piotreknow02/GPT-OSS-Cybersecurity-20B-Merged-heretic-ara
Abliterated vs sainikhiljuluri2015/GPT-OSS-Cybersecurity-20B-Merged. Looks like abliteration (refusal removal): only the output projections (o_proj) changed, in 12 of 24 layers. Abliteration edits exactly these matrices to remove a "refusal direction"; everything else is untouched.
Tensors changed: Embeddings 0/1, Norms 0/49, MLP 0/144, Attention 12/216, lm_head 0/1
1 derivatives checked
- piotreknow02/GPT-OSS-Cybersecurity-Merged-20B-heretic-ara-GGUF: Same weights, converted