Mathematical Reasoning on GSM8K (saturated)
0.96Best Single Router PerformanceGap decomposition method
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Gap decomposition methodNested bootstrap k=30, Temperature T=0.2, N=500, Model pool size=112026.07 | 0.96 | 0.993 | 0.989 | 0.033 | 0.029 | 0.004 | 12 | 3 | 17 | 3.5 | |
| Best-single routerTemperature T=0.2, Model pool size=112026.07 | 0.96 | — | — | — | — | — | — | — | — | — |