[None][test] Waive test_kda_verify_matches_sequential_decode[2-1] (NaN on B200) - #17561
Conversation
…N on B200) Signed-off-by: Brian Nguyen <brnguyen@nvidia.com>
da78d58 to
d7b44eb
Compare
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (1)
WalkthroughThe integration test waiver list now skips ChangesKDA verification waiver
Estimated code review effort: 1 (Trivial) | ~2 minutes Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
/bot run |
|
PR_Github #65602 [ run ] triggered by Bot. Commit: |
|
/bot skip --comment "waive-only PR" |
|
PR_Github #65603 [ skip ] triggered by Bot. Commit: |
|
PR_Github #65602 [ run ] completed with state |
|
PR_Github #65603 [ skip ] completed with state |
Waives
unittest/_torch/modeling/test_kimi_kda_verify_parity.py::test_kda_verify_matches_sequential_decode[2-1], which fails with NaN on DGX B200 (100% mismatched elements, greatest abs difference = nan). Observed in the L0 MergeRequest PR pipeline (build 53227, DGX_B200-PyTorch-4 stage).The
[2-1]case (t_steps=2, batch=1) of the KDA speculative-verify parity check produces NaN output from the verify path. Tracking bug: https://nvbugs/6599150.This is a waive-only change to unblock CI; the underlying KDA verify NaN is tracked separately.
Dev Engineer Review
test_kda_verify_matches_sequential_decode[2-1].6599150.QA Engineer Review
tests/integration/test_lists/waives.txt.