8": { + "anchors": [ + { + "feasible": true, + "pass_rate": 1.0, + "rate": 4.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 8.0, + "ttft_p95_ms": 93.12333977685495 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 16.0, + "ttft_p95_ms": 309.403713596319 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 20.0, + "ttft_p95_ms": 718.1022232545545 + }, + { + "feasible": true, + "pass_rate": 1.0, + "rate": 24.0, + "ttft_p95_ms": 1155.549457433053 + }, + { + "feasible": false, + "pass_rate": 0.765625, + "rate": 28.0, + "ttft_p95_ms": 1515.6009336189102 + }, + { + "feasible": false, + "pass_rate": 0.625, + "rate": 32.0, + "ttft_p95_ms": 1787.92236218991 + }, + { + "feasible": false, + "pass_rate": 0.390625, + "rate": 64.0, + "ttft_p95_ms": 2661.056949107159 + } + ], + "capacity": 24.0, + "capacity_per_gpu": 6.0, + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + } + }, + "simulator_sources": [ + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json", + "sha256": "2e3c4711bf9a5046daed3517c58f23241880467ee8f8e70191dc379413ad027c" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json", + "sha256": "faffabd900cc38c17a349f13840257078543af9d5da39c363a5d77a87d2eef93" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json", + "sha256": "e54e08cb263c376daf92f418a90ad79ae1afaa2edbed69271513218ab104e93c" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json", + "sha256": "b4530266c764e2bb1b420a68052d70e24f63eb9c30f3b3e8f8c9e4865a25f280" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json", + "sha256": "cd06f00b5c6f514896842e9f4482bd6270fedb0bf003823f1b974d136acdeffd" + }, + { + "path": "/home/gahow/phd/aituner/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json", + "sha256": "6ec105c50563e669ef145aa2b9a431e695a6308abb6d5db951b8a1f9d4a9e2f4" + } + ] +} diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json new file mode 100644 index 0000000..48c05b5 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 7.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.505136728286743, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.398303508758545, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 8, + "name": "tp1_mns8", + "tp": 1 + }, + "elapsed_seconds": 10.264978170394897, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.103909015655518, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.05237078666687, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 16, + "name": "tp1_mns16", + "tp": 1 + }, + "elapsed_seconds": 10.048887729644775, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.058743476867676, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.106712818145752, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 32, + "name": "tp1_mns32", + "tp": 1 + }, + "elapsed_seconds": 10.1020348072052, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.897645235061646, + "offered_request_rate": 5.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "914a1261d25b03b812dfa30880c3b0a0d7b697a8c13a9f3a26a7bb6eaecd0734", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.011124187030024, + "ttft_max_ms": 171.58529929215405, + "ttft_p50_ms": 171.58529929215405, + "ttft_p95_ms": 171.58529929215405 + }, + "status": "completed", + "trace_sha256": "98e4ac5df766b1cb5bfb7469f2d76368808c2e085e22eb3553cb8627592cacd7" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.957297325134277, + "offered_request_rate": 6.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "d8f0fe2fbf2ffb42b7e5033f611fc22f3293264859264b5f8d67f13f3a5b66c9", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 5.915042512446176, + "ttft_max_ms": 432.43322440414465, + "ttft_p50_ms": 282.618152646152, + "ttft_p95_ms": 333.9001759332376 + }, + "status": "completed", + "trace_sha256": "e0f9adf28d29542fcf834d482c990781724426f3d5cdcc043af53888393d1f64" + }, + { + "config": { + "mns": 64, + "name": "tp1_mns64", + "tp": 1 + }, + "elapsed_seconds": 9.799486875534058, + "offered_request_rate": 7.