huihui-ai/Huihui-gemma-4-E4B-it-abliterated
Abliterated vs google/gemma-4-E4B-it. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 35 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 35/606, Norms 0/502, other 0/261, Attention 35/755, Embeddings 0/6
5 derivatives checked
- mradermacher/Huihui-gemma-4-E4B-it-abliterated-GGUF: Same weights, converted (fewer layers)
- mradermacher/Huihui-gemma-4-E4B-it-abliterated-i1-GGUF: Quantization (fewer layers)
- EZCon/Huihui-gemma-4-E4B-it-abliterated-4bit-mlx: Quantization
- lagive/Huihui-gemma-4-E4B-it-abliterated-GGUF: Same weights, converted (fewer layers)
- EZCon/Huihui-gemma-4-E4B-it-abliterated-4bit-g32-mxfp4-mixed_4_8-mlx: Quantization
4 other repos hold byte-for-byte identical weight files.