List your company

Safety Evaluation on HarmBench translated (test)

11Success Rate (EN)Vanilla Model
-0.442.535.58.47Apr 13, 2026

Evaluation results

MethodLinks
Vanilla ModelBase Model=Llama-3.1-8B-Instruct2026.0411164827586528.4
Vanilla ModelBase Model=Qwen-2.5-7B-Instruct2026.04981917865225.1
CPOBase Model=Qwen-2.5-7B-Instruct2026.0440139793817.5
CPOBase Model=Llama-3.1-8B-Instruct2026.043273443110.6
MPOBase Model=Qwen-2.5-7B-Instruct2026.0432106723214.7
DPOBase Model=Llama-3.1-8B-Instruct2026.0427197292410.9
MPOBase Model=Llama-3.1-8B-Instruct2026.041110231197.6
LASABase Model=Llama-3.1-8B-Instruct2026.04100016173.9
SFTBase Model=Qwen-2.5-7B-Instruct2026.041042671610.3
ORPOBase Model=Qwen-2.5-7B-Instruct2026.04100056147.5
LASABase Model=Qwen-2.5-7B-Instruct2026.04100425165.6
SFTBase Model=Llama-3.1-8B-Instruct2026.04026445299.7
KTOBase Model=Llama-3.1-8B-Instruct2026.04013225155.4
ORPOBase Model=Llama-3.1-8B-Instruct2026.04012.021.0123154.3
DPOBase Model=Qwen-2.5-7B-Instruct2026.040187703314.5
KTOBase Model=Qwen-2.5-7B-Instruct2026.040075732813.5