keithnull/Qwen3.6-35B-A3B-REAM-192-heretic
Abliterated vs keithnull/Qwen3.6-35B-A3B-REAM-192. Looks like abliteration (refusal removal): only the output projections (down_proj, o_proj, out_proj) changed, in 35 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, Norms 0/241, Attention 28/388, MLP 70/392
2 derivatives checked
- keithnull/Qwen3.6-35B-A3B-REAM-192-heretic-APEX-GGUF: Quantization
- keithnull/Qwen3.6-35B-A3B-REAM-192-heretic-GGUF: Same weights, converted