wangzhang/Qwen3.5-27B-abliterated
Abliterated vs Qwen/Qwen3.5-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 54 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, MLP 54/192, Attention 41/448, Norms 0/209
2 derivatives checked
- mradermacher/Qwen3.5-27B-abliterated-GGUF: Quantization
- mradermacher/Qwen3.5-27B-abliterated-i1-GGUF: Quantization