Take token count quantization of fused attention into consideration for CP results correction

[xrennvidia]

Description

Fused attention has token count quantization. This can make T of sfotmax_lse mismatch with Q and O. For example, both T=1000 and T=500 will create T=1024 in softmax_lse, so we cannot assume half of T will have 2x smaller shape of softmax_lse. Fix CP results correction by taking this quantization into consideration.

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 refractor

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

[xrennvidia]

/te-ci pytorch L1

[cyanguwa]

LGTM