Multi-hop Question Answering on HotpotQA (F1, Toks)
29.1F1 ScoreCurvFlag
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CurvFlagCost target=T, Routing strategy=Curvature-guided2026.02 | 29.1 | 148 | |
| FLARE+TexCost target=T, Routing strategy=Texture trigger2026.02 | 28.3 | 155 | |
| SelfRt+TexCost target=T, Routing strategy=Texture trigger2026.02 | 24.8 | 301 | |
| FLARECost target=T, Routing strategy=Adaptive2026.02 | 21.4 | 163 | |
| Self-RouteCost target=T, Routing strategy=Adaptive2026.02 | 18.6 | 9,271 |