Defending against gradient-based attacks on Llama3 GCG Attack (test)
9.61ASROurs-Fakecom-t
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours-Fakecom-tDefense=Fake Completion Defense with Template2024.11 | 9.61 | |
| Ours-IgnoreDefense=Ignore Defense2024.11 | 12.01 | |
| Ours-FakecomDefense=Fake Completion Defense2024.11 | 13.94 | |
| SandwichDefense=Sandwich defense baseline2024.11 | 19.23 | |
| Ours-EscapeDefense=Escape Defense2024.11 | 19.23 | |
| SpotlightDefense=Spotlight defense baseline2024.11 | 19.71 | |
| ReminderDefense=Reminder defense baseline2024.11 | 24.51 | |
| InstructionalDefense=Instructional defense baseline2024.11 | 28.84 | |
| IsolationDefense=Isolation defense baseline2024.11 | 40.38 | |
| NoneDefense=No defense baseline2024.11 | 87.01 |