fix-causal-fa-infer

[hsiehjackson]

What does this PR do ?

Previously, we check is_causal for FA if we have self.attn_mask_type == AttnMaskType.causal and s_q == s_k. However, we may need causal self-attention for s_q != s_k if we have context parallelism proposed by @ksivaman . I change the condition s_q == s_k to inference_mode in order to prevent using causal when we have s_q==1 during inference.

Collection: [Note which collection this PR will affect]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this 

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[ksivaman]

Wouldn't we ideally want this same behavior (i.e. no causal mask) for the validation part and not just inference?

[hsiehjackson]

@ksivaman our inference already including validation, test, and predict.