[https://nvbugs/6525011][test] Waive TestLagunaXS::test_fp8 on RTXPro6000D - #17661
Conversation
…6000D The CUDA-graph FP8-MoE illegal-memory-access failure tracked in nvbugs/6525011 (already waived on full:B200 and full:B300) also reproduces on RTXPro6000D. Extend the same waiver to that platform. Signed-off-by: Brian Nguyen <brnguyen@nvidia.com>
|
/bot skip --comment "waive only PR" |
|
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 waiver list skips ChangesTest waiver
Estimated code review effort: 1 (Trivial) | ~2 minutes Mergeability Score: ⚪ Minimal · up to This is a localized waiver-only change with no production behavior impact. No actionable merge-blocking risk remains beyond normal checks and review. Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
/bot run |
|
PR_Github #66065 [ skip ] triggered by Bot. Commit: |
|
PR_Github #66069 [ run ] triggered by Bot. Commit: |
|
PR_Github #66065 [ skip ] completed with state |
|
/bot skip --comment "waive only PR" |
|
PR_Github #66074 [ skip ] triggered by Bot. Commit: |
|
PR_Github/17661-5febb1c #66069 was force-killed by a newer pipeline run. |
|
PR_Github #66074 [ skip ] completed with state |
Description
accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_fp8is already waived onfull:B200andfull:B300under nvbugs/6525011 (CUDA-graph runner not holding a strong ref to graph outputs → illegal memory access during generation CUDA-graph warmup on the TRT-LLM Gen FP8 MoE path). The same failure now reproduces on RTXPro6000D (Blackwell, SM120), which was not covered by the existing waivers, so the test breaks pre-merge L0 on that stage (observed on PR CI run L0_MergeRequest_PR #53409, unrelated to that PR).This extends the existing waiver to
full:RTXPro6000D, citing the same bug.Scope note: the test also runs on H100 (Hopper/SM90), but there is no failure evidence there and the faulting kernel is Blackwell-gen FP8 MoE, so H100 is intentionally left unwaived to preserve coverage.
Test Coverage
Waive-only change; no new tests. Removes one known-failing test from the RTXPro6000D pre-merge stage.
PR Checklist
[JIRA/NVBUG/None][type] descriptionDev Engineer Review
TestLagunaXS::test_fp8onfull:RTXPro6000D.nvbugs/6525011, consistent with existing waivers.tests/integration/test_lists/waives.txt.QA Engineer Review
tests/integration/test_lists/waives.txt.RTXPro6000Dwaiver entry.test-db/orqa/files were modified.