jenerallee78/gemma-4-26B-A4B-it-ara-abliterated
Abliterated vs google/gemma-4-26B-A4B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 11 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 11/231, Norms 0/433, other 0/122, Attention 11/223
2 derivatives checked
- mradermacher/gemma-4-26B-A4B-it-ara-abliterated-i1-GGUF: Quantization
- mradermacher/gemma-4-26B-A4B-it-ara-abliterated-GGUF: Quantization