Open-domain Question Answering on Natural Questions (NQ) few-shot (test)
45.5EMCodex + REPLUG LSR
Evaluation results
| Method | Links | |
|---|---|---|
| Codex + REPLUG LSRNumber of shots=16, Evaluation Protocol=In-context learning, Retrieval Corpus=Wikipedia (2018, December)2023.01 | 45.5 | |
| Codex + REPLUGNumber of shots=16, Evaluation Protocol=In-context learning, Retrieval Corpus=Wikipedia (2018, December)2023.01 | 44.7 | |
| Codex + ContrieverNumber of shots=16, Evaluation Protocol=In-context learning, Retrieval Corpus=MS-MARCO2023.01 | 44.2 | |
| AtlasNumber of shots=64, Evaluation Protocol=Finetuned2023.01 | 42.4 | |
| CodexNumber of shots=16, Evaluation Protocol=In-context learning2023.01 | 40.6 | |
| PaLMNumber of shots=64, Evaluation Protocol=In-context learning2023.01 | 39.6 | |
| ChinchillaNumber of shots=64, Evaluation Protocol=In-context learning2023.01 | 35.5 |