null-space/gemma-4-31b-it-abliterated
Abliterated vs google/gemma-4-31B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 40 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 40/261, Attention 40/338
2 derivatives checked
- divinetribe/gemma-4-31b-it-abliterated-4bit-mlx: Quantization
- altomek/Gemma-4-31B-it-Abliterated-XXL-GGUF: Quantization (fewer layers)
2 other repos hold byte-for-byte identical weight files.