Jiunsong/SuperGLM-5.2-abliterated-NVFP4
Abliterated vs nvidia/GLM-5.2-NVFP4. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 62 of 79 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 62/2172, Embeddings 0/1, Norms 0/364, Attention 62/461, lm_head 0/1, other 0/1