ansulev/lfm2.5-2.6b-fable5-coding-agent-heretic
Abliterated vs AyoubChLin/lfm2.5-2.6b-fable5-coding-agent. Looks like abliteration (refusal removal): only the output projections (out_proj, w2) changed, in 30 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/2, other 22/66, MLP 25/90, Norms 0/76, Attention 6/32