Skip to content

[DEBUG] remove qwen3.5-fp8-mi355x-atom perf-changelog entry #1296

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
seungrokj merged 2 commits into from
May 10, 2026
Merged

[DEBUG] remove qwen3.5-fp8-mi355x-atom perf-changelog entry #1296

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c11b281
Remove duplicate qwen3.5-fp8-mi355x-atom perf-changelog entry
seungrokj May 7, 2026
5594bb9
Merge branch 'main' into srok/atom_qwen3.5_fp8_fix
seungrokj May 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 0 additions & 8 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1436,14 +1436,6 @@
- "Image includes upstream SGLang PRs: https://github.com/sgl-project/sglang/pull/21188, https://github.com/sgl-project/sglang/pull/21421, https://github.com/sgl-project/sglang/pull/20736"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1036

- config-keys:
- qwen3.5-fp8-mi355x-atom
- qwen3.5-fp8-mi355x-atom-mtp
description:
- "Add Qwen3.5-397B-A17B FP8 MI355X ATOM benchmark configs with and without MTP"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1040


- config-keys:
- qwen3.5-fp4-mi355x-sglang
description:
Comment on lines 1436 to 1441

@claude claude Bot May 7, 2026

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.

🔴 This PR's premise is incorrect — the removed block is not a duplicate, it is the only changelog entry for qwen3.5-fp8-mi355x-atom / qwen3.5-fp8-mi355x-atom-mtp (and it is the entry referencing PR #1040). After this PR merges, perf-changelog.yaml has zero entries documenting these configs, while the configs themselves still exist in .github/configs/amd-master.yaml. The PR should be reverted; the author may have been thinking of the actual dsv4-fp4-mi355x-atom duplicate at lines 1832/1843, which is a different config-key.

Extended reasoning...

What the bug is

The PR description claims it removes a duplicate perf-changelog entry for qwen3.5-fp8-mi355x-atom / qwen3.5-fp8-mi355x-atom-mtp that was "accidentally included alongside the existing entry referencing PR #1040." This premise is factually wrong: the deleted block is the only entry for those config-keys, and it is the entry referencing PR #1040. There was no second entry alongside it.

How it manifests

Pre-PR (git show HEAD~1:perf-changelog.yaml), exactly one config-keys list item contains qwen3.5-fp8-mi355x-atom / -mtp (lines 1439-1444):

- config-keys:
    - qwen3.5-fp8-mi355x-atom
    - qwen3.5-fp8-mi355x-atom-mtp
  description:
    - 'Add Qwen3.5-397B-A17B FP8 MI355X ATOM benchmark configs with and without MTP'
  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1040

This is precisely the block the PR deletes. Post-PR, grep '- qwen3.5-fp8-mi355x-atom$' on perf-changelog.yaml returns 0 hits as a config-key. The remaining matches at lines 1835/1846 are inside description strings of unrelated dsv4-fp4-mi355x-atom entries ("...matches qwen3.5-fp8-mi355x-atom base..."), not config-key entries.

Step-by-step proof

  1. git show HEAD~1:perf-changelog.yaml | grep -nE '^\s+- qwen3.5-fp8-mi355x-atom($|-mtp$)' → exactly two lines (the config-keys list items at lines 1440-1441), belonging to a single block.
  2. git show HEAD~1:perf-changelog.yaml | sed -n '1439,1444p' → shows the block has pr-link: .../pull/1040. So the "existing entry referencing PR [AMD/ROCM] atom qwen fp8/fp8_mtp3 on mi355x #1040" is this block.
  3. git log --all --oneline -S 'qwen3.5-fp8-mi355x-atom' -- perf-changelog.yaml → only a70a094 (file creation) ever added the token as a config-key, and c11b281 (this PR) removed it. No commit ever added a duplicate.
  4. Post-PR: grep -nE '^\s+- qwen3.5-fp8-mi355x-atom($|-mtp$)' perf-changelog.yaml → 0 hits.
  5. The configs still exist in .github/configs/amd-master.yaml (lines 265 and 288), so they are real, exercised configs whose changelog provenance has now been wiped.

Why the premise is plausible-sounding but wrong

There is a real duplicate in the file: dsv4-fp4-mi355x-atom appears as a config-keys entry at both lines 1832 and 1843 of the post-PR file. The author may have noticed that and confused it with the qwen entry — the description strings on those dsv4-fp4-mi355x-atom blocks textually mention "qwen3.5-fp8-mi355x-atom base" (referring to the shared docker image), which makes the config-key appear to occur multiple times to a casual grep. But those textual mentions are not config-keys list items.

Impact

The changelog now has no entry documenting when/how qwen3.5-fp8-mi355x-atom and qwen3.5-fp8-mi355x-atom-mtp were added. There is repo precedent that this matters: PR #1287/#1293 ("Re-append perf-changelog entry for minimaxm2.5-fp8-mi355x-vllm") was filed specifically to restore a similarly missing changelog entry. The PR does the opposite of its stated intent.

How to fix

Revert this PR (git revert c11b281), restoring the lines 1439-1444 block. If the author intended to clean up the actual dsv4-fp4-mi355x-atom duplicate at file lines 1832/1843, that should be done in a separate PR that explicitly targets that config-key.

Expand Down