List your company

Safety Evaluation on 8 jailbreak attacks (Aggregated)

3.13Average ASRREPBEND
0.829616.357331.88547.4127Apr 2, 2025

Evaluation results

MethodLinks
REPBENDModel=Llama3 8B Instruct2025.043.1383.14
REPBENDModel=Mistral 7B Instruct v0.22025.043.2481.23
NPOModel=Mistral 7B Instruct v0.22025.045.8374.52
CBModel=Llama3 8B Instruct, safety-tuned=true2025.047.9675.87
CBModel=Llama3 8B Instruct2025.048.5681.61
CBModel=Mistral 7B Instruct v0.22025.049.1977.34
NPOModel=Llama3 8B Instruct2025.0410.6271.65
RMUModel=Llama3 8B Instruct2025.0410.7872.9
CBModel=Mistral 7B Instruct v0.2, safety-tuned=true2025.0411.773.62
RMUModel=Mistral 7B Instruct v0.22025.0416.1971.85
R2D2Model=Mistral 7B Instruct v0.2, safety-tuned=true2025.0423.0972.67
TAModel=Mistral 7B Instruct v0.22025.0423.1472.96
TAModel=Llama3 8B Instruct2025.0429.6970.71
Original Weight ModelModel=Llama3 8B Instruct2025.043473.9
Original Weight ModelModel=Mistral 7B Instruct v0.22025.0460.6463.81