moutons/Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED-mxfp8-mlx-mlx-fp16
Same weights, converted (fewer layers) vs nightmedia/Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED-mxfp8-mlx. moutons/Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED-mxfp8-mlx-mlx-fp16 matches part of nightmedia/Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED-mxfp8-mlx (677 of its 1010 tensors; the rest of nightmedia/Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED-mxfp8-mlx has no counterpart). On the shared part: Same weights stored in another dtype (U32/BF16 → U32/U8).
Tensors changed: Embeddings 0/2, Norms 0/105, Attention 0/376, MLP 0/192, lm_head 0/2