Input prompt classification on Skill-Unseen Method (70B)
100AccuracyEvoSafety+ (Guard)-3B
Evaluation Results
| Method | Links | |
|---|---|---|
| EvoSafety+ (Guard)-3Bsize=3B, variant=plus2026.05 | 100 | |
| EvoSafety (Guard)-3Bsize=3B2026.05 | 99.47 | |
| Qwen3Guard-8Bsize=8B2026.05 | 86.8 | |
| WildGuard-7Bsize=7B2026.05 | 85.53 | |
| Llama-Guard-3-8Bsize=8B2026.05 | 65.13 |