WWTCyberLab/gemma-4-26B-A4B-it-abliterated
Abliterated vs google/gemma-4-26B-A4B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 18 of 30 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 36/231, Norms 0/433, other 0/122, Attention 18/223
3 derivatives checked
- mradermacher/gemma-4-26B-A4B-it-abliterated-GGUF: Quantization
- mradermacher/gemma-4-26B-A4B-it-abliterated-i1-GGUF: Quantization
- Vastopian/gemma-4-26B-A4B-it-abliterated-GGUF: Quantization