K0D3IN/SmolLM3-3B-Instruct-heretic
Abliterated vs HuggingFaceTB/SmolLM3-3B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 25 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/73, MLP 25/108, Attention 24/144
3 derivatives checked
- xtools-at/SmolLM3-3B-Instruct-heretic-GGUF: Quantization
- xtools-at/SmolMed3-3B-GGUF: Full fine-tune
- xtools-at/SmolPhish3-3B-GGUF: Full fine-tune