Goekdeniz-Guelmez/Llama-3.2-1B-Instruct-gabliterated
Abliterated vs meta-llama/Llama-3.2-1B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 13 of 16 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/33, MLP 13/48, Attention 13/64
2 derivatives checked
- mradermacher/Llama-3.2-1B-Instruct-gabliterated-i1-GGUF: Quantization
- mradermacher/Llama-3.2-1B-Instruct-gabliterated-GGUF: Same weights, converted