ArliAI/QwQ-32B-ArliAI-RpR-v2
Abliterated vs Qwen/QwQ-32B. 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
3 derivatives checked
- mradermacher/QwQ-32B-ArliAI-RpR-v2-i1-GGUF: Quantization
- mradermacher/QwQ-32B-ArliAI-RpR-v2-GGUF: Quantization
- Yobenboben/QwQ-32B-Brh: Abliterated