[TRTLLM-13600][test] Clean up Qwen3 test cases

[nv-guomingz]

Summary by CodeRabbit

• Bug Fixes
  • Updated several Qwen3 accuracy test scenarios to use more consistent batch sizing and CUDA graph settings, improving stability across model variants.
  • Adjusted coverage for FP8, NVFP4, and sparse attention paths, including additional Qwen3 8B and 30B cases.
  • Refined test selection across different GPU environments so the right model variants run in the right merge stages.

Description

Test Coverage

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[nv-guomingz]

/bot run --disable-fail-fast

[nv-guomingz]

/bot run --disable-fail-fast

[coderabbitai]

📝 Walkthrough

Walkthrough

Accuracy tests for Qwen3-8B, Qwen3-30B-A3B, and Qwen3-235B-A22B are updated to remove cuda_graph/overlap_scheduler pytest parametrization axes and standardize CUDA graph construction to CudaGraphConfig(enable_padding=True, max_batch_size=32) with max_batch_size=32 in LLM constructors. The TestQwQ_32B class is removed. CI test list YAML files for B200, DGX-B200, DGX-B300, and H100 are updated to reflect the revised parametrization IDs and redistribute test cases between pre/post-merge conditions.

Changes

Qwen3 accuracy test parametrization and CUDA graph config standardization

Layer / File(s)	Summary
TestQwen3_8B: CUDA graph and parametrization standardization tests/integration/defs/accuracy/test_llm_api_pytorch.py	test_fp8_block_scales removes cuda_graph/overlap_scheduler parametrization and sets CudaGraphConfig(enable_padding=True, max_batch_size=32); test_bf16 and test_fp8_block_scales_early_first_token_response add padding config; test_eagle3 updates conditional CudaGraphConfig and max_batch_size=32 in LLM; test_w4a8_mxfp4 introduces activation_dtype parametrization and standardizes CudaGraphConfig.
TestQwen3_30B_A3B: parametrization reduction and config standardization tests/integration/defs/accuracy/test_llm_api_pytorch.py	test_fp8_block_scales and test_fp8 remove cuda_graph/overlap_scheduler axes and use CudaGraphConfig(enable_padding=True, max_batch_size=32) with torch_compile_config; test_nvfp4 removes those params and passes fixed CudaGraphConfig; test_w4a8_mxfp4 adds activation_dtype and sets enable_attention_dp=False; test_w4a16_mxfp4 reduces parametrization and standardizes batch size.
TestQwen3_235B_A22B: parametrization changes and TestQwQ_32B removal tests/integration/defs/accuracy/test_llm_api_pytorch.py	test_fp8_block_scales removes cuda_graph/overlap_scheduler and builds pytorch_config with CudaGraphConfig() and moe_config from attention_dp/moe_backend; test_nvfp4 adds eagle3 boolean and uses fixed CudaGraphConfig with moe_config; TestQwQ_32B class definition is deleted.
l0_b200.yml: pre/post-merge Qwen3 test list updates tests/integration/test_lists/test-db/l0_b200.yml	Pre-merge removes old MoE latency/w4a8_mxfp4/w4a16_mxfp4 entries and adds trtllm-focused cases; post-merge adds TestQwen3_8B w4a8_mxfp4, TestQwen3_30B_A3B nvfp4 latency entries, w4a8_mxfp4 CUTLASS, and w4a16_mxfp4 TRITON.
DGX-B200 and DGX-B300 test list migration tests/integration/test_lists/test-db/l0_dgx_b200.yml, tests/integration/test_lists/test-db/l0_dgx_b300.yml	DGX-B200 moves three dep4_latency_moe test_nvfp4 entries to a later condition block; DGX-B300 moves tep4_latency_moe_trtllm-torch_compile=False from pre-merge to post-merge.
l0_h100.yml: eagle3 and fp8 test redistribution tests/integration/test_lists/test-db/l0_h100.yml	Pre-merge removes test_fp8[latency-torch_compile=False] and prunes test_eagle3 to one entry; post-merge adds back three test_eagle3 parametrization combinations and re-adds test_fp8[latency-torch_compile=False].

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#15544: Applies the same pattern of removing cuda_graph/overlap_scheduler from parametrization and standardizing CudaGraphConfig(enable_padding=True, max_batch_size=32) in Qwen3-related accuracy tests.

Suggested reviewers

• xinhe-nv
• QiJune
• YihuiLu512
• sunnyqgg

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The PR description is mostly the template only; it lacks the required issue, solution, test coverage, and checklist details.	Fill in the Description and Test Coverage sections, and complete the PR Checklist with the relevant validation and review notes.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is concise and clearly matches the Qwen3 test cleanup changes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/integration/defs/accuracy/test_llm_api_pytorch.py`:
- Around line 4397-4412: `test_fp8_block_scales` no longer exercises the
CUDA-graph path because `cuda_graph` is effectively fixed to `False`, so the
`CudaGraphConfig` branch is dead. Update `test_fp8_block_scales` in
`test_llm_api_pytorch.py` to either restore a parameterized CUDA-graph case that
actually instantiates `CudaGraphConfig` or remove
`cuda_graph`/`cuda_graph_config` from the test and rename the test to reflect
its real coverage, keeping the `LLM` setup and `pytorch_config` aligned with the
intended behavior.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: cd816e26-864d-41e2-b49c-6fbfe0f5e492

📥 Commits

Reviewing files that changed from the base of the PR and between 71613f9 and 939f9dd.

📒 Files selected for processing (5)

• tests/integration/defs/accuracy/test_llm_api_pytorch.py
• tests/integration/test_lists/test-db/l0_b200.yml
• tests/integration/test_lists/test-db/l0_dgx_b200.yml
• tests/integration/test_lists/test-db/l0_dgx_b300.yml
• tests/integration/test_lists/test-db/l0_h100.yml

[tensorrt-cicd]

PR_Github #55491 [ run ] triggered by Bot. Commit: 939f9dd Link to invocation

[tensorrt-cicd]

PR_Github #55492 [ run ] triggered by Bot. Commit: 939f9dd Link to invocation

[tensorrt-cicd]

PR_Github #55492 [ run ] completed with state FAILURE. Commit: 939f9dd
/LLM/main/L0_MergeRequest_PR pipeline #44417 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[nv-guomingz]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55648 [ run ] triggered by Bot. Commit: 939f9dd Link to invocation

[tensorrt-cicd]

PR_Github #55648 [ run ] completed with state FAILURE. Commit: 939f9dd
/LLM/main/L0_MergeRequest_PR pipeline #44559 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[nv-guomingz]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55724 [ run ] triggered by Bot. Commit: 939f9dd Link to invocation

[tensorrt-cicd]

PR_Github #55724 [ run ] completed with state SUCCESS. Commit: 939f9dd
/LLM/main/L0_MergeRequest_PR pipeline #44626 completed with status: 'SUCCESS'

CI Report

Link to invocation