Pre-load security auditing on SkillGuardBench (Int.-stress)
98.51Overall Exact AccuracySKILLGUARD-ROBUST
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SKILLGUARD-ROBUST2026.04 | 98.51 | 100 | 98.33 | 100 | 97.87 | 100 | 100 | |
| SKILLGUARD-CALIBRATE2026.04 | 96.8 | 100 | 94.5 | 100 | 94.87 | 100 | 98.72 | |
| SKILLGUARD-VERIFY2026.04 | 93.35 | 100 | 89.33 | 100 | 88.31 | 100 | 97.44 | |
| BUNDLEJUDGE2026.04 | 85.13 | 96.57 | 72.83 | 67.94 | 78.91 | 76.32 | 95.11 | |
| Qwen2.5-14BPackage Mode=full-package2026.04 | 79.47 | 93.26 | 65.67 | 60.52 | 72.85 | 70.08 | 91.79 |