Please add support for MLX models

[Emasoft]

Please add support for MLX models

[bodaay]

Good news: MLX models already download fine today — the downloader is format-agnostic, so any mlx-community/... (or other MLX) repo works out of the box, e.g.:

hfdownloader download mlx-community/Meta-Llama-3-8B-Instruct-4bit
hfdownloader download mlx-community/Meta-Llama-3-8B-Instruct-4bit:safetensors   # filter
hfdownloader serve   # then paste the repo in the Web UI

Could you clarify what "support" you're after? A couple of possibilities:

1. Just downloading them — already works (see above).
2. Analyzer recognition — having analyze / the Web UI label them as an "MLX" model type (with MLX-specific download recommendations) instead of the generic "Transformers" type. We can add this; the reliable signal is an MLX-style "quantization": { "group_size", "bits" } block in config.json and/or the mlx-community owner.

If it's #2 (or something else), let me know and I'll add it. Thanks!