Skip to content

[Klaud Cold] Update GLM-5.2 FP4 GB200 Dynamo SGLang disagg AgentX image to nightly-dev-20261021-c1c287a4 / [Klaud Cold] 将 GLM-5.2 FP4 GB200 Dynamo SGLang 分离式 AgentX 镜像更新至 nightly-dev-20261021-c1c287a4 - #4447

Closed
Klaud-Cold wants to merge 1 commit into
mainfrom
klaud/auto-e9686b8790045c65-0dd16aeb42d3dd5b

Conversation

@Klaud-Cold

@Klaud-Cold Klaud-Cold commented Oct 23, 2026 •

Copy link
Copy Markdown
Collaborator

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-20261020-ab580b2e to lmsysorg/sglang:nightly-dev-20261021-c1c287a4.
Baseline: 2026-10-23 · lmsysorg/sglang:nightly-dev-20261020-ab580b2e@sha256:0af411bb320a95bb5e1c58b4c20c9852920a19aa9f12f148f69731b0dca3b576
Mean latency · Sources: API 1, API 2, API 3, API 4

Point Total tok/s/GPU ↑ Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 8,389.72 49.92 969.8 4.88
AgentX c48 PTP1x1/DTP4x4 c4eef4 14,033.17 91.68 1,316.09 5.17
AgentX c64 PTP1x1/DTP8x2 a39e5c 17,885.41 126.39 1,458.46 7.32
AgentX c96 PTP1x1/DTP8x1 aaa2d3 32,056.64 253.75 2,307.03 8.86
AgentX c96 PTP1x1/DTP8x2 a74110 22,692.08 175.77 2,125.03 7.9
AgentX c224 PTP1x2/DTP16x1 18fb54 37,666.39 264.42 2,586.71 9.34
AgentX c336 PTP1x3/DTP16x1 f06e2e 42,272.42 312.16 3,192.43 10.27
Eval Score ↑ Samples
gsm8k/em_strict · c36 96.74% 1,319
gsm8k/em_strict · c48 97.19% 1,319
gsm8k/em_strict · c64 96.97% 1,319
gsm8k/em_strict · c224 97.12% 1,319
gsm8k/em_strict · c336 96.82% 1,319

Stop: Add the klaud-abort label at any time to stop this candidate now; Klaud cancels its runs, closes this PR and does not retry it. Remove the label to let the planner pick it up again.

中文

**目标:**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-20261020-ab580b2e 更新至 lmsysorg/sglang:nightly-dev-20261021-c1c287a4。
**基线:**2026-10-23 · lmsysorg/sglang:nightly-dev-20261020-ab580b2e@sha256:0af411bb320a95bb5e1c58b4c20c9852920a19aa9f12f148f69731b0dca3b576
平均延迟 · 来源: API 1, API 2, API 3, API 4;数值及异常说明见上表。
**停止:**随时添加 klaud-abort 标签即可立即停止该候选:Klaud 会取消其运行、关闭此 PR 且不再重试。移除该标签后,规划器会在下次扫描时重新考虑它。

…261021-c1c287a4

将 GLM-5.2 FP4 GB200 Dynamo SGLang 分离式配置的镜像更新至
nightly-dev-20261021-c1c287a4,并同步配方中的 SGLang 版本标识。

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

Thanks for the contribution!

  • Review: If this PR changes files owned by someone other than a repository admin or @SemiAnalysisAI/core, ask one eligible CODEOWNER to complete the latest PR_REVIEW_CHECKLIST.md before contacting a core maintainer on Slack. Follow the template exactly, including As a PR reviewer and CODEOWNER, I have reviewed this and have, so sign-off verification triggers.
  • PR verification: Sweeps run only when the PR appends an inferencex-e2e/perf-changelog.yaml entry and carries exactly one primary label: full-sweep-fail-fast (strongly recommended; canary plus per-matrix fail-fast), full-sweep-enabled (canary; matrix jobs continue after a failure), or non-canary-full-sweep-enabled (no canary or fail-fast). The modifiers all-evals, evals-only, and agentx-fast require a primary label. On fork PRs, a maintainer applies the label. See sweep labels and reuse.
  • After merging: PR authors must ensure all GitHub Actions jobs pass. Transient failures often pass on rerun; see how to rerun failed jobs.
中文

感谢你的贡献!

