dealignai/MiMo-V2.6-Flash-RL-UNCENSORED
Abliterated vs XiaomiMiMo/MiMo-V2.6-Flash-RL. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 14 of 48 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 0/2419, Attention 14/373, other 0/3, Norms 0/182, lm_head 0/1, Embeddings 0/22
2 derivatives checked
- AliceThirty/MiMo-V2.6-Flash-RL-UNCENSORED-gguf: Quantization
- cafonez/MiMo-V2.6-Flash-RL-UNCENSORED-Gorgon-GGUF: Quantization