Binary Safety Detection on GUARDZOO (out-of-domain)
87.09Unsafe AccuracyROUTEGUARD-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ROUTEGUARD-8BImplementation Strategy=Router (Proposed), Evaluation Setting=Out-of-domain2026.05 | 87.09 | 66.33 | |
| ROUTEGUARD-4BImplementation Strategy=Router (Proposed), Evaluation Setting=Out-of-domain2026.05 | 86.76 | 66.05 | |
| LLAMAGUARD3-8BImplementation Strategy=Fine-tuned Baselines, Evaluation Setting=Out-of-domain2026.05 | 84.77 | 66.11 | |
| GRAPH ROUTERImplementation Strategy=Router Baselines, Evaluation Setting=Out-of-domain2026.05 | 83.9 | 54.09 | |
| SHIELDGEMMA-9BImplementation Strategy=Fine-tuned Baselines, Evaluation Setting=Out-of-domain2026.05 | 83.88 | 65.64 | |
| DC ROUTERImplementation Strategy=Router Baselines, Evaluation Setting=Out-of-domain2026.05 | 83.58 | 51.12 | |
| MLP ROUTERImplementation Strategy=Router Baselines, Evaluation Setting=Out-of-domain2026.05 | 83.36 | 50.99 | |
| QWEN3GUARD-4BImplementation Strategy=Fine-tuned Baselines, Evaluation Setting=Out-of-domain2026.05 | 83.13 | 46.43 | |
| QWEN3GUARD-8BImplementation Strategy=Fine-tuned Baselines, Evaluation Setting=Out-of-domain2026.05 | 81.45 | 36.94 | |
| KNN ROUTERImplementation Strategy=Router Baselines, Evaluation Setting=Out-of-domain2026.05 | 79.53 | 29.47 |