huihui-ai/Llama-3.2-3B-Instruct-abliterated
Abliterated vs meta-llama/Llama-3.2-3B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, embed_tokens, o_proj) changed, in 28 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 1/1, Norms 0/57, MLP 28/84, Attention 28/112
9 derivatives checked
- mradermacher/Llama-3.2-3B-Instruct-abliterated-GGUF: Same weights, converted
- mradermacher/Llama-3.2-3B-Instruct-abliterated-i1-GGUF: Quantization
- MaziyarPanahi/Llama-3.2-3B-Instruct-abliterated-GGUF: Same weights, converted
- Hasaranga85/Llama-3.2-3B-Instruct-abliterated-Q4_K_M-GGUF: Quantization
- Rusgus/Llama-3.2-3B-Instruct-abliterated-GGUF: Same weights, converted
- maicog/Llama-3.2-3B-Instruct-abliterated-Q4_K_M-GGUF: Quantization
- mlx-community/Llama-3.2-3B-Instruct-abliterated-6bit: Quantization
- trollek/Llama-3.2-3B-Danoia: LoRA-style
- N-Bot-Int/MrgrtV2-3B-merged: LoRA-style
6 other repos hold byte-for-byte identical weight files.