ressl/MiniMax-M3-uncensored
Abliterated vs MiniMaxAI/MiniMax-M3. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, w2) changed, in 60 of 60 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/2, Norms 0/485, MLP 557/1784, Attention 60/610, other 0/118
4 derivatives checked
- spiritfather/MiniMax-M3-uncensored-GGUF: Quantization
- frz1/MiniMax-M3-uncensored-GGUF: Quantization
- ressl/MiniMax-M3-uncensored-NVFP4: Quantization
- m9e/MiniMax-M3-uncensored-RP-LC-Q4HQ-GGUF: Quantization
2 other repos hold byte-for-byte identical weight files.