[https://nvbugs/6468821][infra] Split slow B300 attention unit tests - #16709
Conversation
|
/bot run --disable-fail-fast |
Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.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)
🚧 Files skipped from review as they are similar to previous changes (1)
WalkthroughThe PyTorch integration test list excludes three sparse attention tests from the broader attention test invocation and runs them as separate entries. ChangesAttention test selection
Estimated code review effort: 2 (Simple) | ~10 minutes Suggested reviewers: 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
|
PR_Github #60888 [ run ] triggered by Bot. Commit: |
|
PR_Github #60888 [ run ] completed with state |
|
/bot run --disable-fail-fast |
…CBTS Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
|
/bot run --disable-fail-fast |
|
PR_Github #60924 [ run ] triggered by Bot. Commit: |
|
PR_Github #60925 [ run ] triggered by Bot. Commit: |
|
PR_Github/16709-54905a3 #60924 was force-killed by a newer pipeline run. |
|
PR_Github #60925 [ run ] completed with state |
…VIDIA#16709) Signed-off-by: Yuxian Qiu <142763828+yuxianq@users.noreply.github.com>
Dev Engineer Review
tests/integration/test_lists/test-db/l0_b300.ymlto restructure thel0_b300pre_mergePyTorch attention selection:unittest/_torch/attentionnow excludes three slow CuTe DSL sparse attention tests via--ignore, and the excluded sparse tests are then added back as standalone list entries.l0_b300pre-merge PyTorch test list to preserve coverage while reducing aggregate attention runtime and enabling independent Jenkins sharding of the slow cases.QA Engineer Review
tests/integration/test_lists/test-db/l0_b300.ymlunittest/_torch/attentionrun by adding--ignoreexclusions:unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.pyunittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.pyunittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.pyDescription
Split the three slow CuTe DSL attention test files out of the aggregate B300 attention entry. The aggregate entry ignores those files, and each file remains covered through a standalone CI entry so Jenkins can shard them independently.
The measured serial file profile attributed 2,721 of 4,257 seconds to these three files. Separating them reduces the aggregate attention invocation while preserving test coverage.
Test Coverage
PR Checklist