Update kimik2.5-int4-mi300x-vllm vLLM ROCm image to v0.21.0#1404
Conversation
| - kimik2.5-int4-mi300x-vllm | ||
| description: | ||
| - "Update vLLM ROCm image from v0.18.0 to v0.20.2" | ||
| pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX |
There was a problem hiding this comment.
🔴 The new perf-changelog entry adds pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX — a literal XXX placeholder that was never substituted. Every other entry in this file uses the real numeric PR ID (e.g., pull/1310 immediately above). Since this is PR #1404, the link should be https://github.com/SemiAnalysisAI/InferenceX/pull/1404.
Extended reasoning...
What the bug is
perf-changelog.yaml line 2351 (the newly added entry in this PR) sets:
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXXThe XXX is a literal three-character placeholder, not a substituted PR number. This appears to be a templating oversight — the author updated the version description and config key correctly but forgot to fill in the PR number when committing.
Why this breaks convention
Every other entry in perf-changelog.yaml uses a real numeric PR ID:
- Line 2325:
pull/1304 - Line 2332:
pull/1305 - Line 2338:
pull/1308 - Line 2345:
pull/1310(the entry immediately above this new one)
The newly added entry breaks this pattern and produces a broken URL — clicking https://github.com/SemiAnalysisAI/InferenceX/pull/XXX would return a 404. Any downstream tooling that parses these links into clickable URLs, builds release notes, or correlates changelog entries to PRs will choke on this entry or surface a dead link.
Step-by-step proof
- Open
perf-changelog.yamland jump to line 2351. - Observe the value:
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX. - Compare to line 2345 (the immediately preceding entry's
pr-link):https://github.com/SemiAnalysisAI/InferenceX/pull/1310— a valid numeric ID. - Check the PR metadata: this PR is Update kimik2.5-int4-mi300x-vllm vLLM ROCm image to v0.21.0 #1404.
- Conclusion:
XXXshould be1404.
How to fix
Replace pull/XXX with pull/1404 on line 2351:
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1404This is a small documentation/metadata issue (it doesn't affect benchmark runtime), but it's an easy and clearly actionable fix that should land before merge so the changelog remains correctly cross-linkable.
|
|
||
| kimik2.5-int4-mi300x-vllm: | ||
| image: vllm/vllm-openai-rocm:v0.18.0 | ||
| image: vllm/vllm-openai-rocm:v0.20.2 |
There was a problem hiding this comment.
🟡 PR title says 'Update ... to v0.21.0' and the description says 'from v0.18.0 to v0.21.0', but the actual diff bumps the image to v0.20.2 (and the new perf-changelog entry also describes it as v0.20.2). Please reconcile — either fix the title/description to say v0.20.2 (most likely, since the diff and changelog agree), or bump the YAML further if v0.21.0 was actually intended.
Extended reasoning...
The mismatch. The PR is titled "Update kimik2.5-int4-mi300x-vllm vLLM ROCm image to v0.21.0" and the description states "Updates the vLLM ROCm image tag for `kimik2.5-int4-mi300x-vllm` from v0.18.0 to v0.21.0." However, the only code change in .github/configs/amd-master.yaml is at line 510:
kimik2.5-int4-mi300x-vllm:
- image: vllm/vllm-openai-rocm:v0.18.0
+ image: vllm/vllm-openai-rocm:v0.20.2and the companion perf-changelog.yaml entry at line 2348 says "Update vLLM ROCm image from v0.18.0 to v0.20.2". So the title/description disagree with both the on-disk image tag and the changelog entry.
Step-by-step proof.
- Open
.github/configs/amd-master.yamland jump to thekimik2.5-int4-mi300x-vllm:block (line 509). - Line 510 reads
image: vllm/vllm-openai-rocm:v0.20.2— notv0.21.0. - Open
perf-changelog.yaml, scroll to the new entry near line 2347–2351. Thedescriptionline says"Update vLLM ROCm image from v0.18.0 to v0.20.2", againv0.20.2. - Compare with the PR title (
v0.21.0) and description bullet (from v0.18.0 to v0.21.0): mismatch.
Likely interpretation. Since the actual code change and the changelog text are internally consistent on v0.20.2, the PR title/description are most plausibly stale — likely copy/paste leftover from an earlier draft or a different intended target. The change as written is internally consistent; only the PR metadata is wrong.
Impact. No runtime impact — the on-disk YAML is the source of truth for what image actually runs. The cost is purely metadata: reviewers, future commit log readers, and any tooling that parses PR titles/descriptions for release notes will be misled about which vLLM version this PR shipped. Also note pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX in the new changelog entry is a placeholder that should be replaced with the real PR number before merge.
How to fix. Either (a) edit the PR title to "Update kimik2.5-int4-mi300x-vllm vLLM ROCm image to v0.20.2" and fix the description bullet to from v0.18.0 to v0.20.2 — recommended, since the diff/changelog already agree on v0.20.2; or (b) if v0.21.0 was actually intended, bump line 510 of amd-master.yaml to vllm/vllm-openai-rocm:v0.21.0 and update the changelog text accordingly. While there, replace the pull/XXX placeholder with the real PR number.
Ref #1154 The previous PR head was 59 commits behind main and reverted ~20 unrelated image bumps from other merged PRs. This branch was reset to current main and the intended kimik2.5-int4-mi300x-vllm bump applied fresh (v0.18.0 -> v0.21.0, matching the PR title; the prior commit mis-targeted v0.20.2). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
cb67e38 to
bca2e3f
Compare
# Conflicts: # perf-changelog.yaml
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25984483683 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25984994682 |
2 similar comments
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25984994682 |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25984994682 |
|
/reuse-sweep-run |
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26003154083 |
Summary
kimik2.5-int4-mi300x-vllmfrom v0.18.0 to v0.21.0.Ref #1154
Generated with Claude Code