saidutta69/SmolLM2-135M-Instruct-heretic
Abliterated vs HuggingFaceTB/SmolLM2-135M-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 18 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/61, MLP 16/90, Attention 15/120