Sorihon/KansenSakura-Erosion-RP-12B-heretic

Abliterated vs Retreatcost/KansenSakura-Erosion-RP-12b. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 32 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.

Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/81, MLP 3/120, Attention 32/160

2 derivatives checked

On the Hub · Downloads on Model Pulse