[JAX] THD ring attention

[zlsh80826]

Description

Support P2P context parallel (ring attn) with THD format. This feature is only available for self attn + causal + segment_ids/pos + load balancing (reorder before the attn and inverse-reorder after the attn).

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

• Refactor reorder/inverse_reorder_causal_loading_balancing API to support different reorder strategy.
• Support P2P context parallel. The limitations are list above.
• Reduce the number of test configs of kv_groups in test_distributed_fused_attn
• Use AttnBiasType, AttnMaskType, QKVLayout in cpp_extenion/attention.py for maintaining the readibility.

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

[zlsh80826]

/te-ci jax L1

[zlsh80826]

/te-ci jax L1