A Moe model built on top of Qwen1.5-7B-Chat, Qwen1.5-7B and Crystalcareai/CrystalQwen-1.5-7B.

```
pip install mlx-moe

python -m mlx_moe.generate --model mzbac/qwen-1.5-2x3-hf-4bit-mlx --prompt "how back prop works"

```