grimjim/Llama-3-Instruct-abliteration-OVA-8B
Abliterated vs failspy/Meta-Llama-3-8B-Instruct-abliterated-v3. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 27 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 27/96, Attention 27/128, lm_head 0/1
1 derivatives checked
- mradermacher/Llama-3-Instruct-abliteration-OVA-8B-GGUF: Same weights, converted