kikichi/Qwen2.5-3B-Instruct-Uncensored
Abliterated vs Qwen/Qwen2.5-3B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/73, MLP 21/108, Attention 27/252
2 derivatives checked
- mradermacher/Qwen2.5-3B-Instruct-Uncensored-GGUF: Same weights, converted
- mradermacher/Qwen2.5-3B-Instruct-Uncensored-i1-GGUF: Quantization