tvall43/Qwen3.5-2B-heretic
Abliterated vs Qwen/Qwen3.5-2B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 16 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Attention 14/264, Norms 0/177, Embeddings 0/4, MLP 4/172
1 derivatives checked
- tvall43/Qwen3.5-2B-heretic-gguf: Same weights, converted