coder3101/QwenPaw-Flash-9B-heretic
Abliterated vs agentscope-ai/QwenPaw-Flash-9B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 3 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Norms 0/215, Attention 3/332, MLP 3/208
2 derivatives checked
- FadedRedStar/QwenPaw-Flash-9B-heretic-GGUF: Same weights, converted (fewer layers)
- FadedRedStar/QwenPaw-Flash-9B-heretic-imatrix-GGUF: Same weights, converted (fewer layers)