[Klaud Cold] Update qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg SGLang image to nightly-dev-cu13-20260922-4cbf290f / 将 qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg 的 SGLang 镜像更新至 nightly-dev-cu13-20260922-4cbf290f

[Klaud-Cold]

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 to lmsysorg/sglang:nightly-dev-cu13-20260922-4cbf290f@sha256:3ee16325ce408cf4c39fbe89e25a72b564bd7070683257d2c0e99d00ee5e4748.
Baseline: 2026-09-21 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
Mean latency · Sources: API 1, API 2

Point	Total tok/s/GPU ↑	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c4 PTP4x1/DTP4x1 2a653d	3,453.74	37.52	393.39	2.28
AgentX c12 PTP4x1/DTP4x1 66532b	3,699.77	33.46	26,161.95	16.25
AgentX c24 PTP4x1/DTP4x1 5010d8	17,582.7	147.54	619.44	3.39
AgentX c32 PTP2x1/DTP2x1 cabbcf	38,693.69	342.82	1,370.92	6
AgentX c32 PTP4x1/DTP4x1 04656f	20,623.98	184.41	708.11	3.83
AgentX c40 PTP2x1/DTP2x1 cef85a	44,703.25	423.68	1,465.97	7.26
AgentX c44 PTP2x1/DTP2x1 4744bc	46,713.82	470.17	1,893.85	7.9
AgentX c48 PTP2x1/DTP2x1 f3b1b5	5,466.51	45.89	228,254.34	57.98
AgentX c56 PTP2x1/DTP2x1 b4f7f5	57,140.43	595.62	2,679.55	9.82

Eval: N/A

中文

**目标：**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 更新为 lmsysorg/sglang:nightly-dev-cu13-20260922-4cbf290f@sha256:3ee16325ce408cf4c39fbe89e25a72b564bd7070683257d2c0e99d00ee5e4748。
**基线：**2026-09-21 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
平均延迟 · 来源： API 1, API 2；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 35804549054 / attempt 1 · 2026-09-23 01:09 UTC
lmsysorg/sglang:nightly-dev-cu13-20260922-4cbf290f@sha256:3ee16325ce408cf4c39fbe89e25a72b564bd7070683257d2c0e99d00ee5e4748 · c8b8d5a30bdc · Mean latency
Change: Measured PR head 4efb43d bumps the master image and all nine referenced srt-slurm recipes from SGLang nightly 20518d85 to nightly 4cbf290f (295 commits, every recipe flag still resolves, explicit --hicache-size keeps #40135 auto-sizing off, #39973 permits INTRA_NODE_NVLINK), pinned to the linux/amd64 manifest digest; no recipe flags change.

Point	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
AgentX c4 PTP4x1/DTP4x1 2a653d	N/A	N/A	N/A
AgentX c12 PTP4x1/DTP4x1 66532b	N/A	N/A	N/A
AgentX c24 PTP4x1/DTP4x1 5010d8	N/A	N/A	N/A
AgentX c32 PTP4x1/DTP4x1 04656f	N/A	N/A	N/A
AgentX c32 PTP2x1/DTP2x1 cabbcf	N/A	N/A	N/A
AgentX c40 PTP2x1/DTP2x1 cef85a	N/A	N/A	N/A
AgentX c44 PTP2x1/DTP2x1 4744bc	N/A	N/A	N/A
AgentX c48 PTP2x1/DTP2x1 f3b1b5	N/A	N/A	N/A
AgentX c56 PTP2x1/DTP2x1 b4f7f5	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 canonical smoke generator emits eval-all-concs on multi-node agentic eval-only rows, which the workflow schema forbids, so this family needs a maintainer fix to shared tooling before a retry.

中文

初次尝试 · 失败 · Run 35804549054 / attempt 1 · 2026-09-23 01:09 UTC
lmsysorg/sglang:nightly-dev-cu13-20260922-4cbf290f@sha256:3ee16325ce408cf4c39fbe89e25a72b564bd7070683257d2c0e99d00ee5e4748 · c8b8d5a30bdc · 平均延迟
**变更：**实测 PR head 4efb43d 将 master 镜像及其引用的九个 srt-slurm 配方从 SGLang nightly 20518d85 更新为 nightly 4cbf290f（共 295 个提交，docker/Dockerfile 完全一致，所有配方参数仍可解析，显式 --hicache-size 使 #40135 的自动调整保持关闭，#39973 允许 INTRA_NODE_NVLINK），并固定到 linux/amd64 manifest 摘要；配方参数不变。实测数值及异常说明见上表。
**下一步：**停止：规范 smoke 生成器会在多节点 agentic 仅评测行上输出 eval-all-concs，而工作流 schema 禁止该字段，需维护者修复共享工具后方可重试。

[Klaud-Cold]

failed · Cleanup pending. Stop and confirm owned runs before closing.

中文

failed · 清理待完成。先停止并确认自有运行结束，再关闭 PR。

[Klaud-Cold]

failed · Repairs: 0 · Runs: 35804549054
All owned runs ended. PR closed; branch deleted for retry.

中文

failed · 修复次数：0 · 运行：35804549054
所有自有运行均已结束。PR 已关闭；分支已删除，可重新尝试。