Code Generation on LiveCodeBench v5 (880 samples)
0.46Cost4B Iterative (Lower Bound)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| 4B Iterative (Lower Bound)Ratio (S:L)=100:02026.02 | 0.46 | 62.3 | 135.4 | |
| Random SamplingRouting Mode=Static Routing, Ratio (S:L)=43:572026.02 | 1.98 | 67.5 | 34.1 | |
| Random Sampling (Online Ratio)Routing Mode=Online Adaptation, Ratio (S:L)=42:582026.02 | 2.01 | 68.2 | 33.9 | |
| Decision Tree (Gini, Depth=5)Ratio (S:L)=43:572026.02 | 2.02 | 71.2 | 35.2 | |
| Adaptive Hoeffding TreeRouting Mode=Online Adaptation, Ratio (S:L)=42:582026.02 | 2.08 | 73.6 | 35.4 | |
| 32B Iterative (Upper Bound)Ratio (S:L)=0:1002026.02 | 3.17 | 75.2 | 23.7 |