Single-hop Question Answering on SQuAD
86.8F1 ScoreIN-CONTEXT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| IN-CONTEXT2026.02 | 86.8 | — | |
| GEN ADAPTER2026.02 | 70.3 | — | |
| SHINETraining Data Scale=Full, Training Data Source=MQA + 1QA2026.02 | 63.6 | — | |
| SHINETraining Data Scale=1/2, Training Data Source=1QA2026.02 | 62.2 | — | |
| SHINETraining Data Scale=1/2, Training Data Source=MQA2026.02 | 59.3 | — | |
| AmberRetrieval Strategy=Amber, Backbone=Llama3-8b2025.02 | 39.37 | 34.6 | |
| AmberRetrieval Strategy=Amber, Backbone=GPT-3.52025.02 | 39.06 | 35.8 | |
| AmberRetrieval Strategy=Amber, Backbone=Qwen2-7b2025.02 | 38.43 | 36.8 | |
| Adaptive-RAGRetrieval Strategy=Adaptive RAG, Backbone=Llama3-8b2025.02 | 38.3 | 33 | |
| VanillaRetrieval Strategy=Single-time RAG, Backbone=GPT-3.52025.02 | 37.88 | 34.4 | |
| VanillaRetrieval Strategy=Single-time RAG, Backbone=Llama3-8b2025.02 | 36.08 | 30.4 | |
| FLARERetrieval Strategy=Adaptive RAG, Backbone=Llama3-8b2025.02 | 35.81 | 32.9 | |
| Self-RerankRetrieval Strategy=Single-time RAG, Backbone=Llama3-8b2025.02 | 35.19 | 31.1 | |
| ReActRetrieval Strategy=Adaptive RAG, Backbone=Llama3-8b2025.02 | 34.85 | 33.6 | |
| Chain-of-noteRetrieval Strategy=Single-time RAG, Backbone=Llama3-8b2025.02 | 33.94 | 31.8 | |
| Self-RAGRetrieval Strategy=Adaptive RAG, Backbone=Llama3-8b2025.02 | 33.84 | 32.7 | |
| Adaptive-NoteRetrieval Strategy=Adaptive RAG, Backbone=Llama3-8b2025.02 | 33.61 | 29 | |
| Self-RefineRetrieval Strategy=Single-time RAG, Backbone=Llama3-8b2025.02 | 33.04 | 32.1 | |
| VanillaRetrieval Strategy=Single-time RAG, Backbone=Qwen2-7b2025.02 | 27.7 | 32.2 | |
| NAIVE2026.02 | 22 | — | |
| NORRetrieval Strategy=No Retrieval, Backbone=Llama3-8b2025.02 | 18.41 | 12.6 |