Question Answering on AmbigNQ
41.11Exact MatchBM25 + gte reranker (RRPO)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BM25 + gte reranker (RRPO)context setting=top-32026.04 | 41.11 | 52.18 | |
| BM25 + gte rerankercontext setting=top-32026.04 | 40.26 | 51.36 | |
| BM25 + RankZephyrcontext setting=top-32026.04 | 40.06 | 51.58 | |
| gte reranker (RRPO)Retriever=BM25, Reranker=gte-multilingual-reranker-base, Optimization=RRPO, top-k=52026.04 | 36.56 | 49.67 | |
| gte rerankerRetriever=BM25, Reranker=gte-multilingual-reranker-base, top-k=52026.04 | 35.51 | 48.55 | |
| jina reranker (RRPO)Retriever=BM25, Reranker=jina-reranker-v2-multilingual, Optimization=RRPO, top-k=52026.04 | 32.86 | 46.2 | |
| qwen3 reranker (RRPO)Retriever=BM25, Reranker=Qwen3-reranker-0.6B, Optimization=RRPO, top-k=52026.04 | 32.52 | 45.17 | |
| bge reranker (RRPO)Retriever=BM25, Reranker=bge-reranker-base, Optimization=RRPO, top-k=52026.04 | 32.37 | 45.87 | |
| jina rerankerRetriever=BM25, Reranker=jina-reranker-v2-multilingual, top-k=52026.04 | 32.27 | 45.53 | |
| bge rerankerRetriever=BM25, Reranker=bge-reranker-base, top-k=52026.04 | 31.37 | 45.01 | |
| qwen3 rerankerRetriever=BM25, Reranker=Qwen3-reranker-0.6B, top-k=52026.04 | 31.17 | 42.87 | |
| FLAREStrategy=Adaptive Retrieval2026.04 | 25.62 | 37.78 | |
| Naive RAG (BM25)Retriever=BM252026.04 | 20.33 | 32.18 | |
| DRAGINStrategy=Adaptive Retrieval2026.04 | 18.33 | 30.09 | |
| LLM + CoTReader=Qwen2.5-7B, Chain-of-Thought=true2026.04 | 10.69 | 27.85 | |
| LLM Reader onlyReader=Qwen2.5-7B2026.04 | 3.55 | 17.8 |