TransformerEngine icon indicating copy to clipboard operation
TransformerEngine copied to clipboard

Enable AttnFuncWithCPAndKVP2P to support mla

Open SuperCB opened this issue 1 year ago • 0 comments

Description

The function AttnFuncWithCPAndKVP2Pdo not support mla(Multi-latent attention), because it concats K and V into a single tensor for communication, different head_dim of K and V prevents us from doing the concat.I attempted to pad v to align its head dimension with that of k, thereby enabling support for mla with minimal modifications. 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)
  • [x] 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:

Checklist:

  • [x] 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
  • [x] 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

SuperCB avatar Mar 12 '25 06:03 SuperCB