Add MLX 2-bit quantized model with KV cache compression 0f832ed verified majentik commited on 3 days ago