leimroth-lab/Unbound-LR7
Abliterated vs TensorFold/Qwen3.8-Flash-Next-MLX-4bit-MTP. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 29 of 38 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/393, Attention 87/1090, Norms 0/251, MLP 87/1259, other 0/7