Modern Local Models & Training on Personal Mac MLX

[fire17]

Hi there! Really awesome work!
Was wondering if this will work with more modern agentic local models like different varieties of
Qwen 3.6 or Gemma 4

And also, do you think we will be able to continually train small models using local mlx instead of multiple GPUs any time soon?

Thanks a lot and all the best!