[Klaud Cold] Update qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp to SGLang nightly-dev-cu13-20261001-b37e6f79 / [Klaud Cold] 将 qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp 更新至 SGLang nightly-dev-cu13-20261001-b37e6f79

[Klaud-Cold]

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 to lmsysorg/sglang:nightly-dev-cu13-20261001-b37e6f79@sha256:65a8507b99dc1a6ebfc72344f29cbc017bffd5972ff53bc2d5f43aa293c88a84.
Baseline: 2026-09-19 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
AgentX · PTP4x1/DTP4x1 · Mean latency · Sources: API 1, API 2, API 3, API 4

Concurrency	Total tok/s/GPU ↑	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
8	5,531.21	54.96	426.57	2.42
16	12,419.78	106.07	571.48	3.09
24	17,815.92	149.66	731.12	3.19
32	20,703.06	185.19	934.1	3.59
48	27,982.94	283.37	1,401.29	5.18
64	34,825.43	377.64	3,370.29	7.37

Eval	Score ↑	Samples
gsm8k/em_strict · c8	97.57%	1,319
gsm8k/em_strict · c16	97.8%	1,319
gsm8k/em_strict · c24	97.8%	1,319
gsm8k/em_strict · c32	97.42%	1,319
gsm8k/em_strict · c48	97.88%	1,319
gsm8k/em_strict · c64	97.95%	1,319

中文

**目标：**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 更新为 lmsysorg/sglang:nightly-dev-cu13-20261001-b37e6f79@sha256:65a8507b99dc1a6ebfc72344f29cbc017bffd5972ff53bc2d5f43aa293c88a84。
**基线：**2026-09-19 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
AgentX · PTP4x1/DTP4x1 · 平均延迟 · 来源： API 1, API 2, API 3, API 4；数值及异常说明见上表。

[github-actions]

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 only run on labeled PRs. Add full-sweep-fail-fast (strongly recommended); use full-sweep-enabled only when matrix jobs should continue after a failure.
• 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：**扫描仅在带有标签的 PR 上运行。强烈建议添加 full-sweep-fail-fast；仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。
• **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]

Initial attempt · Failed · Run 36864791393 / attempt 1 · 2026-10-01 12:55 UTC
lmsysorg/sglang:nightly-dev-cu13-20261001-b37e6f79@sha256:65a8507b99dc1a6ebfc72344f29cbc017bffd5972ff53bc2d5f43aa293c88a84 · 1c5e6b9f5e7b · AgentX · PTP4x1/DTP4x1 · Mean latency
Change: Measure ref 0ae428f: pin SGLang nightly-dev-cu13-20261001-b37e6f79 by digest in the master config and six recipes; source diff from 20518d85 (sgl-project/sglang@20518d8...b37e6f7) keeps every used flag, the SGLANG_ENABLE_SPEC_V2 env stays a no-op, and flashinfer 0.7.0.post1 / Mooncake 0.3.13.post1 are the coupled bumps; the run stopped in get-jobs because the smoke matrix fails benchmark_schema validation (eval-all-concs on agentic eval-only entries), reproduced on unmodified main.

Concurrency	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
8	N/A	N/A	N/A
16	N/A	N/A	N/A
24	N/A	N/A	N/A
32	N/A	N/A	N/A
48	N/A	N/A	N/A
64	N/A	N/A	N/A

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

Next: Stop: the required smoke matrix for this family fails shared schema validation on main, which is outside this candidate's edit scope.

中文

初次尝试 · 失败 · Run 36864791393 / attempt 1 · 2026-10-01 12:55 UTC
lmsysorg/sglang:nightly-dev-cu13-20261001-b37e6f79@sha256:65a8507b99dc1a6ebfc72344f29cbc017bffd5972ff53bc2d5f43aa293c88a84 · 1c5e6b9f5e7b · AgentX · PTP4x1/DTP4x1 · 平均延迟
**变更：**测量提交 0ae428f：在主配置和六个配方中按 digest 固定 SGLang nightly-dev-cu13-20261001-b37e6f79；与 20518d85 的源码对比（sgl-project/sglang@20518d8...b37e6f79）显示所用参数均保留，SGLANG_ENABLE_SPEC_V2 仍为无效环境变量，耦合依赖升级为 flashinfer 0.7.0.post1 与 Mooncake 0.3.13.post1；运行在 get-jobs 阶段停止，原因是冒烟矩阵未通过 benchmark_schema 校验（agentic 仅评测条目带有 eval-all-concs），在未修改的 main 上可复现。实测数值及异常说明见上表。
**下一步：**停止：该配置族所需的冒烟矩阵在 main 上未通过共享 schema 校验，超出本候选的修改范围。

[Klaud-Cold]

readiness-blocked · Cleanup pending. Stop and confirm owned runs before closing.

中文

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

[Klaud-Cold]

readiness-blocked · Repairs: 0 · Runs: 36864791393
All owned runs ended. PR closed; branch deleted for retry.

中文

readiness-blocked · 修复次数：0 · 运行：36864791393
所有自有运行均已结束。PR 已关闭；分支已删除，可重新尝试。