huihui-ai/Huihui-Qwen3.5-0.8B-abliterated
Abliterated vs Qwen/Qwen3.5-0.8B. 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 25/220, Norms 0/135, Embeddings 0/5, MLP 25/127, other 0/1
7 derivatives checked
- mradermacher/Huihui-Qwen3.5-0.8B-abliterated-GGUF: Same weights, converted
- mradermacher/Huihui-Qwen3.5-0.8B-abliterated-i1-GGUF: Quantization
- bobber/routangseng-qwen35-0.8b-abliterated-gguf: Full fine-tune
- YuYu1015/YuYu1015-Qwen3.5-0.8B-TW-CivicAligned-v1: Partial fine-tune (fewer layers)
- nbeerbower/Huihui-Qwen3.5-0.8B-abliterated-Athanorlite-ORPO: LoRA-style
- bobber/routangseng-qwen35-0.8b-abliterated: Partial fine-tune
- bobber/routangseng-0.8b-hottake: Full fine-tune (fewer layers)
1 other repos hold byte-for-byte identical weight files.