NullpoLab/Agents-A1-4B-Heretic-ARA-Refusals0
Abliterated vs InternScience/Agents-A1-4B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 18 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 16/320, MLP 17/196, Embeddings 0/4, Norms 0/203
1 derivatives checked
- NullpoLab/Agents-A1-4B-Heretic-ARA-Refusals0-GGUF: Same weights, converted