meangrinch/Qwen3.5-35B-A3B-heretic-v3
Abliterated vs unsloth/Qwen3.5-35B-A3B. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 21 of 40 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: lm_head 0/1, Embeddings 0/4, Attention 16/388, MLP 21/312, Norms 0/241
1 derivatives checked
- meangrinch/Qwen3.5-35B-A3B-heretic-v3-GGUF: Quantization