3MPER0RR/gemma-4-31b-it-3MPER0RR-abliterated
Abliterated vs google/gemma-4-31B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 21 of 60 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/523, other 0/62, MLP 21/261, Attention 20/338
2 derivatives checked
- mradermacher/gemma-4-31b-it-3MPER0RR-abliterated-i1-GGUF: Quantization (fewer layers)
- mradermacher/gemma-4-31b-it-3MPER0RR-abliterated-GGUF: Quantization (fewer layers)