DuoNeural/Nanbeige-4.1-3B-Abliterated
Abliterated vs Nanbeige/Nanbeige4.1-3B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, MLP 27/96, Attention 18/128, Norms 0/65, Embeddings 0/1
2 derivatives checked
- mradermacher/Nanbeige-4.1-3B-Abliterated-GGUF: Same weights, converted
- mradermacher/Nanbeige-4.1-3B-Abliterated-i1-GGUF: Quantization