Skip to content

[https://nvbugs/6501394][fix] Amend the attempt 0 commit to also delete the single waives.txt line…#16840

Open
trtllm-agent wants to merge 1 commit into
NVIDIA:mainfrom
tensorrt-cicd:repair-bot-bug6501394
Open

[https://nvbugs/6501394][fix] Amend the attempt 0 commit to also delete the single waives.txt line…#16840
trtllm-agent wants to merge 1 commit into
NVIDIA:mainfrom
tensorrt-cicd:repair-bot-bug6501394

Conversation

@trtllm-agent

@trtllm-agent trtllm-agent commented Jul 24, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Root cause: PR [TRTLLM-13117][feat] Implement Uneven TP Linear for VisualGen models #14875 (Uneven TP Linear) made ROW-parallel Linear correctly handle hidden_size % tp_size != 0, so the accuracy assert in row_linear_forward no longer fails for hidden_size=15 and the stale pytest.raises(AssertionError) guard triggers "DID NOT RAISE"; the test's waiver in waives.txt (line 355) still skips it in CI.
  • Fix: Amend the attempt 0 commit to also delete the single waives.txt line unittest/_torch/multi_gpu/test_linear.py::test_row_linear[2-unbalanced] SKIP (https://nvbugs/6501394); leave adjacent lines (6507113, 6501404) untouched — they are different NVBugs.
  • Automated fix generated by repair-bot

Test plan

  • Verify fix on the same GPU type as the original failure
  • Check for regressions in related tests

Links

Dev Engineer Review

  • Removed the obsolete NVBug 6501394 waiver for test_row_linear[2-unbalanced].
  • Updated test_row_linear to assert successful execution for odd hidden sizes unconditionally.
  • Changes are consistent with the underlying Linear fix; waiver formatting and adjacent entries remain intact.
  • No API, performance, or error-handling regressions identified.

QA Engineer Review

  • Modified test: test_row_linear in tests/unittest/_torch/multi_gpu/test_linear.py.
  • The test is not represented by a corresponding test-db/ or qa/ entry; only its waiver entry was removed from waives.txt.
  • Verdict: insufficient.

…l Linear

After NVIDIA#14875 ("Uneven TP Linear for VisualGen models") landed, ROW-parallel
Linear with hidden_size not divisible by tp_size now works correctly via
_auto_tp_sharding + _calc_shard. For hidden_size=15, tp_size=2 both
torch.chunk(x, 2, dim=-1) and _calc_shard produce sizes [8, 7], and the
subsequent AllReduce reconstructs the full output. Consequently the
row_linear_forward accuracy check now passes for odd hidden sizes.

The remaining "if hidden_size % 2 != 0: pytest.raises(AssertionError)"
branch in test_row_linear is stale and produces "Failed: DID NOT RAISE".
Assert unconditionally, matching the pattern already used by
test_mlp / test_row_linear_norm_fusion / test_fp4_row_linear_allreduce.

Signed-off-by: handongl <handongl@nvidia.com>
@coderabbitai

coderabbitai Bot commented Jul 24, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4cbe1455-8ff2-4068-8a13-503294f6b4f0

📥 Commits

Reviewing files that changed from the base of the PR and between 2eef5fe and 2481aea.

📒 Files selected for processing (2)
  • tests/integration/test_lists/waives.txt
  • tests/unittest/_torch/multi_gpu/test_linear.py
💤 Files with no reviewable changes (1)
  • tests/integration/test_lists/waives.txt

Walkthrough

Changes

Row-linear validation

Layer / File(s) Summary
Require row-linear success
tests/unittest/_torch/multi_gpu/test_linear.py, tests/integration/test_lists/waives.txt
test_row_linear now asserts all rank results succeed for odd hidden_size values, and the matching skip waiver is removed.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Suggested reviewers: mikeiovine

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name Status Explanation Resolution
Description check ⚠️ Warning The description covers the issue and test plan, but it omits the required template sections like Description, Test Coverage, and PR Checklist. Add the template sections explicitly: Description, Test Coverage, and PR Checklist, and ensure the summary is placed under the required @coderabbitai summary header.
✅ Passed checks (4 passed)
Check name Status Explanation
Title check ✅ Passed The title is specific and correctly tied to the NVBug fix, though it emphasizes the waiver removal more than the test change.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Comment @coderabbitai help to get the list of available commands.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants