Quantization support

[starkdmi]

There is a one line change which allows quantized mlx models in Python version of f5-tts-mlx - source

I tried to patch F5TTS.fromPretrained in Swift to accept quantized weights using quantize(model:,groupSize:,bits:,filter:) on f5tts module but did not succeed.

Any idea how to add support for quantized weights like f5-tts-mlx-4bit and f5-tts-mlx-8bit?

[Victoriaali04]

Hi, I hope you're okay. By chance, did you find out how to support it? The current model causes the app to crash on devices like the iPhone 14.