Skip to content

[https://nvbugs/7078630][test] Unwaive FP4 MLA helix-scatter unit tests fixed by #21313 - #22421

Open
reasonsolo wants to merge 1 commit into
NVIDIA:mainfrom
reasonsolo:user/lizhiz/unwaive-fp4-mla-helix-scatter-7078630
Open

reasonsolo wants to merge 1 commit into
NVIDIA:mainfrom
reasonsolo:user/lizhiz/unwaive-fp4-mla-helix-scatter-7078630

Conversation

@reasonsolo

@reasonsolo reasonsolo commented Nov 17, 2026 •

Copy link
Copy Markdown
Collaborator

What

Removes the two stale SKIP entries from tests/integration/test_lists/waives.txt:

  • unittest/_torch/attention/test_fp4_mla.py::test_fp4_mla_helix_scatter_uses_global_rope_and_local_cache_position[active]
  • unittest/_torch/attention/test_fp4_mla.py::test_fp4_mla_helix_scatter_uses_global_rope_and_local_cache_position[inactive]

Both were waived for https://nvbugs/7078630. Unwaive only — no production code changes.

Why

The helix-scatter fix landed in main via #21313 (d7f2a84c), and post-merge verification shows both parametrizations pass:

Evidence Value
Commit 1e658cb1 (contains d7f2a84c)
Job Slurm 6032907, exit 0
Hardware 1 GPU from the GR100 pool (compute capability 10.7)
Harness target CI image, CI test_unittests.py adapter
Result 2 unit-test JUnit records: 2 passed, 0 failures / errors / skips

That verification covers these two cases only, not the other changes in #21313 and not performance.

Scope and safety

Unwaive only. tests/unittest/_torch/attention/test_fp4_mla.py keeps its hardware guard (_is_cutedsl_unavailable() skips unless Rubin SM107 + CuTeDSL runtime), and the file is already part of tests/integration/test_lists/test-db/l0_gr100.yml, so the cases simply leave the waive list.

Verification of this change

  • python3 scripts/check_test_list.py --validate --check-duplicate-waives → OK: 2228 unique test entries validated
  • pre-commit on the commit: duplicate-waives check and AST test-list validation pass
  • git diff is a 2-line deletion in waives.txt; no other nvbugs/7078630 reference remains in the tree

Waiver-tracking follow-up for nvbugs/7078630.

Dev Engineer Review

  • The change removes two SKIP entries from tests/integration/test_lists/waives.txt. The entries cover the active and inactive parametrizations of test_fp4_mla_helix_scatter_uses_global_rope_and_local_cache_position.
  • The diff has no production-code changes. It adds 0 lines and removes 2.
  • The unwaived tests now run in CI. The unwaive depends on the fix from #21313 being in main. The PR text states this, but this review did not verify it.
  • The PR reports post-merge verification of both cases on one GR100 GPU. This evidence covers only these two cases.
  • The PR reports that check_test_list.py --validate --check-duplicate-waives and the pre-commit checks passed. These results are author-reported and were not re-run for this review.

QA Engineer Review

Test-list files modified:

  • tests/integration/test_lists/waives.txt: removed the SKIP entries for the active and inactive cases of test_fp4_mla_helix_scatter_uses_global_rope_and_local_cache_position.

Behavior covered:

  • The two FP4 MLA helix-scatter cases verify that the scatter path uses the global RoPE and local cache position. These cases now run instead of being skipped.
  • The author reports that the test file keeps its hardware guard. The author also reports that tests/integration/test_lists/test-db/l0_gr100.yml already lists the test file. This review did not verify either claim.

Coverage verdict: sufficient for the two unwaived cases, based on the author's report of a passing run on one GR100 GPU. The result does not cover other changes in #21313 or performance.

Per-File QA Perspective

  • tests/integration/test_lists/waives.txt: Removing these two skips makes the FP4 MLA helix-scatter cases run in CI. QA should confirm that both cases pass on the GR100 CI pool.

…ts fixed by NVIDIA#21313

Signed-off-by: reasonsolo <1432185+reasonsolo@users.noreply.github.com>
@coderabbitai

coderabbitai Bot commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

Explain Change →

No actionable comments were generated in the recent review.

Walkthrough

The skip list no longer includes the active and inactive FP4 MLA Helix scatter test cases.

