llmfan46/Qwen3.5-27B-Writer-V2-uncensored-heretic
Abliterated vs ConicCat/Qwen3.5-27B-Writer-V2. Looks like abliteration (refusal removal): only the output projections (o_proj, out_proj) changed, in 20 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Norms 0/319, Attention 20/556, MLP 0/304
4 derivatives checked
- llmfan46/Qwen3.5-27B-Writer-V2-uncensored-heretic-GGUF: Same weights, converted
- mradermacher/Qwen3.5-27B-Writer-V2-uncensored-heretic-GGUF: Quantization
- mradermacher/Qwen3.5-27B-Writer-V2-uncensored-heretic-i1-GGUF: Quantization
- TheCluster/Qwen3.5-27B-Writer-V2-Uncensored-Heretic-MLX-mixed-9.4bit: Quantization
1 other repos hold byte-for-byte identical weight files.