Add experimental code for custom attention op #81
+141
−0
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
jpienaar
reviewed
Jun 11, 2024
| iree_attention_p = core.Primitive('iree_attention') | ||
| iree_attention_p.def_impl(partial(dispatch.apply_primitive, iree_attention_p)) | ||
|
|
||
| transpose_v = False |
| input_shapes = [ | ||
| jax.ShapeDtypeStruct(input.shape, input.dtype) for input in inputs | ||
| ] | ||
| att = export.export( |
There was a problem hiding this comment.
Should this be some generic IREE export somewhere? E.g., have list of "allowed for IREE export" custom calls listed somwehere. Then one always does iree.jax.export instead of jax.experimental.export.
| ###################### | ||
|
|
||
|
|
||
| def iree_attention( |
There was a problem hiding this comment.
Is this the one users would then call explicitly? So one needs to change the model to use it? (potentially wrapped inside some function which dispatches either to the IREE one or a generic one so that they could still test without IREE)
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Outdated suggestions cannot be applied.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 code can (and should) be expanded upon, and would ideally be integrated in the IREE-Jax workflow. At the moment the workflow is broken since Jaxlib has removed ml_program from their Python deps, so I created a new folder for work that doesn't quite fit anywhere.
This is followed by an IREE-side PR that lowers this custom op to IREE attention.
At the moment this is a 1:1 mapping to the IREE version of the op, but it can be expanded upon as needed.