[Klaud Cold] Update qwen3.5-fp4-b300-sglang-agentic-mtp SGLang image to nightly-dev-cu13-20260907-30705c00 and move to cluster:b300-dsxe - #2882
Conversation
…to nightly-dev-cu13-20260907-30705c00 and move to cluster:b300-dsxe Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
fe11256 to
25d9e08
Compare
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
3 similar comments
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Looks good, straightforward config change. Verified the diff only touches the image and runner fields of qwen3.5-fp4-b300-sglang-agentic-mtp (search-space grids, model, precision, framework untouched), the perf-changelog.yaml entry is a pure append at the tail matching the required convention, and the referenced benchmark script (benchmarks/single_node/agentic/qwen3.5_fp4_b300_sglang_mtp.sh) exists unchanged.
Extended reasoning...
Overview
The diff is confined to two files: configs/nvidia-master.yaml (a two-line scalar edit — image and runner for the qwen3.5-fp4-b300-sglang-agentic-mtp recipe) and perf-changelog.yaml (one new append-only block). No script, code, or CI logic is touched.
Security risks
None. This is a pure data/config change: a Docker image tag string and a cluster label string. No secrets, no auth, no injectable input. The dated nightly tag is a dev image rather than a numbered release, which carries some build-stability risk, but that is an operational/perf-testing concern rather than a security one, and the PR description states the tag was verified against Docker Hub with digest/push-date info matching sibling recipes already bumped in merged PRs.
Level of scrutiny
Low. This matches the repo's explicit approval criteria for "simple, mechanical... config tweaks, version bumps... follow established patterns with no ambiguity." I verified directly: (1) the nvidia-master.yaml hunk changes only image and runner, leaving the TP4 and TP2/EP2 search-space grids, model, precision, and framework identical to before; (2) the perf-changelog.yaml addition is appended strictly after the prior entry with no edits to historical bytes, satisfying the repo's append-only/byte-sensitive changelog invariant; (3) the referenced benchmark script path exists in the tree unchanged. I also checked that cluster:b300-dsxe and the retirement narrative for cluster:b300-nv are plausible given many other recipes in the same file still reference cluster:b300-nv (consistent with a rolling, PR-by-PR migration rather than a single cutover) — the sibling glm5.2-fp4-b300-sglang-agentic-mtp recipe is still on cluster:b300-nv in this checkout, which is a minor discrepancy against the PR description's claim that a sibling PR (#2829) already made the same move, but it does not affect the correctness of this PR's own change and is not something this diff needs to fix.
Other factors
No CODEOWNERS/ANTOWNERS restriction applies to config/changelog files here, and the bug-hunting system reported zero findings. The pr-link: PRLINK_PLACEHOLDER is a known, expected placeholder per CONTRIBUTING.md's documented pattern (pr-link: .../pull/XXX) that gets filled once the PR number is known, not a defect.
This review covers commit fe11256, which is no longer the latest commit on this pull request; later commits are not covered by it.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=34177497130 |
1 similar comment
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=34177497130 |
1 participant
Summary
Update the SGLang image for
qwen3.5-fp4-b300-sglang-agentic-mtpfromlmsysorg/sglang:v0.5.16-cu130tolmsysorg/sglang:nightly-dev-cu13-20260907-30705c00, and move the recipe from the retiredcluster:b300-nvrunner tocluster:b300-dsxe.sha256:19b8fa1223cc339c1eae7a5b703f1a8c2543b5b119155bf3d7efaef18f77f007; same tag as the other Qwen3.5 SGLang AgentX bumps.qwen3.5-fp4-b300-sglang-agentic-power-ab, a controlled FP8/FP4 power matrix pinned to one SGLang build across both precisions.cluster:b300-nvwas retired in feat(runners): add the B300 DSXE cluster and retire the B300 NV launcher / 添加 B300 DSXE 集群并下线 B300 NV 启动脚本 #2826 (launcher and runner labels removed), so a sweep on it can never be scheduled. Repointed tocluster:b300-dsxe, the same move [Klaud Cold] glm5.2-fp4-b300-sglang-agentic-mtp: extend the AgentX curve to concurrency 48 and move to cluster:b300-dsxe / 将 GLM-5.2 B300 SGLang AgentX 曲线扩展至并发度 48 并迁移至 cluster:b300-dsxe #2829 makes for the GLM-5.2 FP4 B300 sibling. Because the runner changed this is not an append-only bump; the whole curve reruns on DSXE.Recipes touched:
qwen3.5-fp4-b300-sglang-agentic-mtpTest plan
🤖 Generated with Claude Code
Note
Low Risk
Benchmark config and changelog only; no application code, though runner migration invalidates prior B300-NV curves for this key.
Overview
Updates
qwen3.5-fp4-b300-sglang-agentic-mtpto use SGLangnightly-dev-cu13-20260907-30705c00instead ofv0.5.16-cu130, aligning with the other Qwen3.5 SGLang AgentX B200/H200/H100 bumps.The recipe runner moves from retired
cluster:b300-nvtocluster:b300-dsxeso agentic-coding sweeps can schedule again; that forces a full curve rerun on DSXE, not an append-only image bump.qwen3.5-fp4-b300-sglang-agentic-power-abstays on the old image so FP8/FP4 power A/B stays on one build.perf-changelog.yamldocuments the image, runner, and rationale.Reviewed by Cursor Bugbot for commit 25d9e08. Bugbot is set up for automated code reviews on this repo. Configure here.