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7ba9181767b8cbd741db3794bc7729920d7741d63b795ade4aa1fe28a89f3a63", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 6.89299458621687, + "ttft_max_ms": 432.97081304433016, + "ttft_p50_ms": 297.34516049744997, + "ttft_p95_ms": 427.64609771465524 + }, + "status": "completed", + "trace_sha256": "3734627d83aa731c9f09a4c4fea7d994f110bcbe937e157906354a4eda9cbf48" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 5.0, + 6.0, + 7.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json new file mode 100644 index 0000000..da9191c --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 14.0, + "maximum_tested_feasible_request_rate_per_gpu": 7.0, + "upper_censored": true + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.598209381103516, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.660333633422852, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 8, + "name": "tp2_mns8", + "tp": 2 + }, + "elapsed_seconds": 10.409451007843018, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.360321521759033, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.402549028396606, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 16, + "name": "tp2_mns16", + "tp": 2 + }, + "elapsed_seconds": 10.40966272354126, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.155991315841675, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.103453874588013, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 32, + "name": "tp2_mns32", + "tp": 2 + }, + "elapsed_seconds": 10.255317687988281, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 10.106109380722046, + "offered_request_rate": 10.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "3597984a1acaffc6b3e8b1baa3ff41bb94ee1618f8780fcc6cd8d3bdb3d7201b", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 9.814433634929404, + "ttft_max_ms": 300.1784228926625, + "ttft_p50_ms": 211.39433537475938, + "ttft_p95_ms": 297.1650738307572 + }, + "status": "completed", + "trace_sha256": "92e3001d8467e3f97a243bea8323440ecf254a8ca211e52226fe886ebc9404be" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 10.112579107284546, + "offered_request_rate": 12.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "06641f2758bd3f801fb7b9c934670bc766372ad7353f5e14aea2988dfbec462a", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 11.55742285377119, + "ttft_max_ms": 389.93612032123883, + "ttft_p50_ms": 296.08352841701094, + "ttft_p95_ms": 384.6764910356253 + }, + "status": "completed", + "trace_sha256": "cbc40c05fb9e400fa5edea6dd52c4122a957418fe523d102f1e72a784321e681" + }, + { + "config": { + "mns": 64, + "name": "tp2_mns64", + "tp": 2 + }, + "elapsed_seconds": 9.847446918487549, + "offered_request_rate": 14.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "94cf0f3c510015ba50495eb38fd6fb452dd23d16ce0c788dc926229ecd557812", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 13.374732701347012, + "ttft_max_ms": 388.3013907870674, + "ttft_p50_ms": 309.4635231444016, + "ttft_p95_ms": 386.4490667329015 + }, + "status": "completed", + "trace_sha256": "3e43801bb724218ceaf0d52f17cd4d0fb3863e70f95c4d8d5a43ce7f85c75c01" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 10.0, + 12.0, + 14.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json b/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json new file mode 100644 index 0000000..0190387 --- /dev/null +++ b/runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json @@ -0,0 +1,461 @@ +{ + "capacity": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "lower_censored": false, + "maximum_tested_feasible_request_rate": 24.0, + "maximum_tested_feasible_request_rate_per_gpu": 6.0, + "upper_censored": false + } + ], + "config_results": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 10.649231672286987, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 10.245610237121582, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 8, + "name": "tp4_mns8", + "tp": 4 + }, + "elapsed_seconds": 