fix torch compile when using fp8 fla #1177
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There was a problem hiding this comment.
Pull Request Overview
This PR fixes a bug in torch compile when using FP8 data types with Flash Attention (FLA). The fix ensures that when QKV tensors are in FP8 format, the output tensor is correctly created with BF16 data type instead of inheriting the FP8 type.
- Added conditional logic to handle FP8 input tensors by creating BF16 output tensors
- Maintains existing behavior for non-FP8 data types
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
Comment on lines
113
to
+123
| else: | ||
| out = torch.empty( | ||
| (batch_size, seqlen_q, num_heads, head_size_v), | ||
| dtype=q.dtype, | ||
| device=q.device, | ||
| requires_grad=q.requires_grad, | ||
| ) | ||
| if q.dtype == dtypes.fp8: | ||
| out = torch.empty( | ||
| (batch_size, seqlen_q, num_heads, head_size_v), | ||
| dtype=dtypes.bf16, | ||
| device=q.device, | ||
| requires_grad=q.requires_grad, | ||
| ) | ||
| else: | ||
| out = torch.empty( | ||
| (batch_size, seqlen_q, num_heads, head_size_v), |
There was a problem hiding this comment.
The nested if-else structure creates duplicated tensor creation logic. Consider restructuring to determine the output dtype first, then create the tensor once to reduce code duplication.
Copilot uses AI. Check for mistakes.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
Found bug when using fp8 fla + torch compile
Technical Details
When datatype for QKV is FP8, datatype for out should be bf16