Safety Classification on DiaSafety (test)
66.98AUROCGAUGE-mean
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GAUGE-meanMethod Category=GAUGE (Ours), Aggregation Strategy=mean2025.12 | 66.98 | 64.51 | 64.24 | |
| GAUGE-percentileMethod Category=GAUGE (Ours), Aggregation Strategy=percentile2025.12 | 65.18 | 59.69 | 63.76 | |
| GAUGE-minMethod Category=GAUGE (Ours), Aggregation Strategy=min2025.12 | 64.09 | 59.33 | 63.74 | |
| GAUGE-topkMethod Category=GAUGE (Ours), Aggregation Strategy=top-k2025.12 | 62.66 | 56.36 | 64.03 | |
| Llama-Guard-3-8BMethod Category=External Classifiers2025.12 | 58.84 | 53.15 | 36.28 | |
| HateBERTMethod Category=External Classifiers2025.12 | 50.76 | 45.71 | 62.82 | |
| Safety Prompt (Refusal)Method Category=Prompt-based Baseline2025.12 | 50.43 | 46.02 | 20.62 | |
| ToxicBERTMethod Category=External Classifiers2025.12 | 33.66 | 35.44 | 26.12 |