[https://nvbugs/6274932] [fix] Fix and unwaive step3p7 test cases

[kaiyux]

Summary by CodeRabbit

• Bug Fixes

  • Updated accuracy reference data for one model entry.
  • Adjusted test skipping and waiver coverage so the relevant model tests run under the intended hardware conditions.

• Chores

  • Cleaned up obsolete test waiver entries for specific model configurations.

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.

[coderabbitai]

📝 Walkthrough

Walkthrough

For the stepfun-ai/Step-3.7-Flash model, @skip_pre_blackwell is moved from method-level to class-level in both TestStep3_7 test classes (pytorch and multimodal). Nine FP8 block-scale SKIP waive entries for A100, H100, and B200 are removed from waives.txt. A top-level accuracy: 88 field is added to the gsm8k.yaml reference entry.

Changes

Step-3.7-Flash skip promotion and waive cleanup

Layer / File(s)	Summary
Promote @skip_pre_blackwell to class level for TestStep3_7 tests/integration/defs/accuracy/test_llm_api_pytorch.py, tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py	@skip_pre_blackwell is added at the TestStep3_7 class level in both files and removed from the preceding method-level position near test_nvfp4, so the decorator governs the entire class rather than individual methods.
Remove Step3_7 FP8 block-scale waives and add accuracy reference tests/integration/test_lists/waives.txt, tests/integration/defs/accuracy/references/gsm8k.yaml	Nine full:A100, full:B200, and full:H100 Step3_7 FP8 block-scale SKIP waive entries (including multimodal variants) are deleted from waives.txt. A standalone accuracy: 88 top-level field is added to the stepfun-ai/Step-3.7-Flash entry in gsm8k.yaml.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#15504: Modifies the same H100 waiver blocks in waives.txt for Step3_7 FP8 block-scale and multimodal FP8 block-scale entries at overlapping lines.
• NVIDIA/TensorRT-LLM#15535: Changes TestStep3_7 FP8 block-scale waive entries in waives.txt with additions/remappings that directly overlap with this PR's removals.
• NVIDIA/TensorRT-LLM#15581: Adjusts TestStep3_7-related SKIP/waive entries in waives.txt by adding new SKIP entries for test_fp8_block_scales/test_nvfp4, counterpart to this PR's removals.

Suggested reviewers

• StanleySun639
• jieli-matrix
• crazydemo

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description is mostly the template text and does not explain the issue, solution, or test coverage.	Add a short Description section with the problem and fix, and a Test Coverage section listing the tests run or expected to guard the changes.

✅ Passed checks (4 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title includes the NVBugs ID and fix type and accurately reflects the Step3.7 test unwaive and skip changes.

_{✏️ 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.}

[kaiyux]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55468 [ run ] triggered by Bot. Commit: 8675fd2 Link to invocation

[kaiyux]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55488 [ run ] triggered by Bot. Commit: e50bd64 Link to invocation

[tensorrt-cicd]

PR_Github #55468 [ run ] completed with state ABORTED. Commit: 8675fd2

Link to invocation

[tensorrt-cicd]

PR_Github #55488 [ run ] completed with state FAILURE. Commit: e50bd64
/LLM/main/L0_MergeRequest_PR pipeline #44413 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

Link to invocation

[kaiyux]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55643 [ run ] triggered by Bot. Commit: e50bd64 Link to invocation

[tensorrt-cicd]

PR_Github #55643 [ run ] completed with state FAILURE. Commit: e50bd64
/LLM/main/L0_MergeRequest_PR pipeline #44554 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

[kaiyux]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #55751 [ run ] triggered by Bot. Commit: e50bd64 Link to invocation

[tensorrt-cicd]

PR_Github #55751 [ run ] completed with state SUCCESS. Commit: e50bd64
/LLM/main/L0_MergeRequest_PR pipeline #44650 completed with status: 'SUCCESS'

CI Report

Link to invocation