0xKitkat/Agnes-3.0-Flash-abliterated
Abliterated vs Agnes-AI/Agnes-3.0-Flash. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj, out_proj) changed, in 72 of 72 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 72/616, Norms 0/351, MLP 144/547, Embeddings 1/5, lm_head 0/1, other 0/1
1 derivatives checked
- 0xKitkat/Agnes-3.0-Flash-abliterated-GGUF: Not comparable yet