Skip to content

[Klaud Cold] Update qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg SGLang image to v0.5.22-cu130 / [Klaud Cold] 将 qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg 的 SGLang 镜像更新至 v0.5.22-cu130 - #4224

Merged
adibarra merged 6 commits into
mainfrom
klaud/auto-3a7743c5a4d1dde4-abb39e882feee984
Oct 21, 2026
Merged

adibarra merged 6 commits into
mainfrom
klaud/auto-3a7743c5a4d1dde4-abb39e882feee984

Conversation

@Klaud-Cold

@Klaud-Cold Klaud-Cold commented Oct 17, 2026 •

Copy link
Copy Markdown
Collaborator

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 to lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae.
Baseline: 2026-09-21 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
Mean latency · Sources: API 1, API 2, API 3

Point Total tok/s/GPU ↑ Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c4 PTP4x1/DTP4x1 6570f4 3,453.74 37.52 393.39 2.28
AgentX c12 PTP4x1/DTP4x1 fe42ff 3,699.77 33.46 26,161.95 16.25
AgentX c24 PTP4x1/DTP4x1 9b4a3e 17,582.7 147.54 619.44 3.39
AgentX c32 PTP2x1/DTP2x1 addfc6 38,693.69 342.82 1,370.92 6
AgentX c32 PTP4x1/DTP4x1 a13987 20,623.98 184.41 708.11 3.83
AgentX c40 PTP2x1/DTP2x1 b59a02 44,703.25 423.68 1,465.97 7.26
AgentX c44 PTP2x1/DTP2x1 9ffba9 46,713.82 470.17 1,893.85 7.9
AgentX c48 PTP2x1/DTP2x1 14e1a2 5,466.51 45.89 228,254.34 57.98
AgentX c56 PTP2x1/DTP2x1 e2eb15 57,140.43 595.62 2,679.55 9.82
Eval Score ↑ Samples
gsm8k/em_strict · c4 PTP4EP1/DTP4EP1 97.73% 1,319
gsm8k/em_strict · c12 PTP4EP1/DTP4EP1 97.42% 1,319
gsm8k/em_strict · c24 PTP4EP1/DTP4EP1 97.73% 1,319
gsm8k/em_strict · c32 PTP2EP2/DTP2EP2 97.57% 1,319
gsm8k/em_strict · c32 PTP4EP4/DTP4EP1 97.95% 1,319
gsm8k/em_strict · c40 PTP2EP2/DTP2EP2 97.8% 1,319
gsm8k/em_strict · c44 PTP2EP2/DTP2EP2 97.73% 1,319
gsm8k/em_strict · c48 PTP2EP2/DTP2EP2 97.8% 1,319
gsm8k/em_strict · c56 PTP2EP2/DTP2EP2 97.73% 1,319
中文

**目标:**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85 更新至 lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae。
**基线:**2026-09-21 · lmsysorg/sglang:nightly-dev-cu13-20260918-20518d85
平均延迟 · 来源: API 1, API 2, API 3;数值及异常说明见上表。

…u130

将 qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg 的 SGLang 镜像更新至
v0.5.22-cu130,并按摘要固定。

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 17, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Passed · Run 38989245325 / attempt 1 · 2026-10-17 03:06 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 62175beb470d · Mean latency
Change: At measured ref 51e0d1f, pin the v0.5.22-cu130 digest (SGLang commit baa9befa, 1604 commits after 20518d85) with the recipe unchanged, since every SGLang arg and choice in disagg-variants.yaml is still declared in arg_groups/fields at baa9befa, --disable-cuda-graph remains a deprecated alias, and SGLANG_ENABLE_SPEC_V2 was already a removed no-op at 20518d85; Dynamo 805a77f integration and kernels are unverified until this smoke run.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c4 PTP4x1/DTP4x1 6570f4 38.22 (N/A) 380.83 (N/A) 2.34 (N/A)
AgentX c12 PTP4x1/DTP4x1 fe42ff 84.98 (N/A) 494.62 (N/A) 2.79 (N/A)
AgentX c24 PTP4x1/DTP4x1 9b4a3e 147.1 (N/A) 638.45 (N/A) 3.85 (N/A)
AgentX c32 PTP2x1/DTP2x1 addfc6 346.63 (N/A) 1,268.69 (N/A) 5.68 (N/A)
AgentX c32 PTP4x1/DTP4x1 a13987 183.97 (N/A) 743.96 (N/A) 3.98 (N/A)
AgentX c40 PTP2x1/DTP2x1 b59a02 427.12 (N/A) 1,340.58 (N/A) 7.65 (N/A)
AgentX c44 PTP2x1/DTP2x1 9ffba9 485.6 (N/A) 1,597.23 (N/A) 7.28 (N/A)
AgentX c48 PTP2x1/DTP2x1 14e1a2 508.53 (N/A) 1,949.17 (N/A) 8.43 (N/A)
AgentX c56 PTP2x1/DTP2x1 e2eb15 602.81 (N/A) 2,840.59 (N/A) 10.96 (N/A)

