[PyTorch] Add sliding window support to FlashAttention - #551
Merged
Conversation
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci |
Collaborator
Author
|
/te-ci |
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
cyanguwa
force-pushed
the
fa/sliding_window
branch
from
December 7, 2023 23:41
49e147e to
32db392
Compare
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
1 similar comment
Collaborator
Author
|
/te-ci pytorch |
Collaborator
Author
|
Pipeline 11331817 |
Collaborator
Author
|
With newer cuDNN, pipeline 11412889 is green! |
Signed-off-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
ptrendx
reviewed
Dec 15, 2023
ptrendx
reviewed
Dec 15, 2023
ptrendx
reviewed
Dec 15, 2023
ptrendx
reviewed
Dec 15, 2023
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
ptrendx
reviewed
Dec 15, 2023
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Collaborator
Author
|
/te-ci pytorch |
|
Also want to add a comment that what happens when we want to use grouped query attention with unfused attention. Right now it seems like it errors out. |
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
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.
This PR only makes changes on the PyTorch side. It
TransformerLayer,MultiHeadAttention,DotProductAttentionandFlashAttentionUnfusedDotProductAttentionarbitrary mask, generated based on the window sizeuse_unfused_attentionflag and exception when none of the three DPA backends are available