mlx-community/MOSS-Image-SFT-bf16
Text-to-Image • Updated
MLX snapshots of OpenMOSS MOSS-Image (5B DiT, Qwen3.5-9B encoder, FLUX.2 small decoder) for the Swift/MLX port.
Note Recommended: supervised fine-tune — cleaner text grouping and composition. int8 DiT at load.
Note Pretrained base the SFT starts from; LoRA-training substrate.