coder3101/gemma-4-E4B-it-qat-q4_0-unquantized-heretic
Abliterated vs google/gemma-4-E4B-it-qat-q4_0-unquantized. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 16 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 0/606, Norms 0/484, other 0/261, Attention 16/719, Embeddings 0/6
8 derivatives checked
- SC117/gemma-4-E4B-it-heretic-QAT-GGUF: Quantization (fewer layers)
- mradermacher/gemma-4-E4B-it-qat-q4_0-unquantized-heretic-i1-GGUF: Quantization (fewer layers)
- FadedRedStar/gemma-4-E4B-it-qat-q4_0-heretic-GGUF: Not comparable yet
- mradermacher/gemma-4-E4B-it-qat-q4_0-unquantized-heretic-GGUF: Same weights, converted (fewer layers)
- guga112/gemma-4-E4B-it-heretic-QAT-GGUF: Quantization (fewer layers)
- HTNZ555/gemma-4-E4B-it-heretic-QAT-GGUF: Quantization (fewer layers)
- yunlou2026/gemma-4-E4B-it-heretic-QAT-GGUF: Quantization (fewer layers)
- Disya/gemma-4-E4B-it-qat-q4_0-unquantized-heretic-gguf: Same weights, converted (fewer layers)