huihui-ai/Huihui-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 64 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/5, MLP 65/307, Attention 65/560, other 0/1, Norms 0/325
10 derivatives checked
- mradermacher/Huihui-Qwen3.5-27B-abliterated-GGUF: Quantization
- HeYujie/Qwen3.5-27B-abliterated-GGUF: Quantization
- mradermacher/Huihui-Qwen3.5-27B-abliterated-i1-GGUF: Quantization
- DevQuasar/huihui-ai.Huihui-Qwen3.5-27B-abliterated-GGUF: Same weights, converted
- j-a-a-a-y/Huihui-Qwen3.5-27B-abliterated-AWQ-W4A16: Quantization
- tstepspam/Huihui-Qwen3.5-27B-abliterated-Q8-MLX: Quantization
- tstepspam/Huihui-Qwen3.5-27B-abliterated-Q4-MLX: Quantization
- YuYu1015/Huihui-Qwen3.5-27B-abliterated-int4-AutoRound: Quantization
- nbeerbower/Huihui-Qwen3.5-27B-abliterated-Athanorlite-ORPO: Different architecture
- nbeerbower/Huihui-Qwen3.5-27B-abliterated-Athanorlite-ORPO-v2: LoRA-style
8 other repos hold byte-for-byte identical weight files.