adriaflores/Llama-3-8B-abliterated
Abliterated vs meta-llama/Meta-Llama-3-8B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 32 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/65, MLP 32/96, Attention 32/128, lm_head 0/1
2 derivatives checked
- mradermacher/Llama-3-8B-abliterated-i1-GGUF: Quantization
- mradermacher/Llama-3-8B-abliterated-GGUF: Same weights, converted