Changes

Test skip list

Layer / File(s) Summary
Remove FP4 MLA Helix scatter skip entries
tests/integration/test_lists/waives.txt
The skip list no longer contains the active and inactive test entries.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Change: Other

Suggested reviewers: trtllm-agent

Merge Risk: Minimal · up to 4b0ea

This change re-enables two previously skipped FP4 MLA helix-scatter tests after the fix landed. It does not alter production behavior, and reverting it is simple if the tests fail in CI.

Pre-merge checks: Passed 4

Check name Status Explanation
Title check Passed The title names the NVBugs ID and [test] type, and says the PR unwaives the FP4 MLA helix-scatter unit tests. The change removes the two matching waive entries from waives.txt, so the title matches th…
Description check Passed The description follows the repository's intent. It explains what changed (two stale SKIP entries removed from waives.txt), why (the fix landed via #21313 and post-merge verification passed), the scop…
Linked Issues check Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check Passed Check skipped because no linked issues were found for this pull request.
Full details: Title check

Explanation

The title names the NVBugs ID and [test] type, and says the PR unwaives the FP4 MLA helix-scatter unit tests. The change removes the two matching waive entries from waives.txt, so the title matches the main change. It does not follow the template's [type] values exactly, but [test] is a reasonable type and the title is specific.

Full details: Description check

Explanation

The description follows the repository's intent. It explains what changed (two stale SKIP entries removed from waives.txt), why (the fix landed via #21313 and post-merge verification passed), the scope and safety of the change, and the validation run. The template's Description and Test Coverage sections are not filled in as headings, and the PR checklist is not marked. These gaps are minor because the content covers the same information.

Finishing Touches

Generate unit tests (beta)

  • Create a new PR
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Recent review info

Run configuration

Commits

Reviewing files that changed from the base of the PR and between 29869ba and 4b0ea4d.

Files selected for processing: 1

  • tests/integration/test_lists/waives.txt
Files with no reviewable changes: 1

  • tests/integration/test_lists/waives.txt

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

@sunnyqgg sunnyqgg left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

0 blockers, 0 possible blockers, 0 required, 1 nice to have. The unwaive looks right. The fix (#21313, d7f2a84) is on this branch's base. The waived failure was deterministic: in post-merge 3066–3069 both cases failed at the old whole-buffer prequantized_q assertion (torch.Size([128, 320]) != torch.Size([4, 320]), from the packed-Q padding), which #21313 corrected in the test. The remaining assertions match the code at the base: Q batch capacity 1, RoPE at the Helix global position, and inactive ranks return before any cache store. Nothing in the FP4 MLA path or the test changed between the cited run's commit 1e658cb and the base. The only ask is a CI run of the post-merge GR100 stage on this PR (inline).

@@ -169,8 +169,6 @@ test_doc.py::test_http_url_validity SKIP (https://nvbugs/7066122)
test_e2e.py::test_ptp_quickstart_bert[VANILLA-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity] SKIP (bug pending, tracked in PR 17414)
unittest/_torch/attention/multi_gpu --ignore=unittest/_torch/attention/multi_gpu/test_helix_zero_kv.py SKIP (https://nvbugs/7044056)
unittest/_torch/attention/sparse/csa2/test_module.py::test_fused_native_index_q_and_graph SKIP (https://nvbugs/7039506)

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

[Nice to have] Please link a CI run for the two cases unwaived just below. test_fp4_mla.py runs only on GR100-PyTorch-Post-Merge-1 (l0_gr100.yml, stage: post_merge; every other list skips it via _is_cutedsl_unavailable()), so the default /bot run won't execute these cases. Every earlier CI run (post-merge 3066–3069) stopped at the old whole-buffer prequantized_q assertion, so the later assertions have never run in CI. Suggest /bot run --stage-list "GR100-PyTorch-Post-Merge-1" (an explicitly named post-merge stage doesn't need the ci: post-merge approved label), then check that both [active] and [inactive] show PASSED, not SKIPPED, in the inner test_fp4_mla.py results. Go by those inner results: in post-merge 3076 the stage also had unrelated MiniMax-M3 and test_moe_backend failures.

@github-actions

Copy link
Copy Markdown

Automatically added "ci: full pre-merge approved" because this PR has satisfied the required GitHub review approvals. Unresolved review conversations and other required checks remain independent merge requirements.

@reasonsolo

Copy link
Copy Markdown
Collaborator Author

/bot run --stage-list "GR100-PyTorch-Post-Merge-1"

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #85347 [ run ] triggered by Bot. Commit: 4b0ea4d Link to invocation

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants