3MPER0RR/Ministral-3-3B-Instruct-2512-BF16-3MPER0RR-abliterated
Abliterated vs nchapman/Ministral-3-3B-Instruct-2512-heretic-BF16. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 10 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 8/150, Attention 7/200, other 0/4
2 derivatives checked
- mradermacher/Ministral-3-3B-Instruct-2512-BF16-3MPER0RR-abliterated-i1-GGUF: Quantization (fewer layers)
- mradermacher/Ministral-3-3B-Instruct-2512-BF16-3MPER0RR-abliterated-GGUF: Same weights, converted (fewer layers)