[https://nvbugs/6633268][test] waive DeepSeekV3Lite bfloat16 mtp_nextn=2 v2_kv_cache test on DGX B200 - #17957
Conversation
…n=2 v2_kv_cache test on DGX B200 Signed-off-by: xxi <xxi@nvidia.com>
|
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)
Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review. WalkthroughThe integration test waiver list adds a DGX_B200 waiver for the configured ChangesIntegration test waiver
Estimated code review effort: 1 (Trivial) | ~2 minutes Merge Risk: ⚪ Minimal · up to This localized test-waiver change does not alter product or runtime behavior, and the waiver remains limited to DGX_B200 while other listed coverage stays enabled. No actionable merge-blocking risk remains beyond normal validation of the active test list. Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
/bot skip --comment "waive the flacky test to unblock" |
|
PR_Github #67344 [ skip ] triggered by Bot. Commit: |
|
PR_Github #67344 [ skip ] completed with state |
|
/bot skip --comment "waive the flacky test to unblock" |
|
PR_Github #67393 [ skip ] triggered by Bot. Commit: |
|
PR_Github #67393 [ skip ] completed with state |
Description
Waive
accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]on DGX_B200, tracked by https://nvbugs/6633268.The test failed in pre-merge
LLM/main/L0_MergeRequest_PR#54717, stageDGX_B200-PyTorch-6, withTest terminated unexpectedly.OpenSearch history over the last 30 days on DGX_B200 shows the case is flaky rather than consistently broken: 13 failures out of 1000 runs (1.3%), flip-flop rate 28.9%, longest failure streak 1.
The waive is scoped to
full:DGX_B200/so that coverage on other platforms is preserved. This case also runs froml0_h100.yml, which stays enabled. An existing waive for the same case is scoped tofull:RTX_PRO_6000_Blackwell_Server_Edition/(https://nvbugs/6451323) and does not cover DGX_B200.Test Coverage
N/A - test list change only.
PR Checklist
Please review the following before submitting your PR:
Dev Engineer Review
tests/integration/test_lists/waives.txt.DeepSeekV3Litebfloat16mtp_nextn=2v2_kv_cachetest.full:DGX_B200/.QA Engineer Review
test-db/orqa/files were modified.