Paragraph-level Retrieval on KILT benchmark
62.8FEVER ScoreDPR-PAQ
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DPR-PAQBase model=BERTlarge2021.07 | 62.8 | 66.58 | 66.9 | 42.6 | 42.1 | 37.9 | 23.4 | 48.9 | |
| DPR-PAQBase model=BERTbase2021.07 | 61.4 | 68.4 | 73.28 | 44.1 | 44.6 | 38.9 | 26.5 | 50.6 | |
| Multi-task DPRBase model=BERTbase2021.07 | 52.1 | 61.4 | 54.1 | 40.1 | 41 | 34.2 | 24.6 | 43.9 | |
| BM25Base model=-2021.07 | 40.1 | 51.6 | 53 | 14.2 | 38.4 | 16.2 | 18.4 | 33.1 |