[Klaud-Cold]

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-20261026-8252e3af to lmsysorg/sglang:nightly-dev-20261027-5884e4fe.
Baseline: 2026-10-27 · lmsysorg/sglang:nightly-dev-20261026-8252e3af@sha256:b4e4abb9e1dc038f7e14c85805be97e4f74abb46e541bd4edd31401d50a2e593
AgentX · TP4/EP4 · Mean latency · Sources: API 1, API 2, API 3, API 4

Concurrency	Total tok/s/GPU ↑	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
1	5,613.11	38.28	1,078.13	2.52
2	5,895.94	44.35	622.03	2.72
4	8,186.26	56.41	1,429.7	2.68
6	10,491.55	79.12	729.52	3.27
8	13,333.73	98.31	2,516.49	3.23
10	16,944.25	125.05	738.97	3.86
16	25,942.58	168.06	775.14	5.16
20	29,756.6	209.27	2,293.36	5.96
24	33,383.74	221.24	4,295.75	6.39
28	37,129.15	247.56	5,133.12	7.73
32	38,877.17	247.05	7,319.85	8.59

Eval	Score ↑	Samples
gsm8k/em_strict · c32	97.57%	1,319

Stop: Close this PR at any time to stop this candidate now; Klaud cancels its runs and deletes its branch.

中文

**目标：**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-20261026-8252e3af 更新为 lmsysorg/sglang:nightly-dev-20261027-5884e4fe。
**基线：**2026-10-27 · lmsysorg/sglang:nightly-dev-20261026-8252e3af@sha256:b4e4abb9e1dc038f7e14c85805be97e4f74abb46e541bd4edd31401d50a2e593
AgentX · TP4/EP4 · 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：**随时关闭此 PR 即可立即停止该候选：Klaud 会取消其运行并删除其分支。

[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 · Passed · Run 40607684753 / attempt 1 · 2026-10-28 01:48 UTC
lmsysorg/sglang:nightly-dev-20261027-5884e4fe@sha256:61b9083bdcb7cf3741366f487cb996c05ea33ad2ddddc820d10cbc1c0cfb0a86 · 79a44b9079e8 · AgentX · TP4/EP4 · Mean latency
Change: Bump the master image from SGLang 8252e3af to 5884e4fe (74 upstream commits); the source diff (sgl-project/sglang@8252e3a...5884e4f) renames or removes none of the recipe's flags, and its DFlash changes are gated on new draft-config fields, so the recipe is unchanged; measured ref dc2131f.

Concurrency	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
1	38.24 (N/A)	1,065.92 (N/A)	2.51 (N/A)
2	46.18 (N/A)	607.09 (N/A)	2.75 (N/A)

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

Eval	Score ↑	Samples
gsm8k/em_strict · c1	97.88% (N/A)	1,319
gsm8k/em_strict · c6	97.42% (N/A)	1,319

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

Next: Append the perf-changelog entry, merge main and dispatch the full AgentX final sweep with full-sweep-fail-fast.

中文

初次尝试 · 已通过 · Run 40607684753 / attempt 1 · 2026-10-28 01:48 UTC
lmsysorg/sglang:nightly-dev-20261027-5884e4fe@sha256:61b9083bdcb7cf3741366f487cb996c05ea33ad2ddddc820d10cbc1c0cfb0a86 · 79a44b9079e8 · AgentX · TP4/EP4 · 平均延迟
**变更：**将主镜像从 SGLang 8252e3af 更新到 5884e4fe（74 个上游提交）；源码差异（sgl-project/sglang@8252e3a...5884e4fe）未重命名或移除 recipe 使用的任何参数，DFlash 改动仅在新的草稿配置字段下生效，因此 recipe 保持不变；测量 ref 为 dc2131f。实测数值及异常说明见上表。
**下一步：**追加 perf-changelog 条目，合并 main，并以 full-sweep-fail-fast 发起完整的 AgentX 最终扫描。

[Klaud-Cold]

parked · Session 1 is waiting on 40607684753. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 1 个会话正在等待 40607684753。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

[Klaud-Cold]

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]

Final full sweep · Passed · Run 40617278994 / attempt 1 · 2026-10-28 03:48 UTC
lmsysorg/sglang:nightly-dev-20261027-5884e4fe@sha256:61b9083bdcb7cf3741366f487cb996c05ea33ad2ddddc820d10cbc1c0cfb0a86 · 481f54cc97a5 · AgentX · TP4/EP4 · Mean latency
Change: Validate the complete updated-image family at this head.

