MuXodious/ZwZ-8B-PaperWitch-heresy
Abliterated vs inclusionAI/ZwZ-8B. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, MLP 0/232, Norms 0/261, Attention 36/252, Embeddings 0/4
2 derivatives checked
- mradermacher/ZwZ-8B-PaperWitch-heresy-i1-GGUF: Quantization (fewer layers)
- mradermacher/ZwZ-8B-PaperWitch-heresy-GGUF: Same weights, converted (fewer layers)