Summary

Fixes a bug where post_all_gather_processing created transpose for Float8Tensor weights even when keep_fp8_weight_transpose_cache=False, leading to assertion failures in Linear forward.

Problem

With keep_fp8_weight_transpose_cache=False, quantizer.columnwise_usage is set to False (e.g. on ROCm/AMD). post_all_gather_processing was still creating transpose via _create_transpose() because it did not respect columnwise_usage. This triggered the assertion in Linear forward: expected _transpose to be None or an empty tensor when transpose cache is disabled.

Solution

• Update post_all_gather_processing in utils.py so it only creates transpose when model_weight._quantizer.columnwise_usage is True.
• Parametrize test_cast_master_weights_to_fp8 with keep_fp8_weight_transpose_cache=True and False to cover both cases.

Testing

• test_cast_master_weights_to_fp8 with keep_fp8_weight_transpose_cache=False now passes.

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

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