LLM Routing on CEVAL and GSM8K (OOD)
87.46PerformanceNIRT-Router
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| NIRT-Routeralpha (α)=0.2, beta (β)=0.82025.06 | 87.46 | 0.13 | 15.5 | |
| NIRT-Routeralpha (α)=0.8, beta (β)=0.22025.06 | 87.37 | 0.15 | 69.32 | |
| NIRT-Routeralpha (α)=0.5, beta (β)=0.52025.06 | 87.24 | 0.14 | 42.3 | |
| MIRT-Routeralpha (α)=0.2, beta (β)=0.82025.06 | 87.18 | 0.13 | 15.45 | |
| MIRT-Routeralpha (α)=0.8, beta (β)=0.22025.06 | 87.12 | 0.14 | 69.17 | |
| MIRT-Routeralpha (α)=0.5, beta (β)=0.52025.06 | 87.12 | 0.14 | 42.25 | |
| RouterBenchalpha (α)=0.5, beta (β)=0.52025.06 | 85.75 | 0.16 | 41.37 | |
| RouterBenchalpha (α)=0.8, beta (β)=0.22025.06 | 85.5 | 0.26 | 67.42 | |
| Large LLMalpha (α)=0.8, beta (β)=0.22025.06 | 84.9 | 5.3 | 47.92 | |
| Large LLMalpha (α)=0.5, beta (β)=0.52025.06 | 84.9 | 5.3 | -7.55 | |
| Large LLMalpha (α)=0.2, beta (β)=0.82025.06 | 84.9 | 5.3 | -63.02 | |
| RouterBenchalpha (α)=0.2, beta (β)=0.82025.06 | 84.62 | 0.12 | 15.09 | |
| RouteLLMalpha (α)=0.8, beta (β)=0.22025.06 | 84.39 | 5.25 | 47.7 | |
| RouteLLMalpha (α)=0.5, beta (β)=0.52025.06 | 79.9 | 4.74 | -4.83 | |
| RouteLLMalpha (α)=0.2, beta (β)=0.82025.06 | 75.06 | 3.48 | -37.58 | |
| HybridLLMalpha (α)=0.2, beta (β)=0.82025.06 | 63.79 | 0.65 | 2.98 | |
| HybridLLMalpha (α)=0.8, beta (β)=0.22025.06 | 63.34 | 0.73 | 47.93 | |
| HybridLLMalpha (α)=0.5, beta (β)=0.52025.06 | 62.08 | 0.41 | 27.19 | |
| Small LLMalpha (α)=0.8, beta (β)=0.22025.06 | 59.83 | 0.11 | 45.67 | |
| Small LLMalpha (α)=0.5, beta (β)=0.52025.06 | 59.83 | 0.11 | 28.88 | |
| Small LLMalpha (α)=0.2, beta (β)=0.82025.06 | 59.83 | 0.11 | 10.31 |