Question Answering on NQ
79EMSGIC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SGICBackbone=Llama2-7B, Tuning=LoRA (PEFT), Retrieved documents=102025.06 | 79 | 81.2 | — | |
| GeAR2025.01 | 65.7 | 60.7 | — | |
| Self-RAGBackbone=Llama2-7B, Tuning=Full parameter fine-tuning2025.06 | 61.3 | 70.8 | — | |
| Llama 3.3 70BModel Size=70B2025.01 | 61 | 66.9 | — | |
| Llama 3.2 3BModel Size=3B2025.01 | 57.7 | 59.9 | — | |
| TSPOBackbone=Qwen2.5-7b-Instruct2026.01 | 52.7 | — | — | |
| MT-PPOBackbone=Qwen2.5-7b-Instruct2026.01 | 49.8 | — | — | |
| TSPOBackbone=Qwen2.5-3b-Instruct2026.01 | 49 | — | — | |
| COIECDModel=FLAN-T5-3B2024.02 | 48.84 | 64.45 | — | |
| Self-Correcting RAGCategory=Self-Correcting RAG2026.04 | 48.4 | 56.2 | — | |
| COIECDModel=LLaMA2-13B2024.02 | 47.42 | 61.97 | — | |
| RECOMPCategory=Advanced Selection & Reranking2026.04 | 47.1 | 55.4 | — | |
| CADModel=LLaMA2-13B2024.02 | 46.79 | 62.89 | — | |
| SCModel=LLaMA2-13B2024.02 | 46.66 | 61.76 | — | |
| RegularModel=LLaMA2-13B2024.02 | 46.48 | 61.51 | — | |
| RAG + MMRCategory=Advanced Selection & Reranking2026.04 | 46.2 | 54.5 | — | |
| CDModel=LLaMA2-13B2024.02 | 46.19 | 62.29 | — | |
| SCModel=FLAN-T5-3B2024.02 | 46.14 | 62.51 | — | |
| RegularModel=FLAN-T5-3B2024.02 | 46 | 62.78 | — | |
| FilcoCategory=Advanced Selection & Reranking2026.04 | 45.8 | 54.2 | — | |
| RRRCategory=Standard RAG Baselines2026.04 | 45.1 | 53.8 | — | |
| HyDECategory=Standard RAG Baselines2026.04 | 44.5 | 53.2 | — | |
| ZeroSearchBackbone=Qwen2.5-7b-Instruct2026.01 | 43.8 | — | — | |
| LongLLMLinguaCategory=Advanced Selection & Reranking2026.04 | 43.8 | 52.6 | — | |
| Self-RAGCategory=Iterative & Agentic RAG2026.04 | 43.2 | 52.5 | — | |
| IRCoTCategory=Iterative & Agentic RAG2026.04 | 42.5 | 51.8 | — | |
| ZeroSearchBackbone=Qwen2.5-3b-Instruct2026.01 | 41.4 | — | — | |
| CRAGCategory=Iterative & Agentic RAG2026.04 | 40.5 | 50.1 | — | |
| CADModel=OPT-6.7B2024.02 | 40.1 | 29.15 | — | |
| NaiveCategory=Standard RAG Baselines2026.04 | 39.4 | 52.1 | — | |
| Search-R1Backbone=Qwen2.5-7b-Instruct2026.01 | 39.3 | — | — | |
| CADModel=FLAN-T5-3B2024.02 | 38.91 | 57.77 | — | |
| CDModel=OPT-6.7B2024.02 | 38.16 | 22.9 | — | |
| CDModel=FLAN-T5-3B2024.02 | 37.62 | 55.47 | — | |
| COIECDModel=OPT-6.7B2024.02 | 37.35 | 30.07 | — | |
| DRAGCategory=Iterative & Agentic RAG2026.04 | 36.8 | 50.4 | — | |
| Rejection SamplingBackbone=Qwen2.5-7b-Instruct2026.01 | 36 | — | — | |
| RAGBackbone=Qwen2.5-7b-Instruct2026.01 | 34.9 | — | — | |
| RAGBackbone=Qwen2.5-3b-Instruct2026.01 | 34.8 | — | — | |
| Search-R1Backbone=Qwen2.5-3b-Instruct2026.01 | 34.1 | — | — | |
| DPRtraining=full data2024.02 | 33.49 | — | — | |
| SFTBackbone=Qwen2.5-7b-Instruct2026.01 | 31.8 | — | — | |
| MindRefbackbone=LLaMA-2-13b2024.02 | 31.69 | — | — | |
| MindRefbackbone=LLaMA-2-7b2024.02 | 29.89 | — | — | |
| MindRefbackbone=LLaMA-7b2024.02 | 29.78 | — | — | |
| MindRefbackbone=LLaMA-13b2024.02 | 29.68 | — | — | |
| Rejection SamplingBackbone=Qwen2.5-3b-Instruct2026.01 | 29.4 | — | — | |
| R1-instructBackbone=Qwen2.5-7b-Instruct2026.01 | 27 | — | — | |
| BM252024.02 | 25.84 | — | — | |
| SFTBackbone=Qwen2.5-3b-Instruct2026.01 | 24.9 | — | — | |
| Contriever2024.02 | 24.78 | — | — | |
