Repository navigation
[https://nvbugs/6581067][test] Unwaive test_fp4_gemm_bias_per_backend mnk2/mnk3 - #20581
Conversation
|
/bot run --disable-fail-fast |
|
PR_Github #78827 [ run ] triggered by Bot. Commit: |
… mnk2/mnk3 The waivers were removed by NVIDIA#17505 and re-added by NVIDIA#18232. The parity reference is single-rounded, so the cases pass without a waiver. Signed-off-by: Yiyun Lu <55233584+luyiyun1021@users.noreply.github.com>
d8c2c88 to
44d7a63
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #78830 [ run ] triggered by Bot. Commit: |
|
PR_Github #78827 [ run ] completed with state |
|
PR_Github #78830 [ run ] completed with state |
brnguyen2
left a comment
There was a problem hiding this comment.
Approving — the comments below are optional touch-ups, not blockers.
Two things before merge:
- The description says #18232 re-added these waivers but not why. If #18232 added them back because of a new GB10 failure after the reference fix (rather than a stale waive-list merge), removing them now would bring that failure back. Please state which case it was.
- These cases fail on GB10 in the post-merge stage, so the default pre-merge
/bot runwon't run them. Run/bot run --extra-stage "GB10-PyTorch-Post-Merge-1"(or whichever GB10 post-merge stage carriestest_fp4_linear.py) and link the green result here.
|
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. |
|
/bot run --extra-stage "GB10-PyTorch-Post-Merge-1" |
|
PR_Github #79129 [ run ] triggered by Bot. Commit: |
|
PR_Github #79129 [ run ] completed with state |
Signed-off-by: Yiyun Lu <55233584+luyiyun1021@users.noreply.github.com>
@coderabbitai summary
Description
Remove the four
test_fp4_gemm_bias_per_backend[mnk2/mnk3-cublaslt/cutlass]waivers for https://nvbugs/6581067. #17505 fixed the parity reference (single rounding) and removed these waivers; #18232 added them back, so the fixed cases are skipped on main.Test Coverage
unittest/_torch/thop/parallel/test_fp4_linear.py::test_fp4_gemm_bias_per_backend[mnk2/mnk3-cublaslt/cutlass]run in CI without the waivers; not run locally.PR Checklist
PR description clearly explains what and why.
PR Follows TRT-LLM CODING GUIDELINES.
Test cases are provided for new code paths.
Any new dependencies have been scanned for license and vulnerabilities.
CODEOWNERS updated if ownership changes.
Documentation updated as needed.
Update tava architecture diagram if significant design change.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.