Multilingual Document Retrieval on MIRACL (Evaluation set)
71.7nDCG@10DRAMA1B
Evaluation Results
| Method | Links | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DRAMA1BNon-Emb. Param.=1B, Repre. Dim.=2048, Contra. Pretrain.=false, Data Aug.=true, Multi. Lang.=true2025.02 | 71.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DRAMA0.3BNon-Emb. Param.=265M, Repre. Dim.=1024, Contra. Pretrain.=false, Data Aug.=true, Multi. Lang.=true2025.02 | 71.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DRAMA1B (768d)Non-Emb. Param.=1B, Repre. Dim.=768, Contra. Pretrain.=false, Data Aug.=true, Multi. Lang.=true2025.02 | 70.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DRAMA0.1BNon-Emb. Param.=113M, Repre. Dim.=768, Contra. Pretrain.=false, Data Aug.=true, Multi. Lang.=true2025.02 | 70.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M3-BGE-DenseNon-Emb. Param.=303M, Repre. Dim.=1024, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 69.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mE5-InstNon-Emb. Param.=303M, Repre. Dim.=1024, Contra. Pretrain.=true, Data Aug.=true, Multi. Lang.=true2025.02 | 66 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mE5-largeNon-Emb. Param.=303M, Repre. Dim.=1024, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 65.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ArcticEmb-v2-LNon-Emb. Param.=303M, Repre. Dim.=1024, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 64.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MistralE5Non-Emb. Param.=7B, Repre. Dim.=4096, Contra. Pretrain.=false, Data Aug.=true, Multi. Lang.=true2025.02 | 62.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mGTE-DenseNon-Emb. Param.=113M, Repre. Dim.=768, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 62.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| mE5-baseNon-Emb. Param.=86M, Repre. Dim.=768, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 60.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ArcticEmb-v2-MNon-Emb. Param.=113M, Repre. Dim.=768, Contra. Pretrain.=true, Data Aug.=false, Multi. Lang.=true2025.02 | 59.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GeckoNon-Emb. Param.=1B, Repre. Dim.=768, Contra. Pretrain.=true, Data Aug.=true, Multi. Lang.=true2025.02 | 56.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BM25Contra. Pretrain.=false, Data Aug.=false, Multi. Lang.=true2025.02 | 38.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BM252023.04 | — | 48.1 | 50.8 | 35.1 | 31.9 | 33.3 | 55.1 | 18.3 | 45.8 | 44.9 | 36.9 | 41.9 | 33.4 | 38.3 | 49.4 | 48.4 | 18 | — | — | — | |
| BM25 + mDPR2023.04 | — | 67.3 | 65.4 | 54.9 | 64.1 | 59.4 | 67.2 | 52.3 | 61.6 | 44.3 | 57.6 | 60.9 | 53.2 | 44.6 | 60.2 | 59.9 | 52.6 | — | — | — | |
| mContrieverTraining Data=English MS MARCO2023.04 | — | 55.3 | 54.2 | 37.9 | 34.1 | 42.6 | 51.2 | 31.5 | 40.6 | 36.8 | 38.3 | 46.2 | 39.9 | 44.4 | 48.7 | 52.4 | 27.4 | 32.9 | 32.9 | 41.5 | |
| mContrieverTraining Data=English MS MARCO + MIRACL (16 languages)2023.04 | — | 64.6 | 66.4 | 41.2 | 40.3 | 46.3 | 61.9 | 42.9 | 41.9 | 44.6 | 55.6 | 55.4 | 48.1 | 65.3 | 77.6 | 69.3 | 45.9 | 39.6 | 41.9 | 52.7 | |
| mDPR2023.04 | — | 49.9 | 44.3 | 39.4 | 47.8 | 48 | 47.2 | 43.5 | 38.3 | 27.2 | 43.9 | 41.9 | 40.7 | 29.9 | 35.6 | 35.8 | 51.2 | — | — | — | |
| mXTRbaseTraining Data=English MS MARCO2023.04 | — | 66.1 | 64.7 | 49.4 | 40.5 | 47.9 | 62.2 | 37.5 | 51.4 | 46.9 | 56.8 | 64 | 49.8 | 43 | 67.7 | 69.2 | 47.2 | 34.5 | 40.6 | 52.2 | |
| mXTRbaseTraining Data=English MS MARCO + MIRACL (16 languages)2023.04 | — | 73 | 73.9 | 46.1 | 42.6 | 51 | 70.5 | 39.3 | 51.3 | 54.2 | 62.3 | 67.7 | 54.5 | 69.7 | 80.7 | 76.1 | 51.4 | 36.1 | 46.8 | 58.2 | |
| mXTRxxlTraining Data=English MS MARCO2023.04 | — | 74.1 | 75.5 | 56 | 52.4 | 56.1 | 75.1 | 51.4 | 61.8 | 52 | 68.7 | 67.4 | 61.3 | 69.7 | 76 | 76.9 | 56.9 | 51.7 | 60.3 | 63.5 | |
| mXTRxxlTraining Data=English MS MARCO + MIRACL (16 languages)2023.04 | — | 77.8 | 78.4 | 52.5 | 48.9 | 56 | 76 | 52.9 | 61.5 | 54.9 | 73.4 | 68.5 | 66.2 | 79.4 | 84.3 | 80.7 | 58.9 | 52.8 | 62.4 | 65.9 |