Skip to content

Update minimaxm2.5-fp4-mi355x-vllm vLLM ROCm image to v0.21.0#1409

Merged
Oseltamivir merged 2 commits into
mainfrom
claude/issue-1154-minimaxm2.5-fp4-mi355x-vllm
May 17, 2026
Merged

Update minimaxm2.5-fp4-mi355x-vllm vLLM ROCm image to v0.21.0#1409
Oseltamivir merged 2 commits into
mainfrom
claude/issue-1154-minimaxm2.5-fp4-mi355x-vllm

Conversation

@Klaud-Cold
Copy link
Copy Markdown
Collaborator

Summary

  • Updates the vLLM ROCm image tag for minimaxm2.5-fp4-mi355x-vllm from v0.19.1 to v0.21.0.

Ref #1154

Generated with Claude Code

Ref #1154

Co-authored-by: Klaud Cold <Klaud-Cold@users.noreply.github.com>
@github-actions
Copy link
Copy Markdown
Contributor

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 single node 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.

Comment thread perf-changelog.yaml Outdated
- minimaxm2.5-fp4-mi355x-vllm
description:
- "Update vLLM ROCm image from v0.19.1 to v0.21.0"
pr-link: XXX
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🔴 The new perf-changelog entry at line 2501 has pr-link: XXX instead of an actual PR URL — the placeholder from the changelog template was left in. Every other entry in the file uses pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/<number>; this should be https://github.com/SemiAnalysisAI/InferenceX/pull/1409.

Extended reasoning...

What is the bug

perf-changelog.yaml line 2501 commits the literal placeholder XXX as the pr-link value:

- config-keys:
    - minimaxm2.5-fp4-mi355x-vllm
  description:
    - "Update vLLM ROCm image from v0.19.1 to v0.21.0"
  pr-link: XXX

This is the template placeholder shown in AGENTS.md (the documented template form is pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX). The author bumped the image tag in .github/configs/amd-master.yaml and updated the changelog but never substituted the actual PR number (1409) into pr-link. Every other entry in the file uses a full URL (see e.g. lines 2482, 2488, 2495).

Why existing code does not prevent it

There is no schema/format validation on pr-link in the diff or surrounding code that would reject a bare XXX. YAML parsing happily produces the string "XXX".

Impact — concrete tooling failure

This is more than a cosmetic documentation issue. utils/merge_with_reuse.sh has logic that specifically expects the /pull/XXX URL form, not a bare XXX:

  1. The filter if "XXX" not in link and not link.endswith(f"/pull/{pr}"): continue does keep the entry (because "XXX" in "XXX" is true), so the entry is selected as a contribution.
  2. The substitution block.replace("/pull/XXX", f"/pull/{pr}") then looks for the literal substring /pull/XXX. Since the value here is just XXX (no /pull/ prefix), the replace is a no-op and the placeholder is silently retained in the merged changelog.
  3. Immediately after, the script asserts last["pr-link"].endswith("/$PR"). With pr-link: XXX, "XXX".endswith("/1409") is false, so the assertion fails and the merge-with-reuse path errors out.

Step-by-step proof

Concrete walk-through with PR #1409:

  1. pr-link in YAML is parsed to Python string "XXX".
  2. In the contribution loop: link = "XXX"; "XXX" in "XXX" is True, so the entry is not skipped (good — it is supposed to be picked up).
  3. block.replace("/pull/XXX", "/pull/1409") is called on the raw YAML block. The block contains pr-link: XXX, not pr-link: ...pull/XXX, so no match is found. The block is appended unchanged, still with pr-link: XXX.
  4. The script then parses the rewritten file and runs assert entries[-1]["pr-link"].endswith("/1409"). "XXX".endswith("/1409") is False, so the assertion fires and the merge script exits with an AssertionError.

Result: if merge_with_reuse.sh is used for this PR, it will fail on the assertion. Even if a maintainer bypasses that path (e.g. squash-merges directly), the changelog entry remains untraceable — XXX is not a URL, so any downstream tooling or reader that consumes perf-changelog.yaml cannot link the entry to its PR.

How to fix

Replace line 2501 with the proper URL:

  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1409

This matches every other entry in the file and satisfies both the /pull/XXX substitution and the endswith("/<pr>") assertion in utils/merge_with_reuse.sh.

@github-actions
Copy link
Copy Markdown
Contributor

@Oseltamivir
Copy link
Copy Markdown
Collaborator

/reuse-sweep-run

@Oseltamivir Oseltamivir merged commit 539ea1b into main May 17, 2026
2 of 4 checks passed
@Oseltamivir Oseltamivir deleted the claude/issue-1154-minimaxm2.5-fp4-mi355x-vllm branch May 17, 2026 00:33
@github-actions
Copy link
Copy Markdown
Contributor

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

Development

Successfully merging this pull request may close these issues.

2 participants