Add Unified Sequence Parallel attention #12693
Open
+270
−6
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
Member
|
It would be nice to get a testing script so that we can quickly check things. |
|
I added a basic test script with a simple forward and backward op. Is it better to have a test script with flash_attention_backward and forward?? |
Bissmella
force-pushed
the
unified-SP-attention
branch
from
a244006 to
9dee8f8
Compare
bug fixes, lse calculation - switched to _all_to_all_single helper in _all_to_all_dim_exchange due contiguity issues bug fix bug fix bug fix
Bissmella
force-pushed
the
unified-SP-attention
branch
from
9dee8f8 to
9ebcff5
Compare
Member
|
Let us know if this is ready for a review! |
Author
|
Yep, ready for review! I tested it with a 4-process setup (2×2 mesh, on cpu) and everything checks out, shapes look good and gradients flow correctly. Looking forward for feedback and happy to address any issues. |
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Outdated suggestions cannot be applied.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.
What does this PR do?
This is a draft implementation of the Unified SP attention approach.
_all_to_all_dim_exchangewith custom scatter and gather indicesTemplatedUnifiedAttentionCore implementation complete, needs: