ArliAI/R1-8B-ArliAI-RpR-v2
Abliterated vs deepseek-ai/DeepSeek-R1-Distill-Llama-8B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 31 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Attention 31/128, MLP 31/96, Norms 0/65, lm_head 0/1
2 derivatives checked
- mradermacher/R1-8B-ArliAI-RpR-v2-i1-GGUF: Quantization
- mradermacher/R1-8B-ArliAI-RpR-v2-GGUF: Same weights, converted