DSv4 B200 TRT#1277
Conversation
|
Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook If it is not, please create a PR first before we can merge your PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you 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. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow As a rule of thumb, generally, PR authors should request a review & get a PR approval from the respective companies' CODEOWNERS before requesting a review from core maintainers. If additional help is needed, PR authors can reach out to core maintainers over Slack. |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| description: | ||
| - "Add B200 TensorRT-LLM DeepSeek-V4-Pro single-node coverage using the feat/deepseek_v4 image" | ||
| - "Mirror the B300 TRT launch path with OpenAI chat serving, FP8 KV cache, TRTLLM MoE, NCCL NVLS disabled by default, and fused MHC disabled for hidden size 7168 correctness" | ||
| pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/TBD |
There was a problem hiding this comment.
🟡 The new perf-changelog.yaml entry sets pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/TBD (line 2216), which is a leftover placeholder. Since this PR is #1277, replace TBD with 1277 so the changelog link resolves. Every other entry in the file uses a real PR number.
Extended reasoning...
What the bug is. The newly-added changelog entry for dsv4-fp4-b200-trt ends with:
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/TBDTBD is a placeholder string that was meant to be replaced with the actual PR number once the PR was opened. Visiting that URL today produces a 404 — there is no PR #TBD on GitHub.
Why existing code does not catch it. utils/matrix_logic/validation.py only validates that pr_link is a string, so the YAML loads cleanly and the PR-submission flow described in the description (bash -n, YAML load, generate_sweep_configs.py test-config) all pass. There is no automated check that the link resolves or even matches a numeric pattern. AGENTS.md shows XXX as the placeholder used while drafting before the PR exists, but at PR-submission time it is expected to be the real number.
Step-by-step proof.
- Open the PR diff: the appended entry at
perf-changelog.yaml:2216ispr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/TBD. - The PR metadata in the timeline shows this is PR
#1277. - Compare with surrounding entries in the same file:
/pull/1222,/pull/1256,/pull/1065,/pull/1272— all are real numeric PR IDs. - Resolving
https://github.com/SemiAnalysisAI/InferenceX/pull/TBDdoes not produce a real PR; the link is dangling.
Impact. Documentation-only; benchmark execution is unaffected. The cost is that anyone later auditing the changelog (e.g. trying to figure out which PR introduced the dsv4-fp4-b200-trt config) cannot click through to find the discussion or review.
Fix. Replace TBD with 1277 on the pr-link line of the new entry before merge:
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1277Trivial edit, no other call-sites to update.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25347394482 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25347972124 |
2 similar comments
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25347972124 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25347972124 |
|
@Oseltamivir the upstream branch has ur commit now NVIDIA/TensorRT-LLM#13771 !! |
Removed comments and environment variable related to fused HC.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25403731383 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25407156900 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25408925419 |
2 participants
Summary
Validation