mergekit-community/Perky-Pat-V2
Abliterated vs failspy/Llama-3-8B-Instruct-MopeyMule. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 25 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 25/96, Attention 25/128, lm_head 0/1
1 derivatives checked
- mradermacher/Perky-Pat-V2-GGUF: Same weights, converted