coder3101/gemma-4-31B-it-qat-q4_0-unquantized-heretic
Abliterated vs google/gemma-4-31B-it-qat-q4_0-unquantized. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 24 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 24/338
5 derivatives checked
- mradermacher/gemma-4-31B-it-qat-q4_0-unquantized-heretic-i1-GGUF: Quantization (fewer layers)
- SC117/gemma-4-31B-it-heretic-QAT-For-Edge-16G-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-4-31B-it-qat-q4_0-unquantized-heretic-GGUF: Quantization (fewer layers)
- Stabhappy/gemma-4-31B-it-qat-q4_0-unquantized-heretic-GGUF: Quantization (fewer layers)
- pegasus912/gemma-4-31b-it-qat-heretic-ud-q4-k-xl: Quantization (fewer layers)