Initial support

[dc-dc-dc]

No description provided.

[simonw]

I can't wait to try this out, I want to run https://huggingface.co/microsoft/Phi-3-mini-128k-instruct-onnx on my Mac!

[awni]

Just curious, why not use this one https://huggingface.co/microsoft/Phi-3-mini-128k-instruct with MLX LM? That should already work quite well.

[csukuangfj]

1 month has passed. Is there any update?

[awni]

Can you say more about your use case for this @csukuangfj ?

[csukuangfj]

I want to create sherpa-mlx similar to sherpa-onnx.

But the first obstacle is how to convert models from onnx to mlx.

sherpa-onnx supports different types of models, e.g.

• zipformer, paraformer, whisper, etc., for speech recognition
• vits for text-to-speech
• 3d-speaker, wespeaker, models from NeMo, for speaker embedding extractor

If we can convert models from onnx to mlx, then it should be straightforward to create sherpa-mlx.

[awni]

Makes sense, thanks! This PR needs a bit of work before we can land it. I don't have an ETA yet. But its good to know of your use case to help prioritize.

[csukuangfj]

Almost two months have passed.

Is there any update?

[awni]

No update here yet, sorry 😔