igorls/gemma-4-12B-it-heretic-v1
Abliterated vs google/gemma-4-12B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 16 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/12, Norms 0/289, other 0/48, MLP 13/144, Attention 9/184
2 derivatives checked
- igorls/gemma-4-12B-it-heretic-GGUF: Quantization
- IvanSmit05/gemma-4-12B-it-heretic-mlx-4bit: Quantization