prithivMLmods/gemma-4-E4B-it-qat-heretic_decensored
Abliterated vs google/gemma-4-E4B-it-qat-q4_0-unquantized. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 30 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 30/606, Norms 0/484, other 0/261, Attention 24/719, Embeddings 0/6
3 derivatives checked
- mradermacher/gemma-4-E4B-it-qat-heretic_decensored-i1-GGUF: Quantization (fewer layers)
- prithivMLmods/gemma-4-E4B-it-qat-heretic_decensored-GGUF: Same weights, converted (fewer layers)
- mradermacher/gemma-4-E4B-it-qat-heretic_decensored-GGUF: Same weights, converted (fewer layers)