Open-domain Question Answering on TRIVIAQA (dev)
514.6Word ScoreTop 5 documents
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Top 5 documentsRetrieval Strategy=Retrieval without Noise Filtering, Documents=Top 52024.06 | 514.6 | — | 40.7 | 39.6 | 21.1 | 70.9 | -4.39 | |
| LONGLLMLINGUARetrieval Strategy=Retrieval with Filtering Method, Filtering Method=LONGLLMLINGUA2024.06 | 137.2 | — | 49.8 | 24.4 | 23.9 | 76.2 | -4.61 | |
| Top 1 documentRetrieval Strategy=Retrieval without Noise Filtering, Documents=Top 12024.06 | 102.9 | — | 46.5 | 28.3 | 21.4 | 75.7 | -4.67 | |
| RANKGPTRetrieval Strategy=Retrieval with Filtering Method, Filtering Method=RANKGPT2024.06 | 102.8 | — | 47.5 | 28.3 | 23.3 | 76.1 | -4.7 | |
| LLAMA2-7BRetrieval Strategy=Retrieval with Filtering Method, Filtering Method=LLAMA2-7B2024.06 | 30 | — | 51.4 | 23.1 | 26 | 76.3 | -4.76 | |
| Ours w/SFT w/DPORetrieval Strategy=Retrieval with Filtering Method, Fine-tuning protocol=SFT w/ DPO2024.06 | 13.3 | — | 52.1 | 12.5 | 16.8 | 78.2 | -4.88 | |
| Ours w/SFTRetrieval Strategy=Retrieval with Filtering Method, Fine-tuning protocol=SFT2024.06 | 11.6 | — | 50.3 | 14.6 | 15.2 | 77.4 | -4.79 | |
| LLAMA2-13BRetrieval Strategy=No Retrieval, Generator=LLAMA2-13B2024.06 | 0 | — | 49.9 | — | — | 76.7 | -4.68 |