MuXodious/Qwen2.5-7B-Instruct-1M-Thinking-Claude-Gemini-GPT5.2-DISTILL-PaperWitch-heresy
Abliterated vs DavidAU/Qwen2.5-7B-Instruct-1M-Thinking-Claude-Gemini-GPT5.2-DISTILL. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 23 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 12/84, Attention 21/196, lm_head 0/1