| Search-o1Backbone=Qwen2.5-3b-Instruct2026.01 | 23.8 | — | — | |
| PlugDPost-processing=+DPlug2023.05 | 23.01 | 32.68 | — | |
| IRCoTBackbone=Qwen2.5-7b-Instruct2026.01 | 22.4 | — | — | |
| SCModel=OPT-6.7B2024.02 | 21.4 | 24.24 | — | |
| R1-instructBackbone=Qwen2.5-3b-Instruct2026.01 | 21 | — | — | |
| LLaMA-2-13b2024.02 | 19.74 | — | — | |
| RegularModel=OPT-6.7B2024.02 | 18.15 | 19.74 | — | |
| dual-debiasing frameworkRetriever=TopK, Noise Type=Irrelevant, Noise Ratio=0.62025.05 | 15.93 | — | — | |
| Search-o1Backbone=Qwen2.5-7b-Instruct2026.01 | 15.1 | — | — | |
| T5Post-processing=+Concat2023.05 | 14.45 | 22.16 | — | |
| dual-debiasing frameworkRetriever=DPP, Noise Type=Relevant, Noise Ratio=0.42025.05 | 14.07 | — | — | |
| Naïve ICLRetriever=TopK, Noise Type=None, Noise Ratio=0.02025.05 | 14 | — | — | |
| Direct InferenceBackbone=Qwen2.5-7b-Instruct2026.01 | 13.4 | — | — | |
| T5Post-processing=None2023.05 | 11.35 | 17.11 | — | |
| PlugDPost-processing=None2023.05 | 11.17 | 16.39 | — | |
| IRCoTBackbone=Qwen2.5-3b-Instruct2026.01 | 11.1 | — | — | |
| Direct InferenceBackbone=Qwen2.5-3b-Instruct2026.01 | 10.6 | — | — | |
| LPRRetriever=TopK, Noise Type=Irrelevant, Noise Ratio=0.82025.05 | 8.47 | — | — | |
| COSEARCHMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 58.2 | |
| COSEARCHMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 53.8 | |
| CoTMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 19.8 | |
| CoTMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 19.1 | |
| Direct InferenceMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 27.9 | |
| Direct InferenceMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 19.8 | |
| EPR2023.05 | — | 39.5 | — | |
| Fixed RankerMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 57.8 | |
| Fixed RankerMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 51.5 | |
| Hyperformer++2023.05 | — | 40.24 | — | |
| Muppet2023.05 | — | 40.83 | — | |
| OLTQAAblation=without Pm2023.05 | — | 42.1 | — | |
| OLTQAAblation=without Pk2023.05 | — | 40.2 | — | |
| OLTQAKnowledge Distillation=without MKD2023.05 | — | 42.13 | — | |
| OLTQARetriever=BM252023.05 | — | 41.82 | — | |
| OLTQARetriever=EPR2023.05 | — | 42.13 | — | |
| OLTQARe-ranker=excluded2023.05 | — | 41.5 | — | |
| OLTQAKnowledge Distillation=Static MKD2023.05 | — | 42.67 | — | |
| OLTQAKnowledge Distillation=Back KD2023.05 | — | 42.07 | — | |
| OLTQAConfiguration=Full2023.05 | — | 42.68 | — | |
| OracleMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 68 | |
| OracleMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 54.8 | |
| ProQA2023.05 | — | 39.01 | — | |
| RAGMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 42 | |
| RAGMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 31.7 | |
| Retrieval OnlyMain agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 56.4 | |
| Retrieval OnlyMain agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 46.5 | |
| Search-o1Main agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 34.5 | |
| Search-o1Main agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 32.9 | |
| Search-R1Main agent backbone=Qwen2.5-7B-Instruct2026.04 | — | — | 55.6 | |
| Search-R1Main agent backbone=Qwen2.5-3B-Instruct2026.04 | — | — | 45.8 |