Lambent/Mira-v1.17-27B-Custom-Heretic
Abliterated vs Lambent/Mira-v1.17-Karcher-27B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 37 of 62 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, Norms 0/484, MLP 16/294, Attention 37/464, other 0/1
2 derivatives checked
- mradermacher/Mira-v1.17-27B-Custom-Heretic-i1-GGUF: Quantization (fewer layers)
- mradermacher/Mira-v1.17-27B-Custom-Heretic-GGUF: Quantization (fewer layers)