Passage-level Retrieval on KILT (dev)
67.8FEV ScoreSEAL (LM+FM, intersective)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SEAL (LM+FM, intersective)multi-task=true, retrieval formulation=LM+FM, intersective2022.04 | 67.8 | 58.9 | 78.8 | 43.6 | 54.3 | 41.8 | 36 | 54.5 | |
| MT-DPR† (large)multi-task=true, source=Oğuz et al., 2021, additional synthetic data=true, model size=large2022.04 | 62.8 | 66.6 | 66.9 | 42.6 | 42.1 | 37.9 | 23.4 | 48.9 | |
| MT-DPR†multi-task=true, source=Oğuz et al., 2021, additional synthetic data=true2022.04 | 61.4 | 68.4 | 73.3 | 44.1 | 44.6 | 38.9 | 26.5 | 51 | |
| MT-DPRmulti-task=true, source=Maillard et al., 20212022.04 | 52.1 | 53.5 | 41.7 | 28.8 | 38.4 | 34.2 | 24.1 | 39 | |
| MT-DPRmulti-task=true, source=Oğuz et al., 20212022.04 | 52.1 | 61.4 | 54.1 | 40.1 | 41 | 34.2 | 24.6 | 43.9 | |
| DPRsource=Maillard et al., 20212022.04 | 43.9 | 58.5 | 78.8 | 28.1 | 43.5 | 23.8 | 20.7 | 42.5 | |
| BM252022.04 | 40.1 | 51.6 | 53 | 14.2 | 38.4 | 16.2 | 18.4 | 33.1 | |
| SEAL (LM+FM)multi-task=true, retrieval formulation=LM+FM2022.04 | 31.5 | 42 | 34 | 21.7 | 24.7 | 21.4 | 17.6 | 27.6 |