DuoNeural/Gemma-4-E2B-Heretic
Abliterated vs google/gemma-4-E2B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 29 of 35 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 29/585, Norms 0/433, other 0/240, Attention 20/687, Embeddings 0/6
2 derivatives checked
- mradermacher/Gemma-4-E2B-Heretic-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma-4-E2B-Heretic-GGUF: Same weights, converted (fewer layers)
1 other repos hold byte-for-byte identical weight files.