KDA in torchtitan - #4115
Closed
liangel-02 wants to merge 1 commit into
Closed
Conversation
drisspg
reviewed
Aug 11, 2026
| pillow | ||
| torchvision | ||
| flash-linear-attention | ||
| attention-gym |
Contributor
There was a problem hiding this comment.
the build release flow is pretty easy and the currently published one is pretty old, before land we shoudl push a new one when we have all the features we need
drisspg
reviewed
Aug 11, 2026
drisspg
reviewed
Aug 11, 2026
drisspg
reviewed
Aug 11, 2026
drisspg
reviewed
Aug 11, 2026
drisspg
reviewed
Aug 11, 2026
tianyu-l
reviewed
Aug 12, 2026
| pillow | ||
| torchvision | ||
| flash-linear-attention | ||
| attention-gym |
Contributor
There was a problem hiding this comment.
should we always ping an version; o/w attention-gym needs to promise never BC breaking?
Contributor
Author
There was a problem hiding this comment.
yes will pin version once we create a new release for attention-gym
liangel-02
force-pushed
the
kda
branch
4 times, most recently
from
August 14, 2026 21:09
840cd50 to
531bb2b
Compare
Contributor
Contributor
Author
|
closing for ghstack --> #4164 |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 adds attention module for KDA (relying on attention gym KDA kernels and user facing APIs) and also builds out the skeleton for vLLM
testing is limited to unit tests and also relying on upstream attn-gym tests for kernel correctness
waiting on model impl to run integration tests