Safety Evaluation on 8 jailbreak attacks (Aggregated)
3.13Average ASRREPBEND
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| REPBENDModel=Llama3 8B Instruct2025.04 | 3.13 | 83.14 | |
| REPBENDModel=Mistral 7B Instruct v0.22025.04 | 3.24 | 81.23 | |
| NPOModel=Mistral 7B Instruct v0.22025.04 | 5.83 | 74.52 | |
| CBModel=Llama3 8B Instruct, safety-tuned=true2025.04 | 7.96 | 75.87 | |
| CBModel=Llama3 8B Instruct2025.04 | 8.56 | 81.61 | |
| CBModel=Mistral 7B Instruct v0.22025.04 | 9.19 | 77.34 | |
| NPOModel=Llama3 8B Instruct2025.04 | 10.62 | 71.65 | |
| RMUModel=Llama3 8B Instruct2025.04 | 10.78 | 72.9 | |
| CBModel=Mistral 7B Instruct v0.2, safety-tuned=true2025.04 | 11.7 | 73.62 | |
| RMUModel=Mistral 7B Instruct v0.22025.04 | 16.19 | 71.85 | |
| R2D2Model=Mistral 7B Instruct v0.2, safety-tuned=true2025.04 | 23.09 | 72.67 | |
| TAModel=Mistral 7B Instruct v0.22025.04 | 23.14 | 72.96 | |
| TAModel=Llama3 8B Instruct2025.04 | 29.69 | 70.71 | |
| Original Weight ModelModel=Llama3 8B Instruct2025.04 | 34 | 73.9 | |
| Original Weight ModelModel=Mistral 7B Instruct v0.22025.04 | 60.64 | 63.81 |