9.953459978103638, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 10.059623956680298, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 10.000702381134033, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 16, + "name": "tp4_mns16", + "tp": 4 + }, + "elapsed_seconds": 9.852763652801514, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.849465370178223, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.80930495262146, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 32, + "name": "tp4_mns32", + "tp": 4 + }, + "elapsed_seconds": 9.798201560974121, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "loads": [ + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.800058603286743, + "offered_request_rate": 20.0, + "offered_request_rate_per_gpu": 5.0, + "request_count": 64, + "request_metrics_sha256": "2bf5e46241bf3462e21f707f715b4a9fe2c932b19da35b19b4a875f88605eee0", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.74355310200617, + "ttft_max_ms": 779.2433711652614, + "ttft_p50_ms": 476.39050138849325, + "ttft_p95_ms": 718.1022232545545 + }, + "status": "completed", + "trace_sha256": "da52e3a35a89135dd82d40415cf2dd43738eceb4183f322fcd38d96294a845d4" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.804483413696289, + "offered_request_rate": 24.0, + "offered_request_rate_per_gpu": 6.0, + "request_count": 64, + "request_metrics_sha256": "79986860bea420cd6e42385521fc8377dbc237a15010a4daf014bc9a94f71aa7", + "score": { + "feasible": true, + "pass_rate": 1.0, + "passed": 64, + "throughput_requests_per_second": 16.93744325871021, + "ttft_max_ms": 1219.4533647214066, + "ttft_p50_ms": 708.2221064125775, + "ttft_p95_ms": 1155.549457433053 + }, + "status": "completed", + "trace_sha256": "abae88091c5b7bb5a4b7f3f6fa553e132b10e7943c0dbe18dea203ac85c289fa" + }, + { + "config": { + "mns": 64, + "name": "tp4_mns64", + "tp": 4 + }, + "elapsed_seconds": 9.80262541770935, + "offered_request_rate": 28.0, + "offered_request_rate_per_gpu": 7.0, + "request_count": 64, + "request_metrics_sha256": "7f33dcb51a6e2a47acbc2f4ee968746f86575c0cbf62482ae08cfaf58216d459", + "score": { + "feasible": false, + "pass_rate": 0.765625, + "passed": 49, + "throughput_requests_per_second": 16.898601280793688, + "ttft_max_ms": 1587.02950504691, + "ttft_p50_ms": 885.3220562610811, + "ttft_p95_ms": 1515.6009336189102 + }, + "status": "completed", + "trace_sha256": "603b9e57429538d162af2ab5fd99c15611246ab92ab067b09acb4da3963d40e1" + } + ] + } + ], + "contract": { + "arrival": "open_loop_uniform", + "input_tokens": 2048, + "output_tokens": 1, + "prefix_caching": false, + "rates": [ + 20.0, + 24.0, + 28.0 + ], + "requests_per_anchor": 64, + "target_pass_rate": 0.95, + "ttft_slo_ms": 1256.0 + }, + "frontier": { + "git_head": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "git_status_short": " M frontier/config/config.py\n M frontier/entities/request.py\n M frontier/events/cluster_schedule_event.py\n M frontier/execution_time_predictor/sklearn_execution_time_predictor.py\n M frontier/metrics/constants.py\n M frontier/metrics/metrics_store.py\n M frontier/profiling/common/layers/rotary_embedding.py\n M frontier/profiling/moe/moe_impl.py\n M frontier/profiling/moe/moe_vllm_kernel.py\n M frontier/scheduler/cluster_scheduler/__init__.py\n M frontier/scheduler/cluster_scheduler/base_cluster_scheduler.py\n M frontier/scheduler/cluster_scheduler/cluster_scheduler_registry.py\n M frontier/scheduler/cluster_scheduler/sticky_lor_cluster_scheduler.py\n M frontier/scheduler/replica_scheduler/base_replica_scheduler.py\n M frontier/scheduler/replica_scheduler/vllm_v1_engine_replica_scheduler.py\n M frontier/scheduler/replica_stage_scheduler/replica_stage_schduler.py\n M frontier/simulator.py\n M frontier/types/cluster_scheduler_type.py\n?? data/profiling/compute/h20/\n?? frontier/scheduler/cluster_scheduler/prefix_lor_cluster_scheduler.py\n?? runs/\n?? tests/unit/test_attn_prefill_prediction_fallback.py\n", + "source": "/tmp/replayserve-frontier-rs1b" + }, + "profiles": { + "coverage": { + "attention": { + "1": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "2": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + }, + "4": { + "exact_prefill_2048_rows": 1, + "profile_batch_size": 1 + } + }, + "manifest": { + "attention_tp_coverage": [ + 1, + 2, + 4 + ], + "environment_contract": { + "dtype": "bfloat16", + "frontier_commit": "d9cfeb6d8791fbf2f295dd9744c56a666171776e", + "hardware": "NVIDIA H20", + "model": "Qwen3-30B-A3B", + "tensor_parallel_sizes": [ + 1, + 2, + 4 + ], + "vllm_source_commit": "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1", + "vllm_version": "0.20.0" + }, + "inputs": { + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp2-20260716-v1-dispatch-aware-20260716T140743025781Z/artifacts/artifacts/allreduce-full-tp2-v1/raw/allreduce-tp2.json": "97c3c76b5a04e95bd9192423c2b891667c668f39cc0dfecbd097d749939f2d0a", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-allreduce-full-tp4-20260716-v1-dispatch-aware-20260716T141106009788Z/artifacts/artifacts/allreduce-full-tp4-v1/raw/allreduce-tp4.json": "809df9baa6f468cf12bf0c99827475acc67894dd9f3f948976590b665fac0e76", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp1-20260716-v2-20260716T135132587012Z/artifacts/artifacts/flashattn-kv-full-v2-tp1/raw/flashattn-tp1.json": "dcb4c1bf7e76b9c765f78ddd2b8a734f2d7ba2adac13ce017689a8a77fe69a27", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp2-20260716-v2-20260716T135134194295Z/artifacts/artifacts/flashattn-kv-full-v2-tp2/raw/flashattn-tp2.json": "43ce042556ba887c8860614b43ccf0f564e5cebc1a0cffbce299d0acb9fa8d07", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-flashattn-kv-full-tp4-20260716-v2-20260716T135135197200Z/artifacts/artifacts/flashattn-kv-full-v2-tp4/raw/flashattn-tp4.json": "84eef31bcad0f556907a093318a420959d14fdc94474823d11f659704bdfec73", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-frontier-linear-full-20260716-v2-max-tokens-20260716T144444676943Z/artifacts/artifacts/frontier-linear-full-v2/profiles/compute/h20/qwen3-a3b-30b-moe/linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-moe-full-20260716-v1-local-shard-20260716T141334565164Z/artifacts/artifacts/moe-full-v1/raw/moe-full.json": "588f6ad0d69c9636d1b852e3df0a12d13cfe731f050ea7ec7aea457cceefbde8", + "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/fleet-artifacts/qwen30-vllm020-router-full-20260716-v3-tp-context-20260716T145446098505Z/artifacts/artifacts/router-full-v3/raw/router.json": "1962972e983bff3e06a721ef4ae4ec65728ff669681497a4a7e7f769b88b4931" + }, + "outputs": { + "allreduce.json": "b38d14f990578d668523d25b107aceed433da5020d8ada3b6e44d3562261a3b3", + "attention.csv": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "attention_true_mixed_fused.csv": "43ef4be90bddc9aeac6dbbe339feec24162cd1f2129a08fbd959e6ee4eaf5f60", + "linear_op.csv": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "moe.csv": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + }, + "profile_id": "qwen3-30b-a3b-bf16-vllm020-h20-tp1-2-4-fused-mixed-total-conserving", + "projection_contract": { + "allreduce": "Frozen exact runtime measurements; base profile-only comparison keeps the historical Frontier CC backend fixed to isolate compute profile fidelity", + "attention": "Pure prefill/extend/decode FA3 core plus separately measured KV update; input/output reshape assumed zero; exported mean is used as median target; true mixed rows use a total-conserving compatibility projection", + "attention_true_mixed": "The directly measured fused total is preserved in diagnostics. Frontier's two targets are projected by the same-TP pure prefill/decode reference ratio, with projected prefill + decode exactly equal to the fused total; the split is a schema compatibility attribution, not an observation", + "linear": "Frontier profiler using vLLM 0.20 CUDA operators", + "moe": "Replicated gate and fused top-k plus TP-local modular expert kernel; expert measurement already includes prepare/finalize so shuffling is