  • **审阅:**如果 PR 修改的文件归属于仓库管理员及 @SemiAnalysisAI/core 之外的 CODEOWNER,请先联系一位有资格的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md,再通过 Slack 联系核心维护者。必须严格遵循模板,并保留 As a PR reviewer and CODEOWNER, I have reviewed this and have,才能触发签核验证。
  • **PR 验证:**只有当 PR 在 inferencex-e2e/perf-changelog.yaml 末尾追加了条目,并且恰好带有一个主标签时,才会运行扫描:full-sweep-fail-fast(强烈推荐;canary 加逐矩阵 fail-fast)、full-sweep-enabled(有 canary;矩阵任务在失败后继续运行)或 non-canary-full-sweep-enabled(无 canary,也无 fail-fast)。修饰标签 all-evals、evals-only 和 agentx-fast 必须与主标签一起使用。fork PR 的标签由维护者添加。参见扫描标签与复用。
  • **合并后:**PR 作者必须确保所有 GitHub Actions 任务通过。临时性失败通常可以通过重新运行恢复;参见重新运行失败任务的说明。

@Klaud-Cold

Klaud-Cold commented Oct 23, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 1 · 2026-10-23 19:27 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c N/A N/A N/A
AgentX c96 PTP1x1/DTP8x1 aaa2d3 N/A N/A N/A
AgentX c96 PTP1x1/DTP8x2 a74110 N/A N/A N/A
AgentX c224 PTP1x2/DTP16x1 18fb54 N/A N/A N/A
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: All rows: pending.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.

Next: Servers start on the new image: the c48, c64 and c96 (2D) eval jobs passed; waiting for the remaining 4 evals and 7 benchmarks in run 40005721369.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 1 · 2026-10-23 19:27 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**新镜像上服务已正常启动:c48、c64 与 c96(2D)评测任务已通过;正在等待运行 40005721369 中其余 4 个评测和 7 个基准。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Queued · Run 40005721369 / attempt 2 · 2026-10-23 20:03 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c N/A N/A N/A
AgentX c96 PTP1x1/DTP8x1 aaa2d3 N/A N/A N/A
AgentX c96 PTP1x1/DTP8x2 a74110 N/A N/A N/A
AgentX c224 PTP1x2/DTP16x1 18fb54 N/A N/A N/A
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: All rows: pending.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.

Next: inferencex-dash[bot] cancelled run 40005721369 attempt 1 (preemption, no failure; c48/c64/c96-2D evals had passed); reran the cancelled jobs as attempt 2 on the same head and waiting.

中文

初次尝试 · 排队中 · Run 40005721369 / attempt 2 · 2026-10-23 20:03 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**inferencex-dash[bot] 取消了运行 40005721369 的第 1 次尝试(抢占,无失败;c48/c64/c96-2D 评测已通过);已在同一 head 上以第 2 次尝试重跑被取消任务,正在等待。

@Klaud-Cold

Klaud-Cold commented Oct 23, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 3 · 2026-10-24 09:37 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c N/A N/A N/A
AgentX c96 PTP1x1/DTP8x1 aaa2d3 N/A N/A N/A
AgentX c96 PTP1x1/DTP8x2 a74110 N/A N/A N/A
AgentX c224 PTP1x2/DTP16x1 18fb54 N/A N/A N/A
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: All rows: cancelled.
Note: All rows: request errors unavailable.
Note: All rows: Δ N/A: point unavailable.

Next: Slurm job 30270 (c96 PTP1x1/DTP8x1) was externally cancelled (uid 1010) 53 min into a healthy profile with 0 request errors, and fail-fast cancelled the sibling benchmarks: preemption, so I will rerun the failed and cancelled jobs on the same head once the 4 remaining evals finish.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 3 · 2026-10-24 09:37 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**Slurm 作业 30270(c96 PTP1x1/DTP8x1)在正常 profiling 53 分钟、0 请求错误时被外部取消(uid 1010),fail-fast 随后取消了其余基准:属于抢占;待其余 4 个评测结束后在同一 head 上重跑失败与被取消的任务。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

resumed · The session ended before finishing; session 2 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束;第 2 个会话将在其规划运行的新运行中继续。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

resumed · The session ended before finishing; session 3 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束;第 3 个会话将在其规划运行的新运行中继续。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 4 · 2026-10-24 13:24 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 N/A N/A N/A
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c224 PTP1x2/DTP16x1 18fb54: cancelled.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c224 PTP1x2/DTP16x1 18fb54, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c224 PTP1x2/DTP16x1 18fb54, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.
Note: AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: pending.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: inferencex-dash[bot] cancelled the queued c36 and c224 jobs at 13:22 UTC (preemption, no failure); c336 and c48 are still running, after which I will rerun the cancelled jobs on the same head.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 4 · 2026-10-24 13:24 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**inferencex-dash[bot] 于 13:22 UTC 取消了排队中的 c36 与 c224 任务(抢占,无失败);c336 与 c48 仍在运行,结束后我将在同一 head 上重跑被取消的任务。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

resumed · The session ended before finishing; session 4 continues it in a new run of its planner.

