Support FP8 primary weight in FSDP training

[shjwudp]

Description

This MR modifies the cast_master_weights_to_fp8 function in the FP8 primary weight application, allowing us to use FP8 primary weight in FSDP training.

In FSDP training, the model weight may be incomplete, and model_weight._data may be DTensor(FSDP2) or resized for parameter sharding. We cannot obtain the actual model weight shard address through the slice reading method like model_weight._data.view(-1)[start_offset:end_offset]. This MR extends the cast_master_weights_to_fp8 function to accept the direct input of shard model weight, so that the special use of FSDP can be implemented.

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

[ksivaman]

/te-ci pytorch L0 L1