Question Answering on Natural Questions (NQ) (EM and F1 Metrics)
45.68Exact Match (EM)RAG-GD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RAG-GDBackbone=Llama-3.1-8B, Retriever=E5, K=52026.05 | 45.68 | 55.84 | |
| Base adapterBackbone=Llama-3.1-8B, Retriever=E52026.05 | 43.46 | 54.6 | |
| RAG-GDBackbone=Qwen-2.5-7B, Retriever=E5, K=52026.05 | 42.91 | 52.71 | |
| Base adapterBackbone=Qwen-2.5-7B, Retriever=E52026.05 | 41.77 | 51.22 | |
| Vanilla RAGBackbone=Llama-3.1-8B, Retriever=E52026.05 | 40.72 | 52.23 | |
| RAG-GDBackbone=Llama-3.1-8B, Retriever=BM25, K=52026.05 | 40.22 | 50.01 | |
| Vanilla RAGBackbone=Qwen-2.5-7B, Retriever=E52026.05 | 39.16 | 50.03 | |
| Base adapterBackbone=Llama-3.1-8B, Retriever=BM252026.05 | 38.47 | 49.42 | |
| SAVERBackbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 37.1 | 47.8 | |
| SAVERBackbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 36.8 | 45.9 | |
| MADBackbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 36.6 | 46.9 | |
| MADBackbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 36.2 | 44.6 | |
| B-2Backbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 35.9 | 44.9 | |
| B-2Backbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 35.5 | 45.3 | |
| Self-RefineBackbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 34.6 | 43.7 | |
| RAG-GDBackbone=Qwen-2.5-7B, Retriever=BM25, K=52026.05 | 34.46 | 43.54 | |
| B-2Backbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 34.4 | 42.9 | |
| SAVERBackbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 33.9 | 43.8 | |
| Self-RefineBackbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 33.5 | 43 | |
| MADBackbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 33.4 | 42.4 | |
| CoTBackbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 33 | 41.7 | |
| CoTBackbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 32.7 | 43.4 | |
| Base adapterBackbone=Qwen-2.5-7B, Retriever=BM252026.05 | 32.57 | 41.45 | |
| Vanilla RAGBackbone=Llama-3.1-8B, Retriever=BM252026.05 | 31.41 | 40.95 | |
| Self-RefineBackbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 31.2 | 39.2 | |
| Vanilla LMBackbone=LLaMA-3.1-8B, Inference Setting=Zero-shot2026.04 | 29.5 | 38.2 | |
| CoTBackbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 29.5 | 38.6 | |
| Vanilla LMBackbone=Qwen-2.5-7B, Inference Setting=Zero-shot2026.04 | 28.2 | 36.9 | |
| Vanilla RAGBackbone=Qwen-2.5-7B, Retriever=BM252026.05 | 27.61 | 36.66 | |
| Vanilla LMBackbone=LLaMA-3.2-3B, Inference Setting=Zero-shot2026.04 | 25.1 | 34.8 | |
| Query OnlyBackbone=Llama-3.1-8B, Retriever=None2026.05 | 22.46 | 32.51 | |
| Query OnlyBackbone=Qwen-2.5-7B, Retriever=None2026.05 | 15.95 | 24.28 |