huihui-ai/Huihui-Xing4.0-29B-A4B-abliterated

Abliterated vs XingChen-AGI/Xing4.0-29B-A4B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 40 of 41 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.

Tensors changed: Attention 40/325, MLP 748/2502, Embeddings 0/2, Norms 0/168, lm_head 0/1, other 0/2

2 derivatives checked

On the Hub · Downloads on Model Pulse