Concurrency	Output tok/s/GPU ↑	TTFT ms ↓	TPOT ms ↓
1	38.4 (+0.3%)	1,061.46 (-1.5%)	2.51 (-0.4%)
2	45.95 (+3.6%)	641.15 (+3.1%)	2.72 (+0%)
4	56.26 (-0.3%)	1,232.58 (-13.8%)	2.65 (-1.1%)
6	79.1 (-0%)	689.36 (-5.5%)	3.23 (-1.2%)
8	102.59 (+4.4%)	1,846.13 (-26.6%)	3.28 (+1.5%)
10	125.52 (+0.4%)	690.44 (-6.6%)	3.86 (+0%)
16	168.84 (+0.5%)	781.16 (+0.8%)	5.19 (+0.6%)
20	207.78 (-0.7%)	2,315.33 (+1%)	6.04 (+1.3%)
24	220.99 (-0.1%)	4,154.84 (-3.3%)	6.33 (-0.9%)
28	237.28 (-4.2%)	6,817.69 (+32.8%)	7.1 (-8.2%)
32	248.45 (+0.6%)	7,138.64 (-2.5%)	8.47 (-1.4%)

Note: c32: 1 request error.

Eval	Score ↑	Samples
gsm8k/em_strict · c1	97.73% (N/A)	1,319
gsm8k/em_strict · c6	97.65% (N/A)	1,319
gsm8k/em_strict · c8	97.5% (N/A)	1,319
gsm8k/em_strict · c24	97.88% (N/A)	1,319
gsm8k/em_strict · c28	97.73% (N/A)	1,319
gsm8k/em_strict · c32	97.65% (+0.08 pp)	1,319 each

Note: gsm8k/em_strict · c1, gsm8k/em_strict · c6, gsm8k/em_strict · c8, gsm8k/em_strict · c24, gsm8k/em_strict · c28: Δ N/A: no matched eval baseline.

Next: Mark ready for maintainer review.

中文

最终完整 sweep · 已通过 · Run 40617278994 / attempt 1 · 2026-10-28 03:48 UTC
lmsysorg/sglang:nightly-dev-20261027-5884e4fe@sha256:61b9083bdcb7cf3741366f487cb996c05ea33ad2ddddc820d10cbc1c0cfb0a86 · 481f54cc97a5 · AgentX · TP4/EP4 · 平均延迟
**变更：**验证此提交更新镜像后的完整配置族。实测数值及异常说明见上表。
**下一步：**标记为就绪，等待维护者审查。

[Klaud-Cold]

parked · Session 2 is waiting on 40617278994. Recovery reruns preempted runs and starts the next session once they end.

中文

parked · 第 2 个会话正在等待 40617278994。恢复流程会重跑被抢占的运行，并在运行结束后启动下一个会话。

[github-actions]

View unofficial run (performance): https://inferencex.semianalysis.com/inference?unofficialRun=40617278994

View unofficial run (accuracy): https://inferencex.semianalysis.com/evaluation?unofficialRun=40617278994

[Klaud-Cold]

validated · Repairs: unknown · Runs: 40607684753, 40617258962, 40617278994
All owned runs ended. Full sweep verified; ready for review.

中文

validated · 修复次数：未知 · 运行：40607684753, 40617258962, 40617278994
所有自有运行均已结束。完整 sweep 已验证；已就绪，等待审查。

[claude]

Code review found no issues

No high-confidence issues detected in this change.