Stealthiness Evaluation on LLaMA Guard 7B
3.05Mean Perplexity (PPL)Ascii
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ascii2024.10 | 3.05 | 0.27 | |
| ArtPrompt2024.10 | 3.36 | 1 | |
| Base642024.10 | 10.14 | 2.85 | |
| Morse Cipher2024.10 | 11.18 | 1.89 | |
| ReNeLLM2024.10 | 13.33 | 4.16 | |
| Unicode2024.10 | 29.37 | 15.38 | |
| UTF-82024.10 | 29.37 | 15.38 | |
| Origin2024.10 | 33.44 | 21.14 | |
| Caesar Cipher2024.10 | 202.08 | 139.35 | |
| FlipAttack2024.10 | 563.32 | 234.23 |