Document Retrieval on MS MARCO Document (dev)
0.479MRR@100LLaMA2-Llama2Vec
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaMA2-Llama2VecSize=7B, Fine-tuning strategy=hard2023.12 | 0.479 | 0.941 | |
| LLaMA2-RepLLaMASize=7B, Fine-tuning strategy=hard2023.12 | 0.456 | — | |
| DupMAETraining stage=stage 22023.05 | 0.451 | 0.95 | |
| RetroMAETraining stage=stage 22023.05 | 0.432 | 0.935 | |
| COSTA2023.05 | 0.422 | 0.919 | |
| COSTASize=110M, Fine-tuning strategy=hard2023.12 | 0.422 | 0.919 | |
| Adore2023.05 | 0.405 | 0.919 | |
| ADORESize=110M, Fine-tuning strategy=hard2023.12 | 0.405 | 0.919 | |
| COIL2023.05 | 0.397 | — | |
| COILSize=110M, Fine-tuning strategy=hard2023.12 | 0.397 | — | |
| ICT2023.05 | 0.396 | 0.882 | |
| SEED2023.05 | 0.396 | 0.902 | |
| B-PROP2023.05 | 0.395 | 0.883 | |
| B-PROPSize=110M2023.12 | 0.395 | 0.883 | |
| PROP2023.05 | 0.394 | 0.884 | |
| PROPSize=110M2023.12 | 0.394 | 0.884 | |
| STAR2023.05 | 0.39 | 0.913 | |
| BERT2023.05 | 0.389 | 0.877 | |
| ANCERanking strategy=max-p2023.05 | 0.384 | 0.906 | |
| ANCE (max-p)Size=125M, Fine-tuning strategy=hard2023.12 | 0.384 | 0.906 | |
| ANCERanking strategy=first-p2023.05 | 0.377 | 0.893 | |
| ANCE (first-p)Size=125M, Fine-tuning strategy=hard2023.12 | 0.377 | 0.893 | |
| BM252023.05 | 0.277 | 0.807 | |
| BM252023.12 | 0.277 | 0.809 |