zenlm/zen-pro
Abliterated vs Qwen/Qwen3-8B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj) changed, in 36 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/145, MLP 36/108, Attention 36/144, lm_head 0/1
2 derivatives checked
- mradermacher/zen-pro-i1-GGUF: Quantization
- mradermacher/zen-pro-GGUF: Same weights, converted
10 other repos hold byte-for-byte identical weight files.