LLM-as-a-Judge Routing on 3 datasets Average (test)
90AccuracyRACER
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RACERModel Scale=8B, Budget Constraint (C)=42026.05 | 90 | 3.9 | |
| M-IRTModel Scale=8B, Budget Constraint (C)=42026.05 | 88.9 | 3.4 | |
| RouteLLM-MFModel Scale=8B, Budget Constraint (C)=42026.05 | 88.2 | 4.1 | |
| RouterBench-KNNModel Scale=8B, Budget Constraint (C)=42026.05 | 86.8 | 2.6 | |
| RACERModel Scale=4B, Budget Constraint (C)=42026.05 | 85.8 | 3.4 | |
| RouteLLM-MFModel Scale=4B, Budget Constraint (C)=42026.05 | 84.7 | 3.4 | |
| M-IRTModel Scale=4B, Budget Constraint (C)=42026.05 | 84.3 | 2.7 | |
| RouterBench-KNNModel Scale=4B, Budget Constraint (C)=42026.05 | 84.1 | 2.5 | |
| RACERModel Scale=1.7B, Budget Constraint (C)=42026.05 | 72.2 | 3.6 | |
| M-IRTModel Scale=1.7B, Budget Constraint (C)=42026.05 | 71.6 | 3.4 | |
| RouterBench-KNNModel Scale=1.7B, Budget Constraint (C)=42026.05 | 71.3 | 2.6 | |
| RouteLLM-MFModel Scale=1.7B, Budget Constraint (C)=42026.05 | 69.4 | 3.8 |