saidutta69/RaceBench-MiniCPM5-heretic
Abliterated vs saidutta69/RaceBench-MiniCPM5. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 17 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/49, MLP 7/72, Attention 17/96
2 derivatives checked
- mradermacher/RaceBench-MiniCPM5-heretic-i1-GGUF: Quantization
- mradermacher/RaceBench-MiniCPM5-heretic-GGUF: Same weights, converted