zero" + }, + "row_counts": { + "allreduce": 24, + "attention_frontier_compatible": 132, + "attention_true_mixed_fused_diagnostic": 30, + "linear": 36, + "moe": 72 + }, + "schema_version": "frontier_qwen30_vllm020_frozen_profile.v2" + } + }, + "root": "/home/gahow/phd/aituner/runs/frontier-qwen30-vllm020-profile-v1/frozen/profile-v2", + "sha256": { + "attention": "76dcb767cebb4ec1c4e24bd04d93ddd48b5d271986ebfb51a197ab33e1b3d87d", + "linear": "67666cb0a4901b74599d468df2e31bcaa2a11a7842cc0cefba24ffce62508e0c", + "manifest": "af40545e75aff55c6333cd2d5379ccf042a5a0b7d7fc7df4f745ce256cb290eb", + "moe": "0e4dcba72918a1c4cf4e96ced31ee3829248a19ad54553cebef14417725808b0" + } + }, + "schema": "frontier-qwen30-prefill-surface-v1", + "status": "partial_not_decision_bearing" +} diff --git a/runs/frontier-phase-factorial-v0/test_phase_factorial.py b/runs/frontier-phase-factorial-v0/test_phase_factorial.py index 6e30aa5..4172926 100644 --- a/runs/frontier-phase-factorial-v0/test_phase_factorial.py +++ b/runs/frontier-phase-factorial-v0/test_phase_factorial.py @@ -37,3 +37,9 @@ def test_grid_and_trace(tmp_path: Path) -> None: assert len(lines) == 4 assert lines[1].split(",")[:3] == ["0.000000000000", "2048", "1"] assert lines[3].split(",")[:3] == ["0.500000000000", "2048", "1"] + + +def test_kendall_tau_b() -> None: + analysis = load("analyze_qwen30_prefill_fidelity.py") + assert analysis.kendall_tau_b([1, 2, 3], [1, 2, 3])["kendall_tau_b"] == 1 + assert analysis.kendall_tau_b([1, 2, 3], [3, 2, 1])["kendall_tau_b"] == -1 diff --git a/simulator-fidelity.md b/simulator-fidelity.md index 728ce14..b646d5b 100644 --- a/simulator-fidelity.md +++ b/simulator-fidelity.md @@ -1,6 +1,6 @@ # Frontier simulator fidelity：以 config ranking 为目标的阶段性评测 -更新日期：2026-07-16。 +更新日期：2026-07-17。 统一实验平台：所有新增与重跑实验只使用 `dash0` 的 8×NVIDIA H20。Qwen30B 的真实 P1 artifacts 也来自 `dash0`；早期文档中的主机 provenance 标注错误，本版已按实验主机和远端 artifact 路径更正。 @@ -14,11 +14,12 @@ - 将 operator profile 更新为与真实 serving 一致的 community vLLM `0.20.0`、BF16、H20、TP1/2/4 栈后，新的 profile-only Frontier **没有恢复排序**：92 个真实 anchors 在 simulator 中全部 SLO-infeasible，12 个 config 因而全部并列，最坏 tie-break regret 为 `60.91%`。这否证了“旧 profile 版本不一致是主要原因”这一简单解释，并把问题收敛到 execution context、operator composition 与 mixed-state schema。 - 在 Qwen3-235B-A22B-FP8 prefill-only 上，补齐 FP8/MoE profile 与 serving semantics、但不做端到端 action calibration 后，Frontier 的最优集合与真机完全一致，Spearman ρ 为 `0.9487`，20/20 个可比较非 tie config pair 同序，选择 regret 为 `0`。 - 在 Qwen3-235B-A22B-FP8 fixed-shape mixed 上，Frontier 同样给出了与真机完全相同的 TP4 top set，Kendall τ-b=`0.8944`、worst tie-break regret=`0`。但这次成功掩盖了一个明确的机制错误：Frontier 认为同一 TP family 内四个 MNS/MBT config 完全等价，真机 TP8 capacity 却形成对角为 0.30、非对角为 0.20 req/s/GPU 的 checkerboard interaction。34 个 config-load labels 中有 10 个 false-infeasible；20 个 real non-tie pairs 中只保持 16 个方向。 -- 现有结果表明：**在已对齐的受控 case 中，绝对 capacity 有 gap、甚至 action differential 错误，都不必然妨碍 Frontier 找到当前 surface 的最优 config。** 统一 dash0 campaign 已完整覆盖 Qwen235 FP8 prefill-only 与 fixed-shape mixed，但仍不能外推到 trace-faithful mixed 或 decode-only，也不能忽略达到这种 fidelity 所需的真机 profile、KV capacity、runtime-semantic patch 或 calibration 成本。 +- 新增的 Qwen3-30B-A3B BF16 prefill-only 对照在没有 decode、prefix reuse 和 true-mixed attention 的情况下仍然失败：真机 top set 是四个 TP4 config，Frontier top set 却是全部 TP1/TP2 config，top set 无交集，worst regret=`12.5%`、Kendall τ-b=`-1.0`，32 个 real non-tie pairs 全部反向。这直接否证了“prefill-only 是 simulator ranking 充分容易条件”。 +- 现有结果更支持一个 **margin-aware fidelity** 解释：action-differential residual 可以很大，但只有当它足以穿过 real decision margin 时才会导致选错。Qwen235 mixed 中真机 TP4 相对最好 TP8 的 margin 是 `2×`，足以掩盖同 TP family 内的机制错误；Qwen30 prefill-only 中 TP scaling 差分更细，Frontier 的 residual 直接反转了 topology ordering。模型大小或 execution phase 单独都不足以解释现有结果。 因此，现阶段最准确的 research statement 不是“simulator 排序一定错误”，而是： -> Simulator 的 usefulness 取决于 compatibility envelope。问题是需要多少、什么类型的真机信息，才能使其 config ranking 可被信任；以及当 workload、runtime 或 execution topology 变化时，如何低成本判断该 envelope 是否仍成立。 +> Simulator 的 usefulness 取决于 action-differential residual 是否小于真实 decision margin。