Awni Hannun

Very nicely done! I just left a couple minor comments. Please check them and then we can merge it!

I would recommend you file an issue to [lightning-whisper-mlx](https://github.com/mustafaaljadery/lightning-whisper-mlx) unless there is something we need to change here to enable support of that?

We have these [now via the CLI](https://github.com/ml-explore/mlx-examples/tree/main/whisper#cli)

That's not expected.. I tried doing it both ways with some adapters I just tuned and it works fine (as expected given they both use the same `load` function under...

Probably you need to use the chat template when you are using the API. See the [docs here](https://github.com/ml-explore/mlx-lm?tab=readme-ov-file#python-api)

Interesting.. I've seen that behavior before in lower quality models. Two questions: 1. Are you using 16-bit or 32-bit precision? 2. Did you try the PyTorch implementation on the same...

The repetition problem is a common problem with encoder-decoder style models. Though it usually becomes vanishingly rare for high quality models. Indeed it could be that edge case inputs are...

~There is an option that I don't think is currently implemented in the MLX example:~ Edit: it is actually implemented and should be enabled by default. ```text condition_on_previous_text: bool if...

It should be possible. There is a port of Suno's Bark model already: https://github.com/j-csc/mlx_bark I think it still depends on PyTorch for the encodec model though.

Is it one of the examples in this repo? Can you share anything else? There shouldn't be a segfault from MLX, so if you are getting one from MLX it...