llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic
Abliterated vs Nimbz/Gemma-4-Gembrain-31B. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 42 of 60 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/523, other 0/62, MLP 0/261, Attention 42/338
3 derivatives checked
- llmfan46/Gemma-4-Gembrain-31B-it-uncensored-heretic-GGUF: Same weights, converted (fewer layers)
- mradermacher/Gemma-4-Gembrain-31B-it-uncensored-heretic-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma-4-Gembrain-31B-it-uncensored-heretic-GGUF: Quantization (fewer layers)