问题不是简单区分 prefill 与 mixed，而是定位哪些 scheduler-state-conditioned execution residual 会随 TP、batch composition 和负载被放大，并用最少的真机信息判断 ranking 是否足够可信。 ## 评测口径 @@ -89,6 +90,22 @@ Frontier 原生 profile schema 可以按 `num_tensor_parallel_workers` 选择 TP 该 case 是有意设计的 prefill mechanism-isolation workload：output 固定为 1，prefix cache 关闭，并使用固定 64-request cohort。它不保持原 trace 的 output-length distribution 和 prefix reuse，因此不能被称为 trace-faithful workload，也不能代表 mixed serving。 +### Case D：Qwen3-30B-A3B BF16 prefill-only + +| 项目 | 设置 | +|---|---| +| 真机 | `dash0`，8×NVIDIA H20；每个 job 按 TP 独占 GPU | +| model/runtime | `Qwen/Qwen3-30B-A3B`；community vLLM `0.20.0+cu129`（source `88d34c6409e9fb3c7b8ca0c04756f061d2099eb1`），BF16 weights/activation/KV，FA3，runtime 默认 CUDA graph | +| workload | 64 个不同 token-chain prompts；ISL=2,048、OSL=1、uniform open-loop QPS、prefix cache off | +| config surface | `TP∈{1,2,4} × MNS∈{8,16,32,64}`，MBT=8,192，chunked prefill on，共 12 cells | +| simulator | Frontier commit `d9cfeb6d8791fbf2f295dd9744c56a666171776e` + 当前 compatibility patches + vLLM 0.20 same-stack `profile-v2`；无 E2E calibration、decode graph=`none`、analytical collective | +| load lattice | base system rates `{4,8,16,32,64}` req/s；事先声明的 common refinement 为 `{5,6,7}` req/s/GPU | +| real anchor | 每个 `(config,rate,round)` fresh server；两轮独立运行且第二轮反转顺序；两轮都 pass 才算 feasible | +| SLO | 至少 61/64 requests 满足 TTFT≤1,256 ms | +| primary objective | 最大被测试可行 offered req/s / 实际 TP GPUs | + +这个 case 是对“prefill-only 是 simulator 的容易区间”的直接可证伪测试。它刻意去掉 decode、true-mixed batch、prefix reuse 和 KV-residency feedback，但保留 TP 改变下的 local shard、collective、batching 与 scheduler queue。Frontier surface 在查看真机 ranking 前冻结，没有使用本 surface 的 serving latency 拟合 scale。 + ## Workload fidelity 修正 `thinking_w20260327_1000` 的 600 秒原始窗口包含 15,479 requests，字段包括 exact prompt、arrival、`input_length`、`output_length`、session parent/turn 和 block-size=64 的 `hash_ids`。只读审计得到： @@ -126,10 +143,11 @@ trace 的 source hash block size 为 64，而 community vLLM 0.10.2 的 CUDA KV | Qwen30 mixed / old profile-only | 12 | TP2,MNS32 | TP4,MNS32/64 | miss | τ-b=0.000；exact sign=37.88% | 25.63% | 排序错误 | | Qwen30 mixed / vLLM 0.20 profile-only | 12 | TP2,MNS32 | 全部 12 configs | 不可辨识 | τ-b=0.000；exact sign=7.58% | 60.91% | 同栈 raw profile 仍不足 | | Qwen30 mixed / frozen per-TP calibration | 12 | TP2,MNS32 | TP2,MNS32/64 | hit，非 exact | τ-b=0.9668；exact sign=93.94% | 0.76% | dash0 | +| Qwen30 BF16 prefill-only / vLLM 0.20 profile-only | 12 | 全部 TP4 configs | 全部 TP1/TP2 configs | miss，无交集 | τ-b=-1.0000；real non-tie=0/32 | 12.50% | topology order 反转 | | Qwen235 FP8 prefill / best-effort | 8 | TP4,MBT16K,MNS64/128 | 与 real 完全相同 | exact match | ρ=0.9487；non-tied 20/20 | 0 | 足以选最优 config | | Qwen235 FP8 fixed-shape mixed / frozen full profile | 8 | 四个 TP4 configs | 与 real 完全相同 | exact match | τ-b=0.8944；exact sign=24/28；real non-tie=16/20 | 0 | 选对 topology；漏掉 TP8 MNS×MBT interaction | -最重要的区别是：Qwen30 的 high-fidelity 结果依赖 per-TP E2E calibration；Qwen235 的 high-fidelity 结果不依赖本 case 的 E2E calibration，但依赖同 runtime/hardware 的 operator profile、真实 KV capacity 和多处 compatibility fixes。二者都不能表述为“拿 stock Frontier 零成本预测即可”。 +最重要的区别是：Qwen30 的 high-fidelity mixed 结果依赖 per-TP E2E calibration，而新 prefill-only profile-only 实验证明“去掉 decode/mixed 状态”也不足以恢复 ranking。Qwen235 的 high-fidelity 结果不依赖本 case 的 E2E calibration，但依赖同 runtime/hardware 的 operator profile、真实 KV capacity 和多处 compatibility fixes。两个 model 的 stack 和精度不同，所以不能把差异直接归因为 model size；二者也都不能表述为“拿 stock Frontier 零成本预测即可”。 ## Case A baseline 结果：Qwen30 mixed serving @@ -311,6 +329,53 @@ MNS 128 fail pass 原预注册的 TPOT 40 ms 因最低负载已不可行而失去 capacity-ranking 可辨识性；本文已把 150 ms 明确记录为 post-pilot protocol amendment，并保留 40 ms 全部失败的 sensitivity 结果，没有把 SLO change 隐藏成预注册结论。 +## Case D 结果：Qwen30 BF16 prefill-only + +![Qwen3-30B BF16 prefill-only simulator-vs-real config ranking](docs/assets/simulator-fidelity/qwen30-prefill-ranking.png) + +结果不是“绝对 capacity 有偏差但 rank 可用”，而是完整的 topology ordering reversal： + +| TP | MNS | Real | Frontier profile-only | 差异 | +|---:|---|---:|---:|---:| +| 1 | 8/16/32/64 | 7.0 | **8.0** | +1.0 | +| 2 | 8/16/32/64 | 7.0 | **8.0** | +1.0 | +| 4 | 8/16/32/64 | **8.0** | 6.0 | -2.0 | + +单位为最大被测试 SLO-feasible req/s/GPU。