VoxMel-0.6B-MLX
Native Apple Silicon MLX converted weights for VoxDance-Lab/VoxMel-0.6B (post-trained with frozen-head GRPO).
Includes bf16, 8bit (affine group-size 64), and 4bit variants. Use with the mlx-voxmel inference engine.
Hardware compatibility
Log In to add your hardware
Quantized