llmfan46/gemma-4-12B-it-uncensored-heretic
Abliterated vs google/gemma-4-12B-it. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 16 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 16/184
8 derivatives checked
- llmfan46/gemma-4-12B-it-uncensored-heretic-GGUF: Same weights, converted
- mradermacher/gemma-4-12B-it-uncensored-heretic-GGUF: Quantization
- mradermacher/gemma-4-12B-it-uncensored-heretic-i1-GGUF: Quantization
- FreedomAISVR/Gemma-4-12B-it-Uncensored-Heretic-NVFP4-GGUF: Quantization
- llmfan46/gemma-4-12B-it-uncensored-heretic-NVFP4-GGUF: Same weights, converted (fewer layers)
- claymorecrystal/gemma-4-12B-it-uncensored-heretic-GGUF: Same weights, converted
- llmfan46/gemma-4-12B-it-uncensored-heretic-NVFP4: Quantization
- FreedomAISVR/Gemma-4-12B-it-Uncensored-Heretic-MXFP4-GGUF: Quantization
1 other repos hold byte-for-byte identical weight files.