mondk/MiniCPM5-2B-Abliterated-Uncensored-Safetensors
Abliterated vs openbmb/MiniCPM5-2B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 42 of 42 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 0/85, MLP 42/126, Attention 42/168
5 derivatives checked
- mondk/MiniCPM5-2B-Abliterated-Uncensored-GGUF: Same weights, converted
- 0xSojalSec/MiniCPM5-2B-Abliterated-Uncensored-GGUF: Same weights, converted
- mradermacher/MiniCPM5-2B-Abliterated-Uncensored-Safetensors-i1-GGUF: Quantization
- mradermacher/MiniCPM5-2B-Abliterated-Uncensored-Safetensors-GGUF: Same weights, converted
- mondk/MiniCPM5-2B-Abliterated-Uncensored-MLX-4Bit: Quantization
1 other repos hold byte-for-byte identical weight files.