Issues

[diimdeep]

Hey @Blaizzy, mlx Kokomo model from HF works as expected, that's great but there is few problems

1. Code that supposed to convert from PyTorch is broken
2. from safetensors quantization code from readme is broken.
3. Library code always making network requests downloading the voice file.
4. There is no way to pass local path to model avoiding network calls.

Overall code seems needlessly complicated, vibe-code I mean.