Passage Retrieval on Natural Questions (NQ) (test)
85.2Top-20 AccuracySIMLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SIMLMretrieval_mode=Retriever_distill2022.07 | 85.2 | 89.7 | — | |
| XTRxxlzero-shot=true2023.04 | 84.9 | 90.5 | — | |
| DPR-PAQBase model=RoBERTa_large2021.07 | 84.68 | 89.22 | 76.93 | |
| DPR-PAQBase model=BERT_large2021.07 | 84.4 | 88.9 | 75.3 | |
| coCondenser2022.07 | 84.3 | 89 | — | |
| DPR-PAQBase model=RoBERTa_base2021.07 | 84.01 | 89.2 | 74.15 | |
| DPR-PAQBase model=BERT_base2021.07 | 83.7 | 88.6 | 74.5 | |
| RocketQAv22022.07 | 83.7 | 89 | — | |
| PAIR2022.07 | 83.5 | 89.1 | — | |
| GTRxxlzero-shot=true2023.04 | 83.5 | 89.8 | — | |
| Condenser2022.07 | 83.2 | 88.4 | — | |
| DPR-PAQBase model=DeBERTa_xlarge-v22021.07 | 83 | 88.61 | 73.38 | |
| RocketQAPLMs=ERNIEbase2020.10 | 82.7 | 88.5 | 74 | |
| RocketQABase model=ERNIE_base2021.07 | 82.7 | 88.5 | 74 | |
| RocketQA2022.07 | 82.7 | 88.5 | — | |
| DPRmulti + BM25sparse component=true, in-domain=true2023.04 | 82.6 | 88.6 | — | |
| DPR (ours)Base model=DeBERTa_xlarge-v22021.07 | 82.38 | 87.56 | 72.66 | |
| ANCE (single)PLMs=RoBERTabase2020.10 | 81.9 | 87.5 | — | |
| ANCE (single)Base model=RoBERTa_base2021.07 | 81.9 | 87.5 | — | |
| ANCE2022.07 | 81.9 | 87.5 | — | |
| DPR (ours)Base model=RoBERTa_large2021.07 | 81.27 | 87.01 | 69.67 | |
| DPR (ours)Base model=BERT_large2021.07 | 80.19 | 86.73 | 69.14 | |
| DPRmultiretrieval pre-training=true, zero-shot=true2023.04 | 79.5 | 86.1 | — | |
| DPRTraining=Multi, Retriever=DPR2020.04 | 79.4 | 86 | — | |
| DPRTraining Setting=Multi2021.10 | 79.4 | 86 | — | |
| XTRbasezero-shot=true2023.04 | 79.3 | 88.1 | — | |
| DCSRTraining Setting=Multi2021.10 | 79.1 | 86.6 | — | |
| DPR (ours)Base model=RoBERTa_base2021.07 | 79.03 | 85.42 | 67 | |
| DCSRTraining Setting=Single2021.10 | 78.9 | 86.5 | — | |
| GTRbasezero-shot=true2023.04 | 78.5 | 86.5 | — | |
| DPRTraining=Single, Retriever=DPR2020.04 | 78.4 | 85.4 | — | |
| DPR (single)PLMs=BERTbase2020.10 | 78.4 | 85.4 | — | |
| DPR (single)Base model=BERT_base2021.07 | 78.4 | 85.4 | 65.8 | |
| DPRTraining Setting=Single2021.10 | 78.4 | 85.4 | — | |
| DPR_single2022.07 | 78.4 | 85.4 | — | |
| DPR (ours)Base model=BERT_base2021.07 | 78.3 | 85.6 | 65.5 | |
| BM25 + DPRTraining=Multi, Retriever=BM25 + DPR2020.04 | 78 | 83.9 | — | |
| BM25 + DPRTraining=Single, Retriever=BM25 + DPR2020.04 | 76.6 | 83.8 | — | |
| GAR2020.10 | 74.4 | 85.3 | — | |
| GARBase model=None2021.07 | 74.4 | 85.3 | — | |
| BM25sparse component=true, zero-shot=true2023.04 | 62.9 | 78.3 | — | |
| BM25Training=None, Retriever=BM252020.04 | 59.1 | 73.7 | — | |
| BM25 (anserini)2020.10 | 59.1 | 73.7 | — | |
| BM25 (anserini)Base model=None2021.07 | 59.1 | 73.7 | — | |
| BM252022.07 | 59.1 | 73.7 | — | |
| BM25Knowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | — | 56.1 | |
| ContrieverKnowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | — | 54.6 | |
| GritLM-7BKnowledge Representation Type=Embedding-based Knowledge Representation Augmented, Backbone=GritLM, Number of Parameters=7B2026.06 | — | — | 76.6 | |
| GTE-Qwen2-7B-InstructKnowledge Representation Type=Embedding-based Knowledge Representation Augmented, Backbone=Qwen2-7B-Instruct, Number of Parameters=7B2026.06 | — | — | 74.3 | |
| GTRKnowledge Representation Type=Unstructured Knowledge Representation Augmented2026.06 | — | — | 63.4 | |
| HippoRAG (reproduced)Knowledge Representation Type=Structured Knowledge Representation Augmented, Reproduction Status=reproduced2026.06 | — | — | 44.4 | |
| HippoRAG 2Knowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | — | 78 | |
| HippoRAG 2 (reproduced)Knowledge Representation Type=Structured Knowledge Representation Augmented, Reproduction Status=reproduced2026.06 | — | — | 77 | |
| NV-Embed-v2 (7B)Knowledge Representation Type=Embedding-based Knowledge Representation Augmented, Number of Parameters=7B2026.06 | — | — | 75.4 | |
| RAPTORKnowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | — | 68.3 | |
| SemFlowRAGKnowledge Representation Type=Structured Knowledge Representation Augmented2026.06 | — | — | 77.5 |