Hallucination Detection on MLQE-PE NE-EN (test)
97.64AUROCCometKiwi
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CometKiwiType=External Detector, Note=Trained on MLQE-PE data samples2022.12 | 97.64 | 4.03 | |
| LaBSEType=External Detector2022.12 | 92.34 | 20.03 | |
| Wass-to-DataType=Model-based Detector, Note=Mean and SD across five random seeds2022.12 | 90.18 | 48.52 | |
| Wass-ComboType=Model-based Detector, Note=Mean and SD across five random seeds2022.12 | 90.16 | 48.52 | |
| Wass-to-UnifType=Model-based Detector2022.12 | 81.49 | 64.23 | |
| Attn-ign-SRCType=Model-based Detector2022.12 | 28.66 | 100 | |
| Seq-LogprobType=Model-based Detector2022.12 | 26.38 | 99.94 |