[https://nvbugs/6604925][test] Waive flaky TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM] (SIGSEGV on DGX_B200) - #17640
Conversation
…nvfp4[TRTLLM] (SIGSEGV on DGX_B200) test_nvfp4[TRTLLM] intermittently crashes the pytest process with SIGSEGV (exit 139) on DGX_B200 (~1% over a 14-day window, across many unrelated MRs and users). nvfp4 is Blackwell-only, so this test runs only on B200 test lists (l0_b200.yml, qa/llm_function_core.txt); the plain waive is therefore B200-scoped by construction. Tracked by https://nvbugs/6604925. Signed-off-by: Brian Nguyen <brnguyen@nvidia.com>
|
/bot run |
|
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 waiver list adds an NVBug-linked skip for the Qwen3.6 35B A3B NVFP4 TRTLLM accuracy test. ChangesQwen3.6 NVFP4 test waiver
Estimated code review effort: 1 (Trivial) | ~2 minutes Mergeability Score: ⚪ Minimal · up to This change only adds a narrowly scoped test waiver, and no actionable merge-blocking risk remains after normal checks and review. Possibly related PRs
Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
PR_Github #65923 [ run ] triggered by Bot. Commit: |
|
/bot skip --comment "waive-only PR" |
|
PR_Github #65924 [ skip ] triggered by Bot. Commit: |
|
PR_Github #65923 [ run ] completed with state |
|
PR_Github #65924 [ skip ] completed with state |
What
Waive
accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM]— NVBug https://nvbugs/6604925.Why
The test intermittently crashes the pytest process with SIGSEGV (exit 139) on DGX_B200 — a real in-process crash, not an assertion or an infra timeout (the infra classifier matched no infra signature). Flake history over the last 14 days: ~1% (9/885 runs) across ~330 unrelated MRs and 5 distinct users, all on DGX_B200 — a main-side flake, not PR-correlated. Fail-fast turns each occurrence into a red pipeline, so it's worth waiving while the crash is root-caused under the NVBug.
Scope (DGX_B200, implicitly)
nvfp4is Blackwell-only, so this test appears only in B200 test lists (test-db/l0_b200.yml,qa/llm_function_core.txt) and never runs on non-B200 platforms. A plainwaives.txtentry (nofull:<platform>/prefix) is therefore already scoped to DGX_B200 by construction — and unlike afull:B200/-prefixed entry, a plain entry reliably applies in the L0 context (DGX_B200-PyTorch-3stage) where the crash was observed. Only the[TRTLLM]variant is waived;[CUTEDSL]is left running.Tracked by https://nvbugs/6604925.
Dev Engineer Review
accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM].6604925.[TRTLLM]variant.waives.txtformat and does not alter public entities.QA Engineer Review
tests/integration/test_lists/waives.txt.