huihui-ai/QwQ-32B-Preview-abliterated
Abliterated vs Qwen/QwQ-32B-Preview. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 63 of 64 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/129, MLP 63/192, Attention 63/448, lm_head 0/1
7 derivatives checked
- bartowski/QwQ-32B-Preview-abliterated-GGUF: Same weights, converted
- mradermacher/QwQ-32B-Preview-abliterated-i1-GGUF: Quantization
- mradermacher/QwQ-32B-Preview-abliterated-GGUF: Quantization
- tensorblock/QwQ-32B-Preview-abliterated-GGUF: Quantization
- huihui-ai/QwQ-32B-Coder-Fusion-9010: Full fine-tune
- zypcastles/QwQ-32B-Instruct-abliterated: Full fine-tune
- BenevolenceMessiah/QwQ-32B-Preview-abliterated-Self-instruct-2x-TIES-v1.0: Full fine-tune