Add Flash Attention v2 to Ops #1970
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
ptillet
approved these changes
Jul 20, 2023
|
It seems like the performance significantly decreased. Not sure why 🤔 May be worth comparing the TTIR and TTGIR of the tutorial vs what the ops generate |
|
I think the main change in the forward was the switch back from the MODES to the CAUSAL in 1 pass. I think converting the MODE==3 to FA v2 might require a little work. I'll take a look at the ttir/ttgir as well as the MODEs |
This was referenced Jul 20, 2023
|
Eh, I think this PR forgets to update the block size and |
|
Yep, you are correct, I adjusted blocks and stages, locally on my A6000 forward perf is much improved. |
|
Updated the new forward pass numbers for FA based on CI. |
pingzhuu
pushed a commit
to siliconflow/triton
that referenced
this pull request
Apr 2, 2024
I also dropped the do_scaled as it is no longer needed (no scaling done to the do in v2). --------- Co-authored-by: Philippe Tillet <phil@openai.com>
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Outdated suggestions cannot be applied.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.
I also dropped the do_scaled as it is no longer needed (no scaling done to the do in v2).