FenrirLupus/Qwen-3.5-4B-A90-R10-Heretic
Abliterated vs Qwen/Qwen3.5-4B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 23 of 32 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/4, MLP 23/196, Attention 21/320, Norms 0/203
1 derivatives checked
- mradermacher/Qwen-3.5-4B-A90-R10-Heretic-GGUF: Same weights, converted