Open-domain dialogue on Wizard-of-Wikipedia KILT (test)
15.78F1 ScoreEMAT-FKSV
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EMAT-FKSVModel architecture type=Ours2022.10 | 15.78 | 14.73 | 141 | |
| EMAT-SKSVModel architecture type=Ours2022.10 | 15.35 | 14.68 | 150 | |
| BART + DPRModel architecture type=Retrieval-augmented models2022.10 | 15.19 | 13.23 | 0.7 | |
| T5-baseModel architecture type=Parametric models2022.10 | 13.53 | 12.4 | 160 | |
| RAGModel architecture type=Retrieval-augmented models2022.10 | 13.11 | 11.57 | 3.4 | |
| BART-largeModel architecture type=Parametric models2022.10 | 12.86 | 11.77 | 55 | |
| Trans MemNetModel architecture type=Parametric models2022.10 | 11.85 | 10.11 | — | |
| RePAQ w/ EMAT key encoderModel architecture type=Retrieval-only models2022.10 | 1.84 | 1.48 | — |