Safety Classification on Bloom safety filter benchmark 400 examples (val)
100AccuracyBloom Safety Filter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Bloom Safety FilterCategory=3: Mental Health, Evaluation protocol=Strict, Temperature=02025.10 | 100 | 100 | 100 | 100 | |
| Bloom Safety FilterCategory=5: Inacc. Advice, Evaluation protocol=Strict, Temperature=02025.10 | 99 | 100 | 98 | 99 | |
| Bloom Safety FilterCategory=1: Bodily Harm, Evaluation protocol=Strict, Temperature=02025.10 | 98 | 99 | 96 | 98 | |
| Bloom Safety FilterCategory=4: Neg. Mindsets, Evaluation protocol=Strict, Temperature=02025.10 | 97 | 98 | 97 | 97 | |
| Bloom Safety FilterCategory=2: Body Image, Evaluation protocol=Strict, Temperature=02025.10 | 95 | 100 | 89 | 94 | |
| Bloom Safety FilterCategory=Overall, Evaluation protocol=Relaxed, Temperature=02025.10 | 93 | 89 | 98 | 93 | |
| Bloom Safety FilterCategory=Overall, Evaluation protocol=Strict, Temperature=02025.10 | 92 | 89 | 96 | 92 |