DavidAU/gemma-3-4b-it-heretic-uncensored-abliterated-Extreme
Abliterated vs google/gemma-3-4b-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 34 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/316, MLP 27/210, Attention 24/352, other 0/1
5 derivatives checked
- mradermacher/gemma-3-4b-it-heretic-uncensored-abliterated-Extreme-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-3-4b-it-heretic-uncensored-abliterated-Extreme-i1-GGUF: Quantization (fewer layers)
- DavidAU/gemma-3-4b-it-vl-Heretic-4-Horsemen-Uncensored: LoRA-style
- DavidAU/Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-Thinking: LoRA-style
- DavidAU/Gemma-3-4B-VL-it-GLM-4.7-Flash-Heretic-Uncensored-Thinking: LoRA-style
1 other repos hold byte-for-byte identical weight files.