Jailbreak Defense on TAP
2ASRWildGuard (Pre)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| WildGuard (Pre)2026.04 | 2 | 8 | |
| GuardReasoner (Pre)Latency (Sec.)=14.87, Memory Overhead (MB)=15617.332026.04 | 3 | 7 | |
| WildGuard (Post)2026.04 | 4 | 87 | |
| GradientCuffLatency (Sec.)=26.57, Memory Overhead (MB)=2063.762026.04 | 5 | 10 | |
| Token Highlighter2026.04 | 5 | 7 | |
| GuardReasoner (Post)Latency (Sec.)=16.71, Memory Overhead (MB)=15618.132026.04 | 5 | 87 | |
| SelfDefend (Intent)Latency (Sec.)=3.11, Memory Overhead (MB)=13457.402026.04 | 6 | 12 | |
| SelfGraderLatency (Sec.)=1.24, Memory Overhead (MB)=479.812026.04 | 6 | 20 | |
| GradSafe2026.04 | 9 | 53 | |
| SelfDefend (Direct)Latency (Sec.)=1.19, Memory Overhead (MB)=13452.932026.04 | 12 | 20 | |
| LLama-3-8B-Instruct (No Defense)Latency (Sec.)=1.472026.04 | 14 | — | |
| Perplexity FilterLatency (Sec.)=0.49, Memory Overhead (MB)=13571.942026.04 | 14 | 100 | |
| Prompt GuardLatency (Sec.)=22.67, Memory Overhead (MB)=1864.112026.04 | 14 | 93 | |
| Llama Guard (Pre)Latency (Sec.)=0.96, Memory Overhead (MB)=13910.612026.04 | 14 | 46 | |
| Llama Guard (Post)Latency (Sec.)=1.17, Memory Overhead (MB)=13910.612026.04 | 14 | 100 | |
| FedDetoxSetting=Ours2026.04 | 61 | — | |
| Qwen2.5-1.5B-InstructSetting=Original2026.04 | 67 | — | |
| FedDPOSetting=Poisoned2026.04 | 77 | — |