[NVIDIA] Add a custom layer for cudnn flash attention #53
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
kaixih
force-pushed
the
cudnn_attention_dev
branch
from
a55f006 to
61beeb9
Compare
|
FYI, the prerequisite JAX PR is now merged. |
kaixih
force-pushed
the
cudnn_attention_dev
branch
from
61beeb9 to
79f9279
Compare
|
Any update? It seems the PR is in pull ready state for a while. @zhangqiaorjc |
kaixih
force-pushed
the
cudnn_attention_dev
branch
from
79f9279 to
d8d440d
Compare
|
The sdpa is now in the jax public module (see PR). So, we modify this PR to use the public API. |
|
@kaixih could you add a config option in MoE transformer that preserves the old behavior by default i don't want to break existing checkpoints |
|
Can you articulate why this PR breaks the default behavior of MoE transformer? I think this PR just added a new layer of attention and it will be used only when users config it to the model with USE_CUDNN_FLASH_ATTENTION from this PR. |
|
Sorry i meant to reply to the other PR #80 |
|
Gentle ping. @zhangqiaorjc |
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Outdated suggestions cannot be applied.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 a new layer for the cudnn flash attention and so that users can config it into
tr_atten_tplof the Transformer layer.This PR depends on this Jax PR: jax-ml/jax#20380
cc. @nluehr @zhangqiaorjc