Question Answering on NQ (EM and F1)
44EMSelf-RAG
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Self-RAGBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 44 | 52.2 | |
| Iter-RetGenBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 40.8 | 52.31 | |
| Naive RAGBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 38.2 | 50.08 | |
| VOTE-RAGBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs, parallel agents (N)=32026.03 | 37.2 | 50.85 | |
| DRAGBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs, max debate interactions (r)=3, agents=32026.03 | 36.8 | 50.38 | |
| SuReBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 34.8 | 51.35 | |
| IRCoTBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 28.6 | 37.36 | |
| FLAREBase LLM=Llama-3.1-8B-Instruct, Retriever=E5-base-v2, Top-k=3 paragraphs2026.03 | 19.4 | 27.68 |