coder3101/Ministral-3-3B-Reasoning-2512-heretic
Abliterated vs mistralai/Ministral-3-3B-Reasoning-2512. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 18 of 26 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/103, MLP 18/150, Attention 3/200, other 0/4
1 derivatives checked
- coder3101/Ministral-3-3B-Reasoning-2512-heretic-GGUF: Same weights, converted (fewer layers)