List your company

Adversarial Robustness on HarmBench (DirReq, GCG, A.DAN, A.Pmpt, PAIR, TAP)

17DirReqBase (8B Instruct)
-0.683.918.513.09May 8, 2026

Evaluation results

MethodLinks
Base (8B Instruct)Base Model=Llama 32026.05172712182526
CATBase Model=Llama 22026.059165143734
LPABase Model=Llama 32026.05500036
Base (7B chat-hf)Base Model=Llama 22026.053407171521
LPABase Model=Llama 22026.053346121120
CATBase Model=Llama 32026.0511032619
LPA-overfitBase Model=Llama 32026.05100033
LATBase Model=Llama 32026.05000026
LATBase Model=Llama 22026.05000001
LPA-overfitBase Model=Llama 22026.0501617615