Multi-hop Retrieval on Average (HotpotQA, 2WikiMultihopQA, Musique, Morehopqa) (test)
78.44Average RecallThinkGR
Evaluation Results
| Method | Links | |
|---|---|---|
| ThinkGRModel Parameters=8B2026.05 | 78.44 | |
| w/o ThoughtModel Parameters=8B2026.05 | 73.53 | |
| w/o RLModel Parameters=8B2026.05 | 71.73 | |
| GritHopperModel Parameters=7B2026.05 | 71.58 | |
| R3-RAGModel Parameters=8B2026.05 | 65.88 | |
| RT-RAGModel Parameters=8B2026.05 | 61.47 | |
| IRCoTModel Parameters=70B2026.05 | 59.42 | |
| ITER-RETGENModel Parameters=70B2026.05 | 55.5 | |
| Auto-RAGModel Parameters=7B2026.05 | 55.17 | |
| MDRModel Parameters=110M2026.05 | 54.73 | |
| BGE-largeModel Parameters=326M2026.05 | 49.97 | |
| SEALModel Parameters=406M2026.05 | 46.84 | |
| w/o SFTModel Parameters=8B2026.05 | 46.15 | |
| SelfaskModel Parameters=70B2026.05 | 45.84 | |
| Contriever2026.05 | 45.17 | |
| BM252026.05 | 42.88 |