ressl/gemma-4-31B-it-uncensored
Abliterated vs google/gemma-4-31B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 60 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 60/261, Attention 59/338
4 derivatives checked
- ressl/gemma-4-31B-it-uncensored-GGUF: Quantization (fewer layers)
- bzannah/gemma-4-31B-it-uncensored-MLX-6bit: Quantization
- ressl/gemma-4-31B-it-uncensored-MLX-4bit: Quantization
- ressl/gemma-4-31B-it-uncensored-NVFP4: Quantization
1 other repos hold byte-for-byte identical weight files.