Mathematical Reasoning on GSM8K (test) (Pass@1, Efficiency Metrics)
93.3Pass@1DRGRPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DRGRPOBackbone=Qwen3-4B-Instruct2026.02 | 93.3 | 146.52 | 1.12 | |
| DRGRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 92.6 | 511.05 | 5.63 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 92.4 | 166.97 | 6.12 | |
| WS-GRPOBackbone=Qwen3-4B-Instruct2026.02 | 91.7 | 114.53 | 2.06 | |
| GRPOBackbone=Qwen3-4B-Instruct2026.02 | 90.6 | 161.17 | 7.79 | |
| WS-GRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 85.1 | 79.26 | 2.17 |