llmfan46/gemma-4-12B-it-qat-q4_0-unquantized-uncensored-heretic
Abliterated vs google/gemma-4-12B-it-qat-q4_0-unquantized. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 11 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/12, Norms 0/289, other 0/48, MLP 0/144, Attention 11/184
7 derivatives checked
- llmfan46/gemma-4-12B-it-qat-q4_0-uncensored-heretic-GGUF: Quantization
- ailexleon/gemma-4-12B-it-qat-uncensored-heretic-mlx-4Bit: Quantization
- mradermacher/gemma-4-12B-it-qat-q4_0-unquantized-uncensored-heretic-i1-GGUF: Quantization
- llmfan46/gemma-4-12B-it-qat-q4_0-uncensored-heretic-NVFP4-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-4-12B-it-qat-q4_0-unquantized-uncensored-heretic-GGUF: Quantization
- EZForever/gemma-4-12B-it-qat-uncensored-heretic-UDmerge-GGUF: Quantization
- claymorecrystal/gemma-4-12B-it-qat-q4_0-uncensored-heretic-GGUF: Quantization