中文

resumed · 会话未完成即已结束;第 4 个会话将在其规划运行的新运行中继续。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 5 · 2026-10-24 15:30 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 260.07 (-1.6%) 2,644.68 (+2.2%) 9.5 (+1.7%)
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c336 PTP1x3/DTP16x1 f06e2e: cancelled.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.
Note: AgentX c48 PTP1x1/DTP4x4 c4eef4: pending.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: Slurm job 30313 (c336 PTP1x3/DTP16x1) was externally cancelled (uid 1010) at 15:26 UTC during healthy FlashInfer autotune with no server error, and fail-fast cancelled c36: preemption, so I will rerun both on the same head once c48 finishes.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 5 · 2026-10-24 15:30 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**Slurm 作业 30313(c336 PTP1x3/DTP16x1)于 15:26 UTC 在正常 FlashInfer autotune 期间被外部取消(uid 1010),无服务端错误,fail-fast 随后取消了 c36:属于抢占;待 c48 结束后在同一 head 上重跑两者。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 6 · 2026-10-24 15:46 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 260.07 (-1.6%) 2,644.68 (+2.2%) 9.5 (+1.7%)
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772: pending.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.
Note: AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: cancelled.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: Slurm job 30315 (c48 PTP1x1/DTP4x4) was externally cancelled (uid 1010) at 15:42 UTC while weights were still loading, with no server error, and fail-fast cancelled c336: preemption; c36 is running, after which I will rerun c48 and c336 on the same head.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 6 · 2026-10-24 15:46 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**Slurm 作业 30315(c48 PTP1x1/DTP4x4)于 15:42 UTC 在权重加载期间被外部取消(uid 1010),无服务端错误,fail-fast 随后取消了 c336:属于抢占;c36 正在运行,结束后我将在同一 head 上重跑 c48 与 c336。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 7 · 2026-10-24 16:30 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 260.07 (-1.6%) 2,644.68 (+2.2%) 9.5 (+1.7%)
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4: cancelled.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.
Note: AgentX c336 PTP1x3/DTP16x1 f06e2e: pending.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: Slurm job 30320 (c48 PTP1x1/DTP4x4) was externally cancelled (uid 1010) at 16:26 UTC 21 min into a healthy profile with 0 request errors, and fail-fast cancelled c36: preemption; c336 is still running, after which I will rerun c36 and c48 on the same head.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 7 · 2026-10-24 16:30 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**Slurm 作业 30320(c48 PTP1x1/DTP4x4)于 16:26 UTC 在正常 profiling 21 分钟、0 请求错误时被外部取消(uid 1010),fail-fast 随后取消了 c36:属于抢占;c336 仍在运行,结束后我将在同一 head 上重跑 c36 与 c48。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Running · Run 40005721369 / attempt 8 · 2026-10-24 16:49 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 260.07 (-1.6%) 2,644.68 (+2.2%) 9.5 (+1.7%)
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772: failed.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.
Note: AgentX c48 PTP1x1/DTP4x4 c4eef4: pending.
Note: AgentX c336 PTP1x3/DTP16x1 f06e2e: cancelled.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: Attempt 8: c36 decode worker 0 rank 1 on watchtower-navy-cn09 hit CUDA unspecified launch failure in MoE autotune while identically configured workers 1, 3 and 4 passed it and this topology started cleanly in attempt 3 (node fault likely, no Xid captured); after c48 ends I will rerun c36 and c336 on the same head as infrastructure retry 1.

中文

初次尝试 · 运行中 · Run 40005721369 / attempt 8 · 2026-10-24 16:49 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**第 8 次尝试:c36 的 decode worker 0 rank 1(watchtower-navy-cn09)在 MoE autotune 中出现 CUDA unspecified launch failure,而配置相同的 worker 1、3、4 均已通过,且该拓扑在第 3 次尝试中正常启动(疑似节点故障,未捕获 Xid);待 c48 结束后在同一 head 上以基础设施重试 1 重跑 c36 与 c336。

