Multi-hop Question Answering on MuSiQue (Recall %)
57.7RecallR1 Searcher-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| R1 Searcher-7BAvg Tokens=4,665, FLOPs=6.47 × 1013, Latency (s)=0.1327, Search Count=2.312025.07 | 57.7 | |
| CoRAG-8BAvg Tokens=7,683, FLOPs=1.23 × 1014, Latency (s)=0.1472, Search Count=4.002025.07 | 54 | |
| FrugalRAG-7BAvg Tokens=11,914, FLOPs=1.66 × 1014, Latency (s)=0.4098, Search Count=3.302025.07 | 52.6 | |
| SimpleDeepSearcher-7BAvg Tokens=10,027, FLOPs=1.35 × 1014, Latency (s)=0.7366, Search Count=2.862025.07 | 50.4 | |
| Search-R1-7BAvg Tokens=2,212, FLOPs=3.14 × 1013, Latency (s)=0.0676, Search Count=1.362025.07 | 38.1 | |
| O2 Searcher-3BAvg Tokens=2,923, FLOPs=1.74 × 1013, Latency (s)=0.1009, Search Count=1.952025.07 | 37 |