ymcki/gemma-2-2b-jpn-it-abliterated-24
Abliterated vs google/gemma-2-2b-jpn-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 26 of 26 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/105, MLP 26/78, Attention 26/104
2 derivatives checked
- mradermacher/gemma-2-2b-jpn-it-abliterated-24-i1-GGUF: Quantization
- mradermacher/gemma-2-2b-jpn-it-abliterated-24-GGUF: Same weights, converted