huihui-ai/Huihui-MiniCPM-V-4.6-abliterated
Abliterated vs openbmb/MiniCPM-V-4.6. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 24 of 27 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/195, Attention 24/392, MLP 24/184, other 0/4
2 derivatives checked
- mradermacher/Huihui-MiniCPM-V-4.6-abliterated-GGUF: Same weights, converted
- mradermacher/Huihui-MiniCPM-V-4.6-abliterated-i1-GGUF: Quantization