okohol/Qwen3.5-9B-CyberOps
Abliterated vs Qwen/Qwen3.5-9B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 21 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, MLP 15/208, Attention 21/332, Norms 0/215
1 derivatives checked
- mradermacher/Qwen3.5-9B-CyberOps-GGUF: Same weights, converted