List your company

Adversarial Robustness on HarmBench

56.25DRBase
4.2517.7531.2544.75May 30, 2025

Evaluation results

MethodLinks
BaseModel=Mistral-7B2025.0556.2546.2543.7567.5
CAPOModel=Mistral-7B2025.0533.7527.54053.75
DPOModel=Mistral-7B2025.0532.53333.7558.75
APL (w/o Adv)Model=Mistral-7B2025.0531.2530.537.555
APL (RM)Model=Mistral-7B2025.0522.523.2522.548.75
APLModel=Mistral-7B2025.0511.251623.7541.25
BaseModel=Llama-3-8B2025.057.51417.516.25
CAPOModel=Llama-3-8B2025.057.5141523.19
APL (RM)Model=Llama-3-8B2025.057.51310.6715
DPOModel=Llama-3-8B2025.056.2511.751516.25
APL (w/o Adv)Model=Llama-3-8B2025.056.2512.7513.7513.75
APLModel=Llama-3-8B2025.056.25116.2513.75