rapid-mlx/Qwen3.8-27B-4bit-MTP-fp16-MLX
Same weights, converted vs rapid-mlx/Qwen3.8-27B-4bit-MTP-MLX. Same weights stored in another dtype (U32/BF16 → U32/F16).
Tensors changed: Embeddings 0/6, Norms 0/319, Attention 0/1164, MLP 0/688, lm_head 0/3
Same weights, converted vs rapid-mlx/Qwen3.8-27B-4bit-MTP-MLX. Same weights stored in another dtype (U32/BF16 → U32/F16).
Tensors changed: Embeddings 0/6, Norms 0/319, Attention 0/1164, MLP 0/688, lm_head 0/3