MLX model support #124
Open
MLX model support #124
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
* Added MLX feature extractor implementation * CI fix * added better multiarray conversion * CI fix * CI fix * fixed `asMLMultiArray` implementation, fixed CI * update xcode, trigger pr when targeting not main branch * check if vision os builds * update watch os version * conditional watchos compilation * conditional package.swift * conditional package.swift * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * ci fix * add other tests targest back * package.swift cleanup * general cleanup * revert to xcode 15.2
…into mlx-support
* added mlx audio encoder * fixed model protocols * removed not needed
|
@jkrukowski relevant: https://ml-explore.github.io/mlx/build/html/install.html#binary-size-minimization We'll want to minimize the binary size as much as reasonable for iOS deployment |
…into mlx-support
* fixes for mlx models * fixed asMLXArray * fixed tests, mlx doesn't run on simulators * fix
* Added more tests for MLX, cleanup * bumped timeout * fixed tests * reverted cache id
* update mlx-swift * - reverted mlx version - updated readme - updated makefile * reversed * fixed tests * updated mlx-swift * updated makefile * remove device change, fft can run on gpu now * updated readme, added tests * updated readme * review changes * review changes * CI model cache path fix * tests failed * Update package.swift * Keep setupModels with adjustments * Test CI skip cache * Test CI package name change * Test CI optional CLI settings * Use correct logits size for MLX --------- Co-authored-by: ZachNagengast <[email protected]>
atiorh
reviewed
Nov 17, 2024
| } | ||
|
|
||
| private func qkvAttention(_ q: MLXArray, _ k: MLXArray, _ v: MLXArray, _ mask: MLXArray?) -> (MLXArray, MLXArray) { | ||
| let (nBatch, nCtx, nState) = (q.shape[0], q.shape[1], q.shape[2]) |
There was a problem hiding this comment.
|
is this still on the roadmap? |
|
@neo773 Yes it is, we plan to finalize this before v1.0.0 sometime early this year. It is one of the last remaining pieces of feature work to finish up for the v1.0.0 and we will also refine a lot of the docs and do a bit of refactoring for a stable, production-ready release. |
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.
Draft PR for the early stages of supporting MLX based whisper models directly in WhisperKit. (To be updated)
Initial TODOs:
FeatureExtractingprotocolAudioEncodingprotocolTextDecodingprotocol orTextDecodersubclass (research needed)Compiling whisperkit-cli optionally with/without MLXCrossing this out for now: whisperkit-cli does not have the same bundle size constraints as the core libraryxcodebuildinstead of iOS simulator (simulator is not supported for MLX)