ToxiGen
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Toxicity Detection | ToxiGen | Score84.23 | 95 | |
| Safety Evaluation | ToxiGen | Safety100 | 77 | |
| Hate speech classification | ToxiGen (test) | AUC99 | 24 | |
| Toxicity Generation | ToxiGen | ToxiGen Score1,633 | 24 | |
| Toxicity Classification | Toxigen | Accuracy60.41 | 22 | |
| Harmlessness | Toxigen | Toxigen (%)100 | 17 | |
| Detoxification | ToxiGen (test) | MTV97.4 | 16 | |
| Influence Estimation | ToxiGen (test) | Spearman Correlation0.44 | 14 | |
| Machine Unlearning | ToxiGen (test) | Accuracy ($D_f$)86.9 | 13 | |
| Machine Unlearning | ToxiGen (train) | Accuracy ($D_f$)85.06 | 13 | |
| Text Classification | ToxiGen (test) | Accuracy85 | 12 | |
| Bias Detection | Toxigen (test) | Accuracy90.3 | 12 | |
| Safety Evaluation | ToxiGen Pretrained Evaluation | Toxicity Rate14.53 | 12 | |
| Toxicity Detection | TOXIGEN (val) | AUC96 | 8 | |
| Safety Over-triggering | ToxiGen | Over-trigger Rate: Jewish0.02 | 7 | |
| Implicit Hate Speech Detection | Toxigen | Macro-F193.41 | 5 | |
| Formal verification of safety classifiers | Toxigen | tau* Score0.9 | 3 | |
| Misuse Detection | ToxiGen Homophobia (external) | TPR98 | 1 | |
| Misuse Detection | ToxiGen Ethnoracial (external) | TPR91 | 1 | |
| Detoxification Dataset Quality Evaluation | ToxiGen 500 neutral-toxic pairs | Overall O.2.475 | 1 |
Showing 20 of 20 rows