saidutta69/MiniCPM5-2B-Claude-Fable5-heretic
Abliterated vs SauravMahalik/MiniCPM5-2B-Claude-Fable5. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 36 of 42 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/85, MLP 22/126, Attention 36/168, lm_head 0/1
2 derivatives checked
- mradermacher/MiniCPM5-2B-Claude-Fable5-heretic-i1-GGUF: Quantization
- mradermacher/MiniCPM5-2B-Claude-Fable5-heretic-GGUF: Same weights, converted