MobiusDevelopment/gemma4-E2B-it-qat-Q4-unsloth-heretic
Abliterated vs unsloth/gemma-4-E2B-it-qat-q4_0-unquantized. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 33 of 35 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 33/585, Norms 0/433, other 0/240, Attention 32/687, Embeddings 0/6