四个 TP4 configs 是真机 top set；Frontier 却把全部八个 TP1/TP2 configs 判为 top set，两者无交集。部署任一 simulator top config 都只能在真机上得到 7 req/s/GPU，因此 optimistic 与 worst tie-break regret 都为 `1-7/8=12.5%`。Kendall τ-b=`-1.0`；32 个 real non-tie pairs 中 0 个同序。96 个 config-load decisions 中 80 个一致，但同时有 8 个 false-feasible 和 8 个 false-infeasible，它们刚好跨过了不同 TP family 的 capacity boundary。 + +### 错误从哪里开始放大 + +代表性 `MNS=8` anchors 表明，低负载单次 execution time 并没有数倍偏差，错误主要在接近饱和时被 queue 放大： + +| Config / system rate | Real TTFT p95（两轮） | Frontier TTFT p95 | Real / sim SLO | +|---|---:|---:|---| +| TP1 @ 4 req/s | 154.1 / 153.6 ms | 171.6 ms | pass / pass | +| TP1 @ 8 req/s | 1289.2 / 1254.4 ms | 516.4 ms | fail / pass | +| TP2 @ 8 req/s | 97.7 / 93.8 ms | 122.2 ms | pass / pass | +| TP2 @ 16 req/s | 1389.9 / 1365.0 ms | 962.8 ms | fail / pass | +| TP4 @ 4 req/s | 82.3 / 69.6 ms | 93.1 ms | pass / pass | +| TP4 @ 32 req/s | 1136.6 / 1134.0 ms | 1787.9 ms | pass / fail | + +这排除了“只要给所有 kernel 乘一个 global scale 就能修好”的解释：相同 stack 在低负载下接近，但在不同 TP 的饱和边界上向相反方向偏移。更符合数据的抽象是：小的 per-step composition residual 改变 service rate，再通过 scheduler queue 的非线性反馈放大成 TTFT 和 capacity 边界错误。 + +当前证据可以排除一些原因，但还不能唯一定位根因： + +1. **Decode/true-mixed schema 不是必要条件。** 这个 workload 没有 decode、FULL decode CUDA graph、fused true-mixed attention 或 initial-KV state，排序仍然失败。这些机制可能在 mixed/decode 中进一步增大误差，但不是本次失败的前提。 +2. **Communication 可能解释部分 TP scaling，但不能单独解释全部误差。** `profile-v2/allreduce.json` 已有 24 个 TP2/TP4 实测 rows，但 base Frontier 为了保持历史对照使用 analytical 600-Gbps/1-µs collective，没有注入这些数据。下一步必须单独做 measured-collective ablation；但 TP1 无 collective 仍在 8 req/s 出现 773 ms 的 p95 低估，所以 collective 不会是唯一根因。 +3. **Profile 覆盖了 token count，却未必覆盖 scheduler 实际产生的 batch composition。** linear/MoE rows 覆盖到 8,192 tokens，但 exact-2,048 pure-prefill attention 在每个 TP 下只有 batch=1 的直接样本；实际 MBT=8,192 可以产生多个 2,048-token sequence 的组合。这是一个需要补测的 coverage hypothesis，不是已证明根因。 +4. **Routing、step composition 和 scheduler batch state 仍然是联合候选。** pure-prefill 只排除了 phase mixing，没有排除 MoE routing、TP-specific local shape、launch/fusion 和持续负载下 batch/queue trajectory 的耦合。 + +### 为什么 235B 能选对，30B 却选错 + +现有数据不支持“235B 大所以 simulator 更容易”这种模型大小因果。两个 case 的 precision、vLLM 版本、attention backend、graph mode、EP 和 profile closure 都不同。能被数据直接支持的区别是 **decision margin**： + +- 235B mixed 真机中 TP4 最优 capacity 为 0.60 req/s/GPU，最好 TP8 为 0.30，有 `2×` margin；Frontier 虽然漏掉 TP8 的 MNS×MBT interaction，仍预测 TP4/TP8 为 0.40/0.15，所以 residual 没有穿过全局拓扑边界。 +- 30B prefill-only 真机 TP4 相对 TP1/TP2 只有 8 vs. 7 req/s/GPU 的 margin。Frontier 预测为 6 vs. 8，TP-dependent residual 大于真实 margin，因而把最优 topology 完整反转。 + +因此，“prefill-only 容易，decode/mixed 困难”的强假设已被否证。phase 仍可能改变 residual 的幅度，但它不是 fidelity 的充分条件。按预注册 decision rule，此时不应继续无区分度地扩展更多 phase cases，而应在 Qwen30 同一 model/workload 上按单变量顺序做：`measured collective injection` → `batch-composition-conditioned pure-prefill attention/step profile` → 在 TP1@8、TP2@16、TP4@32 对齐 real/sim scheduler batch、queue 与 per-step critical path → 最后再测 routing/graph。 + +本 case 接受的 ground truth 共 24 个 fleet jobs、192 个 fresh-server anchors、12,288 个 measured requests 和 4,512 个 warmup requests，消耗 12.07 H20-GPU-hours。运行中曾因 fleet controller 在 fresh-server 空窗误判 GPU 为空闲而产生重叠 launch；这些 attempts 未被合并到 accepted artifact root，已整体隔离并用独立 queue state 重跑。最终 comparison SHA256 为 `c9a9cac9f60c7be804d1cb9466c455f8fe9e3a8dc60b9cec3329bde6a8c19334`。 + ## 尚不能纳入 simulator ranking 的 case ### 原 internal-runtime Qwen235 prefill-only @@ -328,11 +393,12 @@ MNS 128 fail pass 当前证据的结论是： 1. **绝对 gap 不是否决 simulator 的理由。** Qwen235 中 11%--50% 的 capacity 低估仍可保持 zero-regret config selection。 -2. **但 rank 成功不能证明模型机制正确。** Qwen235 prefill 的 TP8 MBT differential 错误，fixed-shape mixed 又漏掉 TP8 的 MNS×MBT 非加性交互；Qwen30 还有 28/92 anchor SLO labels 错误。只是当前 decision margin 足以容忍这些 residual。 -3. **alignment 是结果的一部分。