Jailbreak Attack on AdvBench (JSR, H-HS, A-HS, FPT)
0.2JSR (Jailbreak Success Rate)Direct
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DirectModel=LLaDA-V2026.05 | 0.2 | — | — | — | |
| FPOModel=LaViDA, NA=12026.05 | 1.1 | 1.25 | 1.55 | 5 | |
| DirectModel=LaViDA2026.05 | 1.2 | — | — | — | |
| FPOModel=LaViDA, NA=22026.05 | 1.4 | 1.52 | 1.81 | 6 | |
| FPOModel=LaViDA, NA=32026.05 | 1.6 | 1.82 | 2.04 | 7 | |
| FPOModel=LLaDA-V, NA=12026.05 | 8.5 | 1.81 | 1.02 | 7 | |
| FPOModel=LLaDA-V, NA=22026.05 | 9.8 | 2.01 | 1.14 | 8 | |
| FPOModel=LLaDA-V, NA=32026.05 | 11.1 | 2.12 | 1.24 | 8 | |
| GPO-VModel=LLaDA-V, NA=32026.05 | 81.1 | 7.25 | 6.54 | 15 | |
| GPO-VModel=LaViDA, NA=12026.05 | 82.3 | 7.12 | 7.45 | 12 | |
| GPO-VModel=LaViDA, NA=22026.05 | 85.6 | 7.45 | 7.88 | 18 | |
| GPO-VModel=LLaDA-V, NA=12026.05 | 86.4 | 7.12 | 6.98 | 12 | |
| GPO-VModel=LaViDA, NA=32026.05 | 88.9 | 7.71 | 8.12 | 12 | |
| GPO-VModel=LLaDA-V, NA=22026.05 | 93.2 | 7.85 | 7.21 | 10 |