Use Expand for Attention mask broadcasting instead of Concat #18159
Conversation
This file contains 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
PatriceVignola
deleted the
user/pavignol/improve-attention-mask-broadcasting
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Concatenating the same tensors many times with itself doesn't scale as well as simply broadcasting it using Expand. At least for DirectML, using Expand instead of Concat improves the perf quite a bite, but I can't imagine Concat being faster than Expand in any implementation since Expand can make more assumptions (e.g. it knows it has to deal with a single tensor).
It doesn't affect the GQA path since that path completely gets rid of the mask anyway.