mlx-vlm v0.1.4 is here 🎉

New models:
- @GoogleDeepMind Paligemma 2

Up next 🚧:
- Refactoring

Get started:
> pip install -U mlx-vlm

Please leave us a star and send a PR :)

github.com/Blaizzy/mlx-...

After some changes,

@hf.co SmolVLM now does +110 tok/s on M1 Max using MLX🚀

Thanks Awni from Apple for some of the suggestions ❤️

@hugg Idefics 3 and SmolVLM now on MLX 🎉🚀

You can now run inference and fine-tune locally on your Mac.

pip install -U mlx-vlm

I’m getting ~140 tok/s on M3 Max 96GB 🔥

Thanks to @pcuenq.hf.co for PR!

Model Cards 👇🏽

mlx-vlm v0.1.1 is here 🎉🔥

New models🤖:
- Allen AI Molmo
- Microsoft Florence 2

Changes 🚀:
- Fixed Pixtral image prompt h/t Nils
- 30-60% faster Qwen2-VL inference h/t Awni
- Fixed Qwen2-VL OCR
- Skip quant for vision encoder or layers.
- New notebooks

Please leave us a star and send a PR❤️