Add Whisper model family#192
Merged
Merged
Conversation
Supports every openai/whisper-* size (tiny → large-v3-turbo, plus .en variants) in both HuggingFace and OpenAI/mlx-whisper checkpoint layouts. Streaming via decode-and-diff, 30-second chunked decoding for long audio, tokenizer fallback fetch when mlx-community repos ship weights only. GLMASR's Whisper* layer types are renamed to GLMASRWhisper* to free the canonical names for the new model.
lucasnewman
previously approved these changes
Jun 5, 2026
lucasnewman
left a comment
lucasnewman
left a comment
Collaborator
There was a problem hiding this comment.
Thanks, this looks great!
Signed-off-by: Lucas Newman <lucas@future.fit>
lucasnewman
approved these changes
Jun 5, 2026
lucasnewman
approved these changes
Jun 12, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
WhisperModelinSources/MLXAudioSTT/Models/Whisper/conforming toSTTGenerationModel.openai/whisper-*size (tiny → large-v3-turbo, plus.en) andmlx-community/whisper-*mirrors. Both the HuggingFacetransformerslayout and the OpenAI / mlx-whisper layout (encoder.blocks.*,n_audio_state, etc.) load through the same path.openai/whisper-*repo on demand (no weight re-download).WhisperConfig/WhisperAttention/WhisperEncoder/WhisperEncoderLayertoGLMASRWhisper*(the SmartTurn module already follows this prefix style) to free up the canonical names.Test plan
xcodebuild build -scheme MLXAudioSTT— cleanxcodebuild test … -only-testing:MLXAudioTests/WhisperTests -only-testing:MLXAudioTests/GLMASRModuleSetupTests— 26 tests passopenai/whisper-tiny/openai/whisper-tiny.en/mlx-community/whisper-large-v3-turboon short + 66 s tiled clips, English + French + 中文 (Breeze-ASR-25 fine-tune)