List your company

Jailbreak Safety Evaluation on StrongREJECT (test)

100Overall ScoreSTAIR-DPO
98.668899.014499.3699.7056Sep 15, 2025

Evaluation results

MethodLinks
STAIR-DPOBase Model=Llama3.1-8B-Instruct, Evaluator=LlamaGuard, Prefilling=No2025.0910068.3794.2598.7290.34
BaseBase Model=Llama3.1-8B-Instruct, Evaluator=LlamaGuard, Prefilling=No2025.0999.6835.1464.8652.463.02
WJ-SFTBase Model=Llama3.1-8B-Instruct, Evaluator=LlamaGuard, Prefilling=No2025.0999.3644.7376.0498.479.63
ReSA-SFTBase Model=Llama3.1-8B-Instruct, Evaluator=LlamaGuard, Prefilling=No2025.0999.3689.7896.8199.3696.33
ReSA-SFTBase Model=Llama3.1-8B-Instruct, Evaluator=LlamaGuard, Prefilling=Yes2025.0998.7278.2795.8597.7692.65