[Klaud Cold] kimik2.5-fp4-b200-vllm: v0.26.0 image, extend conc to 512, add DEP4 arm / 升级 B200 镜像至 v0.26.0,扩展并发,新增 DEP4 - #2443
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 |
| - "Bump Kimi K2.5 NVFP4 B200 vLLM image from v0.22.0 to v0.26.0" | ||
| - "Extend TP8/TP4 concurrency sweep to 512" | ||
| - "Add DEP4 arm (tp4 ep4 dp-attn) with --prefill-schedule-interval 4, conc 128-512" | ||
| pr-link: TBD |
There was a problem hiding this comment.
🔴 The new changelog entry uses pr-link: TBD, but the placeholder validators only accept XXX or the canonical .../pull/XXX URL — any other value raises ChangelogValidationError. Please change it to pr-link: XXX, which gets auto-canonicalized to the real PR URL on merge.
Extended reasoning...
perf-changelog.yaml (line 5365) sets pr-link: TBD for the new Kimi-K2.5 FP4 B200 vLLM entry. This value is not one of the accepted placeholders used elsewhere in this same file (every other entry uses either XXX or the fully-qualified https://github.com/SemiAnalysisAI/InferenceX/pull/<N> URL).
utils/validate_perf_changelog.py defines PR_LINK_PLACEHOLDERS = {"XXX", "https://github.com/SemiAnalysisAI/InferenceX/pull/XXX"} (line 24). validate_added_pr_link (called from compare_entries for every newly appended entry when running in PR-context, i.e. pr_number is set) checks each new entry's pr-link against this set and against the expected canonical URL for the current PR. For link = "TBD", neither condition is satisfied, so it raises ChangelogValidationError("new PR entry must use {expected!r} or an XXX placeholder; found 'TBD'") at lines 156-160.
Independently, utils/prepare_perf_changelog_merge.py performs its own canonicalization pass during merge-prep (the step invoked by utils/merge_with_reuse.sh, the standard merge path for this repo). Lines 111-114 check the same PR_LINK_PLACEHOLDERS set and raise ChangelogValidationError(f"appended entry {index + 1} has unexpected pr-link {link!r}") if the link isn't a recognized placeholder — which TBD is not.
So this single value trips two separate, independent checks: PR-time changelog validation and merge-time canonicalization. Neither existing check is lenient enough to let an arbitrary human-readable placeholder like TBD through — they intentionally accept only XXX or the literal templated URL so that automation can reliably find-and-replace the placeholder with the real PR link at merge time.
Step-by-step proof:
- PR [Klaud Cold] kimik2.5-fp4-b200-vllm: v0.26.0 image, extend conc to 512, add DEP4 arm / 升级 B200 镜像至 v0.26.0,扩展并发,新增 DEP4 #2443 appends the entry at
perf-changelog.yaml:5365withpr-link: TBD. - When
validate_perf_changelog.pyruns in CI for this PR (withpr_number=2443), it callscompare_entries(), which callsvalidate_added_pr_link(entry, pr_number=2443, ...)for this newly appended entry. - Inside that function:
expected = "https://github.com/SemiAnalysisAI/InferenceX/pull/2443". The checkif link not in PR_LINK_PLACEHOLDERS and link != expectedevaluates:"TBD" not in {"XXX", ".../pull/XXX"}→ True, and"TBD" != expected→ True, so theandis True → raisesChangelogValidationError. - Even if that check were somehow bypassed,
prepare_perf_changelog_merge.py(invoked bymerge_with_reuse.shduring the actual merge) performs the same style of check on every appended entry and raises"appended entry N has unexpected pr-link 'TBD'"for the same reason. - Either failure blocks the standard PR/merge automation for this repo.
Fix: change line 5365 from pr-link: TBD to pr-link: XXX (matching the convention used elsewhere in this file, e.g. other pending entries). The XXX placeholder is automatically canonicalized to the real pull/<PR-number> URL by the merge-prep tooling once the PR is merged, so no other action is required.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30685082409 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30685321976 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30685953189 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30686066517 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30686416802 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30686475645 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30698907623 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30699342260 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30728903087 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30729663761 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30730792817 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30733232626 |
1 similar comment
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30733232626 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30751791884 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=30754226067 |
Summary
v0.22.0→vllm/vllm-openai:v0.26.0conc-end: 512tp: 4, ep: 4, dp-attn: true, conc-start: 128, conc-end: 512kimik2.5_fp4_b200.sh):DP_ATTENTIONhandling added — switches to--data-parallel-size, adds--enable-expert-paralleland--prefill-schedule-interval 4in DEP4 mode中文说明
v0.22.0→vllm/vllm-openai:v0.26.0conc-end: 512tp: 4, ep: 4, dp-attn: true, conc-start: 128, conc-end: 512kimik2.5_fp4_b200.sh):新增DP_ATTENTION处理逻辑,DEP4 模式下切换为--data-parallel-size,并启用--enable-expert-parallel和--prefill-schedule-interval 4