dopaemon/Gemma4-E4B-8B-Heretic-Ultra
Abliterated vs google/gemma-4-E4B-it. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 24 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 0/606, Norms 0/502, other 0/261, Attention 24/755, Embeddings 0/6
3 derivatives checked
- mradermacher/Gemma4-E4B-8B-Heretic-Ultra-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma4-E4B-8B-Heretic-Ultra-GGUF: Same weights, converted (fewer layers)
- dopaemon/Gemma4-E4B-8B-Heretic-Ultra-MLX-4Bit: Quantization
2 other repos hold byte-for-byte identical weight files.