Skip to content

[Klaud Cold] Update DeepSeek-V4.1-Flash FP4 H100 SGLang AgentX DPA image to v0.5.22-cu130 / [Klaud Cold] 将 DeepSeek-V4.1-Flash FP4 H100 SGLang AgentX DPA 镜像更新至 v0.5.22-cu130 - #4324

Merged
adibarra merged 5 commits into
mainfrom
klaud/auto-cd8687954dc9dc75-adea5e4c0ce8ed67
Oct 21, 2026
Merged

adibarra merged 5 commits into
mainfrom
klaud/auto-cd8687954dc9dc75-adea5e4c0ce8ed67

Conversation

@Klaud-Cold

@Klaud-Cold Klaud-Cold commented Oct 19, 2026 •

Copy link
Copy Markdown
Collaborator

Goal: Update SGLang image from lmsysorg/sglang:nightly-dev-cu13-20260922-582389ce@sha256:0e1b14e302619a42ef5581b87db6804651d4f946301cf543d74a3a1eb1c33b40 to lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae.
Baseline: 2026-09-24 · lmsysorg/sglang:nightly-dev-cu13-20260922-582389ce@sha256:0e1b14e302619a42ef5581b87db6804651d4f946301cf543d74a3a1eb1c33b40
AgentX · TP8/EP8 · Mean latency · Sources: API 1, API 2

Concurrency Total tok/s/GPU ↑ Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
4 709.32 3.46 64,605.05 15.79
8 973.15 6.98 71,563.38 23.39
16 1,876.38 13.76 83,479.64 23.75
20 1,102.78 8.36 195,648.76 26.74

Eval: N/A

Stop: Add the klaud-abort label at any time to stop this candidate now; Klaud cancels its runs, closes this PR and does not retry it. Remove the label to let the planner pick it up again.

中文

**目标:**将 SGLang 镜像从 lmsysorg/sglang:nightly-dev-cu13-20260922-582389ce@sha256:0e1b14e302619a42ef5581b87db6804651d4f946301cf543d74a3a1eb1c33b40 更新至 lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae。
**基线:**2026-09-24 · lmsysorg/sglang:nightly-dev-cu13-20260922-582389ce@sha256:0e1b14e302619a42ef5581b87db6804651d4f946301cf543d74a3a1eb1c33b40
AgentX · TP8/EP8 · 平均延迟 · 来源: API 1, API 2;数值及异常说明见上表。
**停止:**随时添加 klaud-abort 标签即可立即停止该候选:Klaud 会取消其运行、关闭此 PR 且不再重试。移除该标签后,规划器会在下次扫描时重新考虑它。

将 dsv41flash-fp4-h100-sglang-agentic-dspark-dpa 的 SGLang 镜像从
nightly-dev-cu13-20260922-582389ce 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 19, 2026 •

Copy link
Copy Markdown
Collaborator Author

Initial attempt · Passed · Run 39305089312 / attempt 1 · 2026-10-19 23:01 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 0e5b07c78319 · AgentX · TP8/EP8 · Mean latency
Change: Measure PR commit 769a003: pin SGLang v0.5.22-cu130 (sgl-project/sglang baa9befa = main 684fea8a plus two release cherry-picks, a descendant of old nightly 582389ce) by digest with no recipe change; the server_args/arg_groups diff keeps every recipe flag and env var, and deprecated --enable-dp-attention now resolves to attn_dp_size=8, dp_size=1, which the bundled sgl-model-gateway counts as 8 DP ranks.

Concurrency Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
4 13.9 (N/A) 16,396 (N/A) 8.05 (N/A)
8 15.1 (N/A) 31,624.67 (N/A) 10.31 (N/A)
16 18.24 (N/A) 78,780.49 (N/A) 13.04 (N/A)
20 19.67 (N/A) 80,305.04 (N/A) 13.2 (N/A)

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

Eval Score ↑ Samples
gsm8k/em_strict · c20 97.27% (N/A) N/A/1,319 (old/new)

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

Next: Smoke benchmarks and the gsm8k eval passed; append the perf-changelog entry and run the full sweep.

中文

初次尝试 · 已通过 · Run 39305089312 / attempt 1 · 2026-10-19 23:01 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 0e5b07c78319 · AgentX · TP8/EP8 · 平均延迟
**变更:**测量 PR 提交 769a003:按摘要固定 SGLang v0.5.22-cu130(sgl-project/sglang baa9befa,即 main 684fea8a 加两个发布 cherry-pick,是旧 nightly 582389ce 的后代),配方不变;server_args/arg_groups 对比显示配方的全部参数和环境变量仍然保留,已弃用的 --enable-dp-attention 现解析为 attn_dp_size=8、dp_size=1,随附的 sgl-model-gateway 将其计为 8 个 DP rank。实测数值及异常说明见上表。
**下一步:**冒烟基准测试和 gsm8k 评测均已通过;追加 perf-changelog 条目并运行完整 sweep。

github-actions Bot and others added 2 commits October 19, 2026 23:02
合并 origin/main。

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

为 DeepSeek-V4.1-Flash H100 SGLang DPA 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 19, 2026
@Klaud-Cold

Klaud-Cold commented Oct 19, 2026 •

Copy link
Copy Markdown
Collaborator Author

Final full sweep · Passed · Run 39339355047 / attempt 1 · 2026-10-20 03:39 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 32753f81a43a · AgentX · TP8/EP8 · Mean latency
Change: Validate the complete updated-image family at this head.

