Blackfrost-AI/KOLIBRI-1-DERISKED-BF16
Abliterated vs Aleph-Alpha/Kolibri-1-BF16. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 48 of 50 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 0/301, MLP 803/2494, Attention 48/200, other 0/3