Safety Classification on Jailbreak prompts (test)
100Adv-suffix AccuracyLlama Guard 3
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Llama Guard 3Base Model=LLAMA2-CHAT-7B2025.07 | 100 | 0 | 76 | |
| Latent GuardBase Model=LLAMA2-CHAT-7B2025.07 | 100 | 41.6 | 100 | |
| Latent GuardBase Model=QWEN2-INSTRUCT-7B2025.07 | 100 | 75 | 53.5 | |
| Llama Guard 3Base Model=LLAMA3-INSTRUCT-8B2025.07 | 99.2 | 6.8 | 50 | |
| Llama Guard 3Base Model=QWEN2-INSTRUCT-7B2025.07 | 97.8 | 17.8 | 91.4 | |
| Latent GuardBase Model=LLAMA3-INSTRUCT-8B2025.07 | 91 | 65 | 100 |