MuXodious/gemma-4-E4B-it-ARA-heresy
Abliterated vs google/gemma-4-E4B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 17 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 17/606, Norms 0/502, other 0/261, Attention 17/755, Embeddings 0/6
2 derivatives checked
- mradermacher/gemma-4-E4B-it-ARA-heresy-i1-GGUF: Quantization (fewer layers)
- mradermacher/gemma-4-E4B-it-ARA-heresy-GGUF: Same weights, converted (fewer layers)