Concurrency Output tok/s/GPU ↑ TTFT ms ↓ TPOT ms ↓
4 14.83 (N/A) 13,751.61 (N/A) 8.17 (N/A)
8 18.66 (N/A) 24,657.47 (N/A) 10.56 (N/A)
16 15.96 (N/A) 91,473.97 (N/A) 12.86 (N/A)
20 21.56 (N/A) 81,159.97 (N/A) 12.41 (N/A)

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

Eval Score ↑ Samples
gsm8k/em_strict · c20 97.19% (N/A) N/A/1,319 (old/new)

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

Next: Mark ready for maintainer review.

中文

最终完整 sweep · 已通过 · Run 39339355047 / attempt 1 · 2026-10-20 03:39 UTC
lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae · 32753f81a43a · AgentX · TP8/EP8 · 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

Copy link
Copy Markdown
Collaborator Author

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。

@github-actions

Copy link
Copy Markdown
Contributor

@Klaud-Cold
Klaud-Cold marked this pull request as ready for review October 20, 2026 03:39
@Klaud-Cold
Klaud-Cold requested a review from a team October 20, 2026 03:39
@Klaud-Cold

Copy link
Copy Markdown
Collaborator Author

validated · Repairs: 0 · Runs: 39305089312, 39339285261, 39339355047
All owned runs ended. Full sweep verified; ready for review.

中文

validated · 修复次数:0 · 运行:39305089312, 39339285261, 39339355047
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 39339355047

@claude

claude Bot commented Oct 20, 2026 •

Copy link
Copy Markdown
Contributor

Claude finished @Klaud-Cold's task in 16s —— View job


LGTM, no blocking issues found.

  • Image: The pinned image is a public Docker Hub image with a digest (lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f060…).
  • Changelog: The perf-changelog.yaml entry is at the end of the file and covers the one changed config key, dsv41flash-fp4-h100-sglang-agentic-dspark-dpa.
  • Sibling config: dsv41flash-fp4-h100-sglang-agentic-dspark (the non-DPA config) still uses the nightly 20260922-582389ce image. That looks intentional, since this PR only changes the DPA config.

@github-actions

github-actions Bot commented Oct 20, 2026 •

Copy link
Copy Markdown
Contributor

@Klaud-Cold staged run 39339355047.

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 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 39339355047

@nvpohanh

Copy link
Copy Markdown
Collaborator

[by Claude Code] The SGLang cookbook update of this PR's AgentX config (dsv41flash-fp4-h100-sglang-agentic-dspark-dpa) will happen in sgl-project/sglang#46807, in the Agentic Long-Context Deployment section of the DeepSeek-V4.1 cookbook page. The config is already live in that section; sgl-project/sglang#46807 refreshes it with the pin below.

The commands mirror the launch measured in run 39339355047, pinned by this PR's /reuse-sweep-run comment.

@nvpohanh

Copy link
Copy Markdown
Collaborator

[by Claude Code] sgl-project/sglang#46807 has been merged. This PR's AgentX config (dsv41flash-fp4-h100-sglang-agentic-dspark-dpa) is now in the Agentic Long-Context Deployment section of the SGLang DeepSeek-V4.1 cookbook: https://docs.sglang.io/cookbook/autoregressive/DeepSeek/DeepSeek-V4_1#agentic-long-context-deployment

The commands mirror the launch measured in run 39339355047, pinned by this PR's /reuse-sweep-run comment.

@nvpohanh

Copy link
Copy Markdown
Collaborator

[by Claude Code] The SGLang cookbook update of this PR's AgentX config (dsv41flash-fp4-h100-sglang-agentic-dspark-dpa) will happen in sgl-project/sglang#46948, in the Agentic Long-Context Deployment section of the DeepSeek-V4.1 cookbook page. The config has been live in that section since sgl-project/sglang#46807; sgl-project/sglang#46948 refreshes it with the pin below.

The commands mirror the launch measured in run 39339355047, pinned by this PR's /reuse-sweep-run comment.

@nvpohanh

Copy link
Copy Markdown
Collaborator

[by Claude Code] sgl-project/sglang#46948 has been merged. This PR's AgentX config (dsv41flash-fp4-h100-sglang-agentic-dspark-dpa) is now in the Agentic Long-Context Deployment section of the SGLang DeepSeek-V4.1 cookbook: https://docs.sglang.io/cookbook/autoregressive/DeepSeek/DeepSeek-V4_1#agentic-long-context-deployment

The commands mirror the launch measured in run 39339355047, pinned by this PR's /reuse-sweep-run comment.

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.

4 participants