[Google's Flan-T5 medium](https://huggingface.co/google/flan-t5-small) converted to [MLX](https://github.com/ml-explore/mlx-examples) format and quantized up to 4 bits with group size 64.