[Klaud Cold] Update qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp SGLang image to v0.5.21-cu130 / [Klaud Cold] 将 qwen3.5-fp8-b200-dynamo-sglang-agentic-disagg-mtp 的 SGLang 镜像更新为 v0.5.21-cu130

[Klaud-Cold]

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 to lmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469.
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:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469。
**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 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]

Initial attempt · Failed · Run 38191257011 / attempt 1 · 2026-10-12 01:18 UTC
lmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469 · c801c9f2dbbd · AgentX · PTP4x1/DTP4x1 · Mean latency
Change: Pin the official SGLang v0.5.21 CUDA 13 image (commit e00930c, which contains old nightly 20518d85; FlashInfer 0.6.18, sglang-kernel 0.4.7, Mooncake 0.3.13 and torch 2.13.0 unchanged) with no recipe flag changes, since every used flag still parses and SGLANG_ENABLE_SPEC_V2 was already a no-op (sgl-project/sglang#40795).

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 is invalid for multi-node AgentX families until shared tooling is fixed.

中文

初次尝试 · 失败 · Run 38191257011 / attempt 1 · 2026-10-12 01:18 UTC
lmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469 · c801c9f2dbbd · AgentX · PTP4x1/DTP4x1 · 平均延迟
**变更：**固定官方 SGLang v0.5.21 CUDA 13 镜像（提交 e00930c，包含旧 nightly 20518d85；FlashInfer 0.6.18、sglang-kernel 0.4.7、Mooncake 0.3.13 与 torch 2.13.0 均未变），配方参数不变，所有参数仍可解析，SGLANG_ENABLE_SPEC_V2 此前已为空操作（sgl-project/sglang#40795）。实测数值及异常说明见上表。
**下一步：**停止：在共享工具修复之前，所需的冒烟矩阵对多节点 AgentX 配置族无效。

[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: 38191257011
All owned runs ended. PR closed; branch deleted for retry.

中文

failed · 修复次数：0 · 运行：38191257011
所有自有运行均已结束。PR 已关闭；分支已删除，可重新尝试。