byroneverson/gemma-2-27b-it-abliterated
Abliterated vs google/gemma-2-27b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 46 of 46 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/185, MLP 46/138, Attention 46/184
5 derivatives checked
- mradermacher/gemma-2-27b-it-abliterated-i1-GGUF: Quantization
- QuantFactory/gemma-2-27b-it-abliterated-GGUF: Quantization
- mradermacher/gemma-2-27b-it-abliterated-GGUF: Quantization
- tensorblock/gemma-2-27b-it-abliterated-GGUF: Quantization
- byroneverson/gemma-2-27b-it-abliterated-gguf: Quantization