@Klaud-Cold

Klaud-Cold commented Oct 24, 2026 •

Copy link
Copy Markdown
Collaborator Author

Infrastructure retry 1/2 · Running · Run 40005721369 / attempt 9 · 2026-10-24 17:32 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · Mean latency
Change: Measure ref 247fe1a: pin nightly-dev-20261021-c1c287a4 by digest; recipe identity sglang 0.0.0.dev1+gc1c287a42. Flags, Dynamo 1.6.0.dev20261019, flashinfer 0.7.2, sgl-kernel 0.4.9, CUDA 13.0.3 unchanged. Source ab580b2e..c1c287a4 (sgl-project/sglang@ab580b2...c1c287a): server_args.py identical, all recipe flags parse; changes on this path: NVFP4 MoE w13 global-scale order for flashinfer_trtllm (#44858, #45289), draft MoE out of expert placement (#45299), SGLANG_ENABLE_PREFILL_WAR_READ_DONE default on (#37958). Effect on GB200 accuracy/speed unproven until measured.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c36 PTP1x1/DTP4x6 538772 N/A N/A N/A
AgentX c48 PTP1x1/DTP4x4 c4eef4 N/A N/A N/A
AgentX c64 PTP1x1/DTP8x2 a39e5c 126.75 (+0.3%) 1,452.16 (-0.4%) 7.05 (-3.7%)
AgentX c96 PTP1x1/DTP8x1 aaa2d3 254.43 (+0.3%) 2,362.93 (+2.4%) 8.9 (+0.5%)
AgentX c96 PTP1x1/DTP8x2 a74110 176.56 (+0.4%) 2,114.31 (-0.5%) 7.78 (-1.5%)
AgentX c224 PTP1x2/DTP16x1 18fb54 260.07 (-1.6%) 2,644.68 (+2.2%) 9.5 (+1.7%)
AgentX c336 PTP1x3/DTP16x1 f06e2e N/A N/A N/A

Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: pending.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: request errors unavailable.
Note: AgentX c36 PTP1x1/DTP4x6 538772, AgentX c48 PTP1x1/DTP4x4 c4eef4, AgentX c336 PTP1x3/DTP16x1 f06e2e: Δ N/A: point unavailable.

Eval Score ↑ Samples
gsm8k/em_strict · c96 PTP1x1/DTP8x2 96.66% (N/A) 1,319
gsm8k/em_strict · c48 PTP1x1/DTP4x4 97.35% (N/A) 1,319
gsm8k/em_strict · c64 PTP1x1/DTP8x2 96.74% (N/A) 1,319
gsm8k/em_strict · c336 PTP1x3/DTP16x1 96.89% (N/A) 1,319
gsm8k/em_strict · c96 PTP1x1/DTP8x1 96.59% (N/A) 1,319
gsm8k/em_strict · c36 PTP1x1/DTP4x6 96.51% (N/A) 1,319
gsm8k/em_strict · c224 PTP1x2/DTP16x1 96.74% (N/A) 1,319

Note: All rows: Δ N/A: no matched eval baseline.

Next: Infrastructure retry 1 (run 40005721369 attempt 9): c48 and c336 running on gb200-nv since about 17:00 UTC, c36 queued as of 17:32 UTC Oct 24; waiting.

中文

基础设施重试 1/2 · 运行中 · Run 40005721369 / attempt 9 · 2026-10-24 17:32 UTC
lmsysorg/sglang:nightly-dev-20261021-c1c287a4@sha256:b0473834763b2753e069cb96c1acaf235ba7fd632ccd9bbeb44da809c437fdeb · 7afd3b8564f4 · 平均延迟
**变更:**测量 ref 247fe1a:按摘要固定 nightly-dev-20261021-c1c287a4;配方 identity 为 sglang 0.0.0.dev1+gc1c287a42。参数、Dynamo 1.6.0.dev20261019、flashinfer 0.7.2、sgl-kernel 0.4.9、CUDA 13.0.3 不变。源码 ab580b2e..c1c287a4(sgl-project/sglang@ab580b2...c1c287a42928):server_args.py 相同,配方参数均可解析;此路径相关变化:flashinfer_trtllm 的 NVFP4 MoE w13 全局缩放顺序(#44858、#45289)、草稿 MoE 不参与专家放置(#45299)、SGLANG_ENABLE_PREFILL_WAR_READ_DONE 默认开启(#37958)。对 GB200 精度与速度的影响需实测。实测数值及异常说明见上表。
**下一步:**基础设施重试 1(运行 40005721369 第 9 次尝试):截至 10 月 24 日 17:32 UTC,gb200-nv 上 c48 与 c336 自约 17:00 UTC 起运行,c36 排队中;正在等待。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

unexpected-error · Cleanup pending. Stop and confirm owned runs before closing.

中文

unexpected-error · 清理待完成。先停止并确认自有运行结束,再关闭 PR。

@Klaud-Cold Klaud-Cold closed this Oct 25, 2026
@Klaud-Cold
Klaud-Cold deleted the klaud/auto-e9686b8790045c65-0dd16aeb42d3dd5b branch October 25, 2026 00:46
@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

unexpected-error · Repairs: unknown · Runs: 40005721369
All owned runs ended. PR closed; branch deleted for retry.

中文

unexpected-error · 修复次数:未知 · 运行:40005721369
所有自有运行均已结束。PR 已关闭;分支已删除,可重新尝试。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

1 participant