Fix Qwen2Audio flash attention mask format for generation #41843
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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
@remi-or ready for review 🙂 |
vasqu
reviewed
Oct 24, 2025
vasqu
reviewed
Oct 24, 2025
|
run-slow: qwen2_audio |
|
This comment contains run-slow, running the specified jobs: models: ['models/qwen2_audio'] |
vasqu
approved these changes
Oct 24, 2025
|
Can you push an empty commit? Cant merge with red CI |
|
[For maintainers] Suggested jobs to run (before merge) run-slow: qwen2_audio |
vasqu
reviewed
Oct 24, 2025
| dummy_embeds = torch.zeros( | ||
| (batch_size, max_seq_len, 1), | ||
| dtype=self.audio_tower.conv1.weight.dtype, | ||
| device=self.audio_tower.conv1.weight.device, |
There was a problem hiding this comment.
Ah sorry maybe one last nit: can we change the device/dtype here? Inputs_embeds should suffice?
|
Thx a lot 🤗 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.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.
What does this PR fix?
This PR fixes the
test_eager_matches_fa2_generatetest failure for Qwen2Audio by using thecreate_bidirectional_maskutility function to properly handle attention masks across different attention implementations.The Qwen2Audio model was manually creating a 4D attention mask with
-infvalues for the audio encoder, regardless of the attention implementation being used. This caused issues with Flash Attention 2/3, which requires a 2D boolean mask (shape(batch_size, seq_len)) with1for valid tokens and0for padding.