huihui-ai/Huihui-Qwen3.5-2B-abliterated
Abliterated vs Qwen/Qwen3.5-2B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 24 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 25/268, Norms 0/183, Embeddings 0/5, MLP 25/175, other 0/1
6 derivatives checked
- mradermacher/Huihui-Qwen3.5-2B-abliterated-GGUF: Same weights, converted
- mradermacher/Huihui-Qwen3.5-2B-abliterated-i1-GGUF: Quantization
- interimlabs/InterimLabs-Huihui-Qwen3.5-2B-abliterated-GGUF-Q4_K_M: Quantization
- noTrabajo/Huihui-Qwen3.5-2B-abliterated-Q4_K_M-GGUF: Quantization
- interimlabs/InterimLabs-Huihui-Qwen3.5-2B-abliterated-MLX-4bit: Quantization
- nbeerbower/Huihui-Qwen3.5-2B-abliterated-Athanorlite-ORPO: LoRA-style
1 other repos hold byte-for-byte identical weight files.