Fu01978/SmolLM2-360M-Instruct-Heretic
Abliterated vs HuggingFaceTB/SmolLM2-360M-Instruct. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 22 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/65, MLP 20/96, Attention 10/128
2 derivatives checked
- mradermacher/SmolLM2-360M-Instruct-Heretic-GGUF: Same weights, converted
- Fu01978/SmolMoE-4x360M-Instruct: MoE built from dense models