heretic-org/IBM-granite-4.1-8b-heretic

Abliterated vs ibm-granite/granite-4.1-8b. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 13 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.

Tensors changed: Embeddings 0/1, Norms 0/81, MLP 6/120, Attention 7/160

2 derivatives checked

On the Hub · Downloads on Model Pulse