DavidAU/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored
Abliterated vs meta-llama/Llama-3.2-3B-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 21 of 28 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/57, MLP 5/84, Attention 21/112
8 derivatives checked
- tinyopsec/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored-GGUF: Same weights, converted
- mradermacher/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored-GGUF: Same weights, converted
- mradermacher/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored-i1-GGUF: Quantization
- Fu01978/Llama-3.2-3B-MoE-4Expert: MoE built from dense models
- Guilherme34/Firefly: Full fine-tune
- UmbrellaInc/Progenitor-Virus-3.2-3B: Full fine-tune
- Kanjdoes/L3.2-3B-Herthea: Full fine-tune
- blascotobasco/L3.2-Ascendant-8E-A6B: MoE built from dense models