Code Generation on APPS (Pass@4)
83.2Pass@4CPPO
Evaluation Results
| Method | Links | |
|---|---|---|
| CPPOModel=Gemma 4 E4B, K budget=42026.05 | 83.2 | |
| PKPOModel=Gemma 4 E4B, K budget=42026.05 | 81.5 | |
| Pass@K Training / RLVRModel=Gemma 4 E4B, K budget=42026.05 | 79.8 | |
| PlanSearchModel=Gemma 4 E4B, K budget=42026.05 | 79 | |
| Plan-and-SolveModel=Gemma 4 E4B, K budget=42026.05 | 77 | |
| Direct SolveModel=Gemma 4 E4B, K budget=42026.05 | 75 | |
| CPPOModel=Gemma 4 E2B, K budget=42026.05 | 74.8 | |
| PKPOModel=Gemma 4 E2B, K budget=42026.05 | 72.5 | |
| Pass@K Training / RLVRModel=Gemma 4 E2B, K budget=42026.05 | 70.8 | |
| PlanSearchModel=Gemma 4 E2B, K budget=42026.05 | 69 | |
| Plan-and-SolveModel=Gemma 4 E2B, K budget=42026.05 | 66 | |
| Direct SolveModel=Gemma 4 E2B, K budget=42026.05 | 64 |