Note: All rows: Δ N/A: dataset/scenario/statistic mismatch.

Eval Score ↑ Samples
gsm8k/em_strict · c24 PTP4EP1/DTP4EP1 97.65% (-0.08 pp) 1,319 each
gsm8k/em_strict · c32 PTP2EP2/DTP2EP2 97.73% (+0.15 pp) 1,319 each
gsm8k/em_strict · c32 PTP4EP4/DTP4EP1 97.8% (-0.15 pp) 1,319 each
gsm8k/em_strict · c40 PTP2EP2/DTP2EP2 97.95% (+0.15 pp) 1,319 each
gsm8k/em_strict · c44 PTP2EP2/DTP2EP2 97.8% (+0.08 pp) 1,319 each
gsm8k/em_strict · c48 PTP2EP2/DTP2EP2 97.95% (+0.15 pp) 1,319 each
gsm8k/em_strict · c56 PTP2EP2/DTP2EP2 97.95% (+0.23 pp) 1,319 each

Next: Append the perf-changelog entry, then run the full fail-fast sweep on the exact head.

中文

初次尝试 · 已通过 · Run 38989245325 / attempt 1 · 2026-10-17 03:06 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 62175beb470d · 平均延迟
**变更:**在测量提交 51e0d1f 上,固定 v0.5.22-cu130 摘要(SGLang 提交 baa9befa,位于 20518d85 之后 1604 个提交),配方保持不变,因为 disagg-variants.yaml 中的所有 SGLang 参数及取值在 baa9befa 的 arg_groups/fields 中仍有声明,--disable-cuda-graph 仍为已弃用别名,且 SGLANG_ENABLE_SPEC_V2 在 20518d85 时已是被移除的空操作;Dynamo 805a77f 集成与内核需本次 smoke 运行验证。实测数值及异常说明见上表。
**下一步:**追加 perf-changelog 条目,然后在该提交上运行完整的 fail-fast 扫描。

github-actions Bot and others added 2 commits October 17, 2026 03:06
合并 main 分支。

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…mage bump

为 qwen3.5-fp8-b300-dynamo-sglang-agentic-disagg 的 SGLang 镜像更新添加
perf-changelog 条目。

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Klaud-Cold Klaud-Cold added the full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended) label Oct 17, 2026
@Klaud-Cold

Klaud-Cold commented Oct 17, 2026 •

Copy link
Copy Markdown
Collaborator Author

Final full sweep · Failed · Run 38997725541 / attempt 1 · 2026-10-17 03:27 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · b8e528b7a4db
Change: Validate the complete updated-image family at this head, with main merged in and the perf-changelog entry appended.

Next: Let the remaining owned jobs finish, then rerun the failed and cancelled jobs on the same run and head as an infrastructure retry.

中文

最终完整 sweep · 失败 · Run 38997725541 / attempt 1 · 2026-10-17 03:27 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · b8e528b7a4db
**变更:**在合并 main 并追加 perf-changelog 条目后,于此提交验证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:**等待其余自有任务结束,然后作为基础设施重试,在同一运行和提交上重跑失败及被取消的任务。

@github-actions

Copy link
Copy Markdown
Contributor

@Klaud-Cold

Klaud-Cold commented Oct 17, 2026 •

Copy link
Copy Markdown
Collaborator Author

Final full sweep · Passed · Run 38997725541 / attempt 2 · 2026-10-17 05:39 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · b8e528b7a4db · Mean latency
Change: Validate the complete updated-image family at this head.

