[Qwen] avoid creating attention masks when there is no padding #12987
+183
−153
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
6 tasks
|
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. |
Contributor
Author
|
will do! |
yiyixuxu
reviewed
Jan 17, 2026
| encoder_attention_mask = torch.stack( | ||
| [torch.cat([u, u.new_zeros(max_seq_len - u.size(0))]) for u in attn_mask_list] | ||
| ) | ||
| prompt_embeds, encoder_attention_mask = build_prompt_embeds_and_mask(split_hidden_states) |
Collaborator
There was a problem hiding this comment.
curiously, cant we just do
if encoder_attention_mask.all():
encoder_attention_mask = None
Contributor
Author
There was a problem hiding this comment.
the issue was that this was giving trouble with torch.compile...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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Outdated suggestions cannot be applied.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 do?
We now have a unified Qwen prompt-embed handling to avoid creating attention masks when there is no padding (keeps flash SDPA eligible, removes unnecessary masks). This fixes the inference speed regression.
Before submitting
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.