-
Notifications
You must be signed in to change notification settings - Fork 459
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add FlashAttention #24
Merged
Merged
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
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Outdated suggestions cannot be applied.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.
Closes #23
This is blocked right now on Dao-AILab/flash-attention#132(update: I got this to work with an older version of triton, we'll keep tracking that issue and update triton when it's fixed - #26)Uses the official FlashAttention implementation. I've prebuilt a Python 3.10, PyTorch 1.13.1, CUDA 11.7 wheel for this which you can install with:
This will be built in to the Docker images.
For now we'll have to use their triton version since the CUDA version doesn't support arbitrary attention biases, meaning we can't use ALiBi.
The advantages and disadvantages of the Triton implementation are discussed here:
https://github.com/HazyResearch/flash-attention/blob/57ee618170e1adecbf787365cdf330c63768abd2/flash_attn/flash_attn_triton.py#L1-L35
And to add to that, one of our contacts at MosaicML says:
This PR is based on: