Converted from alphaZimuth/Hy-MT2-1.8B-SpecDraft-30B/Hy-MT2-1.8B-SpecDraft-30B-Q4_K_M.gguf for MLX/oMLX speculative decoding.

The checkpoint is a standard tied-embedding hunyuan_v1_dense model with the source GGUF precision tiers mapped to MLX affine modules using group size 64. Quantization parameters and ordinary floating weights use BF16.

Source tensor types: {'F32': 129, 'Q4_K': 192, 'Q6_K': 33}.