[https://nvbugs/6388157][fix] [nvbugs/6388157][fix] Unwaive Qwen3-30B-A3B skip-softmax attention test - #15804
Open
tensorrt-cicd wants to merge 1 commit into
Open
[https://nvbugs/6388157][fix] [nvbugs/6388157][fix] Unwaive Qwen3-30B-A3B skip-softmax attention test#15804tensorrt-cicd wants to merge 1 commit into
tensorrt-cicd wants to merge 1 commit into
Conversation
Contributor
|
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 (2)
💤 Files with no reviewable changes (1)
📝 WalkthroughWalkthroughThis PR makes two unrelated edits: it changes the TLLM_CHECK_INFO macro to log failures at debug level instead of info level in trtllmGenExportCompat.h, and it removes one SKIP waiver entry for a specific Qwen3 test case from the integration test waivers list. ChangesLog Level Adjustment
Test Waiver Update
Estimated code review effort🎯 1 (Trivial) | ⏱️ ~3 minutes Possibly related PRs
Suggested reviewers
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
The trtllm-gen FMHA path's TLLM_CHECK_INFO macro fires "Consider using numInstsQ = 2 for better performance." from inside KernelTraits<>'s constructor. Because KernelTraits<> is constructed per kernel selection/launch, on long-context workloads with TRTLLM attention backend (Qwen3-30B-A3B SkipSoftmax + TP4 + AttentionDP running LongBenchV1 4750 samples) the advisory fires hundreds of thousands of times per test. The resulting >100 MB of stdout per worker rank saturates the IPC pipes between MPI workers and the orchestrator, stalling the Fetching responses loop until the PyExecutor HangDetector trips its 300-second watchdog -- surfacing as 'Hang detected on rank 0 in PyExecutor'. The check is a perf hint, not a correctness check, and KernelTraits is not on a path where INFO-level visibility is required. Demote the TLLM_CHECK_INFO macro body from TLLM_LOG_INFO to TLLM_LOG_DEBUG so the hint is still available under opt-in verbose logging but is filtered at the default log level. Also drop the nvbugs/6388157 waiver for the test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False] variant, since this fix resolves it. Signed-off-by: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com>
trtllm-agent
force-pushed
the
repair-bot-bug6388157
branch
from
July 11, 2026 09:19
579803b to
ed252b5
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False]was waived after a flaky hang detected on rank 0 in PyExecutor on DGX B200 4 GPUs. Investigation revealed the underlying hang was caused byTLLM_CHECK_INFOin the FMHA export compat header logging at INFO level, which spammed CI logs on everyKernelTraits<>construction (per kernel launch) and contributed to the apparent rank-0 stall during the sparse attention path.TLLM_CHECK_INFOmacro toTLLM_LOG_DEBUGso the per-launch diagnostic no longer floods CI logs, and remove the waive entry now that the test passes (verified exit 0, 1 passed in 450.81s). The verify monitor timeout was a harness artifact; the test itself had already completed successfully, confirming the fix is sufficient and no further code changes are required.Test plan
Links
Summary by CodeRabbit
Bug Fixes
Tests