saidutta69/Mistral-7B-Instruct-v0.3-heretic

Abliterated vs mistralai/Mistral-7B-Instruct-v0.3. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 25 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.

Tensors changed: Embeddings 0/1, Norms 0/65, MLP 14/96, Attention 25/128, lm_head 0/1

On the Hub · Downloads on Model Pulse