Mathematical Reasoning on GSM8K (Result, Process)
91.66AccuracyStaRPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| StaRPOModel=Qwen 7B2026.04 | 91.66 | 91.43 | |
| λ–GRPOModel=Qwen 7B2026.04 | 90.9 | 89.08 | |
| CPPOModel=Qwen 7B2026.04 | 90.52 | 88.93 | |
| EntropyModel=Qwen 7B2026.04 | 90.37 | 89.23 | |
| GRPOModel=Qwen 7B2026.04 | 90.3 | 89.39 | |
| PlannerModel=Qwen 7B2026.04 | 89.76 | 89.08 | |
| OriginalModel=Qwen 7B2026.04 | 89.46 | 88.63 | |
| StaRPOModel=Qwen 1.5B2026.04 | 79.68 | 78.54 | |
| λ–GRPOModel=Qwen 1.5B2026.04 | 78.17 | 76.95 | |
| CPPOModel=Qwen 1.5B2026.04 | 76.8 | 73.62 | |
| GRPOModel=Qwen 1.5B2026.04 | 73.92 | 72.02 | |
| EntropyModel=Qwen 1.5B2026.04 | 65.05 | 64.29 | |
| PlannerModel=Qwen 1.5B2026.04 | 60.35 | 58.83 | |
| OriginalModel=Qwen 1.5B2026.04 | 55.72 | 52.39 |