llmfan46/gpt-oss-120b-ultra-heretic
Abliterated vs openai/gpt-oss-120b. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 17 of 36 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: MLP 0/288, Norms 0/73, Attention 17/324, lm_head 0/1, Embeddings 0/1
1 derivatives checked
- llmfan46/gpt-oss-120b-ultra-heretic-GGUF: Same weights, converted