perf(agentx): refresh dsv4-fp4-b200-vllm-agentic-mtp harness - #2516
Conversation
|
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 |
b9a8538 to
045fee2
Compare
|
Claude finished @cquil11's task in 1m 29s —— View job Review
LGTM - no blocking issues found Verified: the new entry is appended at the end of |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
LGTM — trivial changelog-only entry, matches the pattern of the immediately preceding commit.
Extended reasoning...
Overview
The PR adds a single new entry to perf-changelog.yaml documenting a harness refresh for the dsv4-fp4-b200-vllm-agentic-mtp config. No code, harness scripts, or master-config files are touched.
Security risks
None. The change is a static YAML changelog entry with a description string and a PR link; it contains no executable code, credentials, or config that affects runtime behavior.
Level of scrutiny
Minimal scrutiny is warranted. This is a documentation/metadata-only change to a changelog file, and it follows the exact same shape as the immediately preceding commit (045fee2) which added an analogous entry for the sibling dsv4-b200-vllm-agentic-mtp config. I confirmed the referenced config key dsv4-fp4-b200-vllm-agentic-mtp exists in configs/nvidia-master.yaml, so the changelog entry points at a real, valid config.
Other factors
No prior reviews exist on this PR from me. The bug-hunting system found no issues, and the one candidate concern raised (changelog claims a harness refresh without an accompanying code change) was verified as consistent with existing repo convention, where the actual harness/image bump lands in a separate commit and this PR only adds the changelog record.
…00-vllm-agentic-mtp
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31150497460 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31152052267 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31153425262 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31192602558 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31192602558 |
1 similar comment
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31192602558 |
|
/stage-results 31192602558 |
|
@cquil11 staged run 31192602558: https://inferencemax-app-git-staging-semianalysisai.vercel.app/inference?i_dates=2026-08-07~r31192602558 This run remains available across future @cquil11 已将运行 31192602558 发布到预发布环境：https://inferencemax-app-git-staging-semianalysisai.vercel.app/inference?i_dates=2026-08-07~r31192602558 后续的 |
|
/reuse-sweep-run 31192602558 |
1 participant
Summary
dsv4-fp4-b200-vllm-agentic-mtp.vllm:metrics before publishing results. This preserves automatic direct-server collection for pure TP and adds engine metrics alongside router metrics for DP-attention.Validation
bash -n benchmarks/single_node/agentic/dsv4_fp4_b200_vllm_mtp.shpython3 utils/validate_perf_changelog.py --changelog-file perf-changelog.yaml --base-ref origin/main --head-ref HEADpython3 -m pytest utils/test_process_changelog.py utils/matrix_logic/test_generate_sweep_configs.py -q(116 passed)--server-metrics.git diff --check