Point Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
AgentX c4 PTP4x1/DTP4x1 6570f4 37.6 (N/A) 391 (N/A) 2.36 (N/A)
AgentX c12 PTP4x1/DTP4x1 fe42ff 84.9 (N/A) 494.19 (N/A) 2.79 (N/A)
AgentX c24 PTP4x1/DTP4x1 9b4a3e 146.51 (N/A) 638.3 (N/A) 3.66 (N/A)
AgentX c32 PTP2x1/DTP2x1 addfc6 354.2 (N/A) 1,325.46 (N/A) 5.39 (N/A)
AgentX c32 PTP4x1/DTP4x1 a13987 183.8 (N/A) 747.16 (N/A) 4.06 (N/A)
AgentX c40 PTP2x1/DTP2x1 b59a02 424.47 (N/A) 1,612.93 (N/A) 7.35 (N/A)
AgentX c44 PTP2x1/DTP2x1 9ffba9 473.76 (N/A) 1,646.86 (N/A) 7.85 (N/A)
AgentX c48 PTP2x1/DTP2x1 14e1a2 507.94 (N/A) 1,918.36 (N/A) 8.4 (N/A)
AgentX c56 PTP2x1/DTP2x1 e2eb15 567.76 (N/A) 2,669.07 (N/A) 12.24 (N/A)

Note: All rows: Δ N/A: dataset/scenario/statistic mismatch.

Eval Score ↑ Samples
gsm8k/em_strict · c24 97.88% (+0.15 pp) 1,319 each
gsm8k/em_strict · c32 97.5% (-0.45 pp) 1,319 each
gsm8k/em_strict · c32 97.65% (+0.08 pp) 1,319 each
gsm8k/em_strict · c40 97.8% (+0.00 pp) 1,319 each
gsm8k/em_strict · c44 97.88% (+0.15 pp) 1,319 each
gsm8k/em_strict · c48 97.35% (-0.45 pp) 1,319 each
gsm8k/em_strict · c56 98.03% (+0.30 pp) 1,319 each

Next: Mark ready for maintainer review.

中文

最终完整 sweep · 已通过 · Run 38997725541 / attempt 2 · 2026-10-17 05:39 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · b8e528b7a4db · 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
Klaud-Cold marked this pull request as ready for review October 17, 2026 05:39
@Klaud-Cold
Klaud-Cold requested a review from a team October 17, 2026 05:39
@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

validated · Repairs: 0 · Runs: 38989245325, 38997684119, 38997725541
All owned runs ended. Full sweep verified; ready for review.

中文

validated · 修复次数:0 · 运行:38989245325, 38997684119, 38997725541
所有自有运行均已结束。完整 sweep 已验证;已就绪,等待审查。

@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

/use 38997725541

@claude

claude Bot commented Oct 17, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @Klaud-Cold's task in 13s —— View job


LGTM - no blocking issues found

@github-actions

github-actions Bot commented Oct 17, 2026 •

Copy link
Copy Markdown
Contributor

@Klaud-Cold staged run 38997725541.

This run remains available across future /stage-results requests. Staging the same run ID again updates its staged data. Staging workflow

@claude claude Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code review found no issues

No high-confidence issues detected in this change.

Resolve perf-changelog.yaml by taking main's file and re-appending this PR's entry.

中文:合并 origin/main;perf-changelog.yaml 采用 main 版本,并在末尾重新追加本 PR 的条目。
Resolve conflicts: take main's refactored recipes and re-apply this PR's non-image edits; take main's perf-changelog.yaml and re-append this PR's entry.

中文:合并 origin/main;配方采用 main 重构后的版本并重新应用本 PR 的非镜像改动,perf-changelog.yaml 采用 main 版本并在末尾重新追加本 PR 的条目。
@adibarra

Copy link
Copy Markdown
Collaborator

/reuse-sweep-run 38997725541

@adibarra
adibarra merged commit 195d135 into main Oct 21, 2026
26 checks passed
@adibarra
adibarra deleted the klaud/auto-3a7743c5a4d1dde4-abb39e882feee984 branch October 21, 2026 19:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

full-sweep-fail-fast Full sweep with canary gate; first failure cancels the rest of that matrix (recommended)

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

3 participants