Refresh MiniMax-M3 NVFP4 B300 8k1k disaggregated EAGLE Dynamo-vLLM / 刷新 MiniMax-M3 NVFP4 B300 8k1k 分离式 EAGLE Dynamo-vLLM 配置 - #2314
Conversation
Bump the six existing 8k1k Dynamo-vLLM EAGLE recipes to the refreshed upstream vLLM nightly and remove the deprecated TRT-LLM FlashInfer all-reduce override. 中文：将现有六个 8k1k Dynamo-vLLM EAGLE 配方更新到新版上游 vLLM nightly 镜像，并移除已弃用的 TRT-LLM FlashInfer all-reduce 覆盖配置。
|
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 |
jasonlizhengjian
added
the
non-canary-full-sweep-enabled
|
Claude finished @jasonlizhengjian's task in 2m 15s —— View job ReviewLGTM - no blocking issues found
Verified: all six recipe YAMLs and the Two things I could not verify from this sandbox (network/exec restrictions): that the |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
LGTM — mechanical image-tag refresh and deprecated env-var removal across the six MiniMax-M3 B300 EAGLE recipes.
What was reviewed:
- Confirmed all six recipe diffs are identical mechanical changes (image tag bump + drop
VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllmfrom prefill/decode env), no topology, speculative-config, or concurrency changes. - Verified
configs/nvidia-master.yamlimage reference andperf-changelog.yamlentry match the recipe changes. - Cross-checked against the established precedent (#2120) for the same env-var removal pattern.
Extended reasoning...
Overview
Refreshes the vLLM nightly container image tag and removes a deprecated VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm env override across six existing MiniMax-M3 NVFP4 B300 8k1k disaggregated EAGLE3 recipe YAMLs, plus the corresponding configs/nvidia-master.yaml image pin and a perf-changelog.yaml entry. No topology, speculative-decoding, chat-template, or concurrency settings are touched.
Security risks
None. This is a benchmark-recipe config change (container tag bump, env var removal) with no code execution paths, auth, or data-handling logic involved.
Level of scrutiny
Low. All six file diffs are byte-for-byte identical in shape (only the file name/topology differs), following an established precedent (#2120) for dropping the same deprecated env var. I independently verified the current main branch already reflects the identical image tag and env-var removal (landed via a separate direct commit, d36d9b3), and that a follow-up commit (8985f5f) explicitly registers this PR number in the changelog to drive its validation sweep — this matches the repo'''s established workflow for this kind of recipe refresh rather than indicating drift or a stale/duplicate PR.
Other factors
The PR description reports 300 passing tests for the matrix-logic/changelog-gate test suites, and the change is scoped narrowly (no B200 or non-EAGLE recipes touched, per the author'''s own description). No outstanding review comments need addressing.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30030112902 |
2 similar comments
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30030112902 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30030112902 |
Signed-off-by: Jason Li <lijas@nvidia.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30038295118 |
Signed-off-by: Jason Li <lijas@nvidia.com>
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30042427571 |
1 similar comment
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30042427571 |
|
/reuse-sweep-run |
Ankur-singh
changed the title
|
As a PR reviewer and CODEOWNER, I have reviewed this and have:
Additional detail section:
Signed: |
✅✅✅ Verdict: PASS ✅✅✅✅ Check 0 (CODEOWNER): PASS — |
|
recipe. vllm-project/recipes#682 |
check-changelog rejected the previous push: main had appended the MiniMax-M3 NVFP4 B300 entry (#2314) after this branch's last merge, so relative to current main the branch looked like it deleted those 19 lines. Resolved per KLAUD_DEBUG.md 1.1 -- perf-changelog.yaml is rebuilt as current main's bytes verbatim plus this branch's own five entries re-appended at the tail, verified byte-exact. No 3-way merge of that file: it silently rewrites whitespace and re-trips the deletion check. 中文：上一次推送被 check-changelog 拒绝：本分支上次合并之后，main 追加了 MiniMax-M3 NVFP4 B300 条目（#2314），因此相对于当前 main，本分支看起来删除了那 19 行。 已按 KLAUD_DEBUG.md 1.1 处理：perf-changelog.yaml 重建为当前 main 的原始字节，再将本 分支自身的五个条目重新追加到文件末尾，并校验为字节级纯追加。该文件不做三方合并—— 三方合并会静默改动空白字符并再次触发删除检查。
6 participants
Summary
vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9.VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllmoverride from those five refreshed recipes, matching the image-compatibility change established in Update vLLM MiniMax-M3 image tags / 更新 vLLM MiniMax-M3 镜像标签 #2120.Validation
python -m pytest utils/matrix_logic/ utils/changelog_gate_tests/ -q: 300 passed.中文说明
vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9。VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm覆盖配置，与 Update vLLM MiniMax-M3 image tags / 更新 vLLM MiniMax-M3 镜像标签 #2120 的镜像兼容性调整保持一致。验证
python -m pytest utils/matrix_logic/ utils/changelog_gate_tests/ -q：300 项测试通过。