Safety Classification on Chinese Prompt Datasets
86.31PolyGPolyGuard-7B
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PolyGuard-7BScale=4∼7B2026.06 | 86.31 | 83.54 | 68.13 | 27.6 | 63.84 | 73.83 | 67.21 | |
| CHILLGuard-4BScale=4∼7B2026.06 | 85.05 | 82.36 | 77.17 | 59.38 | 75.56 | 89.43 | 78.16 | |
| CHILLGuard-8BScale=8B+2026.06 | 84.82 | 80.34 | 77.44 | 56.49 | 73.44 | 89.77 | 77.05 | |
| Qwen3Guard-4B-LooseScale=4∼7B2026.06 | 83.86 | 78.48 | 63.01 | 17.32 | 61.57 | 43.84 | 58.01 | |
| CHILLGuard-1.7BScale=0∼3B2026.06 | 82.34 | 78.85 | 75.95 | 53.68 | 77.83 | 82.72 | 75.23 | |
| Qwen3Guard-4B-StrictScale=4∼7B2026.06 | 80.97 | 74.36 | 46.96 | 28.28 | 43.38 | 76.32 | 58.38 | |
| Qwen3Guard-0.6B-StrictScale=0∼3B2026.06 | 80.58 | 76.34 | 60.3 | 27.88 | 50.06 | 73.97 | 61.52 | |
| Qwen3Guard-0.6B-LooseScale=0∼3B2026.06 | 80.5 | 76.92 | 54.58 | 16.99 | 50.84 | 46.65 | 54.41 | |
| Qwen3Guard-8B-StrictScale=8B+2026.06 | 80.27 | 73.13 | 59.82 | 27.15 | 58.54 | 77.44 | 62.72 | |
| WildGuard-7BScale=4∼7B2026.06 | 80.12 | 75.12 | 51.29 | 16.2 | 54.31 | 50.72 | 54.63 | |
| LlamaGuard3-8BScale=8B+2026.06 | 78.46 | 77.11 | 58.97 | 16.2 | 60.32 | 44.53 | 55.93 | |
| Qwen3Guard-8B-LooseScale=8B+2026.06 | 75.62 | 70.02 | 41.25 | 17.44 | 40.78 | 52.36 | 49.58 | |
| ShieldGemma-9BScale=8B+2026.06 | 67.29 | 64.85 | 42.41 | 12.6 | 38.23 | 28.39 | 42.3 | |
| LlamaGuard4-12BScale=8B+2026.06 | 63.1 | 61.61 | 43.49 | 16.89 | 33.98 | 50.69 | 44.96 | |
| NemoGuard-8BScale=8B+2026.06 | 45.54 | 40.48 | 44.29 | 23.92 | 43.06 | 58.2 | 42.58 | |
| ShieldGemma-27BScale=8B+2026.06 | 45.45 | 41.64 | 42.38 | 12.49 | 39.65 | 28.13 | 34.96 |