[Pytorch] Skip the Single Grouped Param Test if NVTE_GROUPED_LINEAR_SINGLE_PARAM=0

[vthumbe1503]

Description

Please include a brief summary of the changes, relevant motivation and context.

Fixes # (issue)

Type of change

• Documentation change (change only to the documentation, either a fix or a new content)
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Infra/Build change
• Code refactoring

Changes

Please list the changes introduced in this PR:

• Change A
• Change B

Checklist:

• I have read and followed the contributing guidelines
• The functionality is complete
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[vthumbe1503]

/te-ci pytorch

[greptile-apps]

Greptile Summary

Adds a pytest.skip guard to test_sanity_grouped_linear so that single_param=True test variants are skipped unless NVTE_GROUPED_LINEAR_SINGLE_PARAM=1 is set in the environment, preventing spurious assertion failures when the feature flag is absent.

• The guard os.environ.get("NVTE_GROUPED_LINEAR_SINGLE_PARAM", "0") == "0" correctly handles both the unset case (defaults to "0") and the explicitly-disabled case, matching the logic in transformer_engine/pytorch/utils.py and the parallel guard already present in tests/pytorch/test_fusible_ops.py.
• CI (qa/L0_pytorch_unittest/test.sh) already invokes test_sanity.py with NVTE_GROUPED_LINEAR_SINGLE_PARAM=1, so the single-param variants continue to run in the full test suite.

Confidence Score: 5/5

The change is a narrow, well-scoped skip guard in a test file; it cannot affect production code and matches the established pattern already used in the test suite.

The guard correctly defaults to '0' when the env var is absent, so both the unset case and the explicitly-disabled case are handled the same way as utils.py and test_fusible_ops.py. No production code is touched, and CI already sets the flag so the single-param tests still run in the full suite.

No files require special attention.

Important Files Changed

Filename	Overview
tests/pytorch/test_sanity.py	Adds a pytest.skip guard for test_sanity_grouped_linear when single_param=True but NVTE_GROUPED_LINEAR_SINGLE_PARAM is unset or '0', preventing assertion failures from running the single-param path without the required feature flag.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["test_sanity_grouped_linear(single_param=True/False)"] --> B{"NVTE_GROUPED_LINEAR_SINGLE_PARAM == '0'\nAND single_param == True?"}
    B -- Yes --> C["pytest.skip()\n'requires NVTE_GROUPED_LINEAR_SINGLE_PARAM=1'"]
    B -- No --> D["Continue test setup"]
    D --> E["GroupedLinear(\n  single_grouped_weight=single_param,\n  single_grouped_bias=single_param\n)"]
    E --> F{"single_param == True?"}
    F -- Yes --> G["check_grouped_weight()\ncheck_grouped_bias()"]
    F -- No --> H["Run inference + backward"]
    G --> H

Loading

_{Reviews (2): Last reviewed commit: "Apply suggestion from @greptile-apps[bot..." | Re-trigger Greptile}

[timmoon10]

[vthumbe1503]

B200 is the only pending test in CI. And they are passing when manually tested. Merging the PR