** Qwen30 需要 per-TP E2E calibration 才能恢复排序；Qwen235 不需要本 case E2E calibration，但需要同栈 FP8 profiles、真机 KV capacity 和 simulator patches。论文必须报告这部分真机成本，不能把它隐藏在“offline profiles”中。 -4. **目前不能建立“simulator 全局选优普遍失败”的 premise。** 受控 prefill、fixed-shape mixed 和历史 mixed surface 上，经过 alignment 的 Frontier 已经能选到最优或 0.76% 内的近最优 config。更有证据的 research premise 是：aggregate selection success 会掩盖 action interaction 与 state-transition model 的错误；若要声称全局选优失败，仍必须在 dash0 的 trace-faithful mixed、decode-only、prefix cache 或跨 topology case 上得到稳定反例。 +2. **rank 成功不能证明模型机制正确，rank 失败则表明 residual 穿过了 decision margin。** Qwen235 prefill 的 TP8 MBT differential 错误，fixed-shape mixed 又漏掉 TP8 的 MNS×MBT 非加性交互，但大 topology margin 保住了 top set；Qwen30 prefill-only 的较小 margin 被 TP-dependent saturation residual 穿过，导致 τ-b=-1 和 12.5% regret。 +3. **prefill-only 不是 fidelity 的充分容易条件。** Qwen30 在没有 decode、prefix reuse、initial KV 和 true-mixed batch 时仍选错 topology。phase 可以改变 residual，但不能单独作为 compatibility-envelope 边界。 +4. **alignment 是结果的一部分。** Qwen30 mixed 需要 per-TP E2E calibration 才能恢复排序，同栈 operator profile 或去掉 mixed phase 都不足；Qwen235 不需要本 case E2E calibration，但需要同栈 FP8 profiles、真机 KV capacity 和 simulator patches。论文必须报告这部分真机成本，不能把它隐藏在“offline profiles”中。 +5. **现在有了可复现的全局选优反例，但尚不足以声称 simulator 普遍失败。** Qwen30 BF16 prefill-only 是一个无 top-set overlap 的稳定反例；Qwen235 两个 case 又证明 Frontier 在某些 envelope 内足以选优。更有研究价值的 premise 是：**什么 state-conditioned action residual 决定 ranking 是否能跨过 real margin，以及如何在不做全 surface 真机 sweep 的情况下检测这个风险。** -后续每个 case 建议使用同一 gate：worst selected-config regret ≤5%、tie-aware rank correlation ≥0.8、足够数量的 informative pairs、ground-truth bracket 不足以反转最优决策，并单独报告达到该结果所需的 real-GPU profiling/calibration cost。 +后续每个 case 建议使用同一 gate：worst selected-config regret ≤5%、tie-aware rank correlation ≥0.8、足够数量的 informative pairs、ground-truth bracket 不足以反转最优决策，并单独报告达到该结果所需的 real-GPU profiling/calibration cost。当前下一 gate 是在 Qwen30 prefill-only 上做同模型单变量 context ablation，而不是继续扩展跨模型 phase 矩阵。 ## 数据与复现 @@ -340,6 +406,10 @@ MNS 128 fail pass - 画图脚本：[plot_simulator_fidelity.py](scripts/plot_simulator_fidelity.py) - Qwen30 audit：[report.md](runs/frontier-multicase-sufficiency-v0/results/qwen30-baseline/report.md) - Qwen30 aligned metrics：[metrics.json](runs/frontier-slo-alignment-v0/results/metrics.json) +- Qwen30 prefill-only experiment card：[experiment-card.md](runs/frontier-phase-factorial-v0/experiment-card.md) +- Qwen30 prefill-only comparison：[comparison.json](runs/frontier-phase-factorial-v0/results/final/comparison.json) +- Qwen30 prefill-only capacity table：[capacity.csv](runs/frontier-phase-factorial-v0/results/final/capacity.csv) +- Qwen30 prefill-only analyzer：[analyze_qwen30_prefill_fidelity.py](runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py) - Qwen235 fixed-cohort comparison：[v2_refined_comparison.json](runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/v2_refined_comparison.json) - Qwen235 full report：[report.md](runs/frontier-multicase-sufficiency-v0/best_effort/fixed_cohort_evidence/report.md) - Fixed-shape mixed comparison：[comparison.json](runs/frontier-multicase-sufficiency-v1/results/t0-final/comparison.json) @@ -357,4 +427,17 @@ MNS 128 fail pass ```bash python3 scripts/plot_simulator_fidelity.py + +python3 runs/frontier-phase-factorial-v0/analyze_qwen30_prefill_fidelity.py \ + --fleet-artifacts runs/frontier-phase-factorial-v0/fleet-artifacts-exclusive \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp1/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp2/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-tp4/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp1/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp2/frontier_surface_frozen.json \ + --simulator-manifest runs/frontier-phase-factorial-v0/simulator-refine-tp4/frontier_surface_frozen.json \ + --output-root runs/frontier-phase-factorial-v0/results/final + +cp runs/frontier-phase-factorial-v0/results/final/qwen30-prefill-ranking.png \ + docs/assets/simulator-fidelity/qwen30-prefill-ranking.png ```