Mathematical Reasoning on Five-Bench Unweighted Mean
31.06Average ScoreRollout-level |Ai| replay
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Rollout-level |Ai| replayModel=Qwen3-4B2026.06 | 31.06 | 50.98 | |
| GRPOModel=Qwen3-4B2026.06 | 26.71 | 40.92 | |
| uniformModel=Qwen3-4B2026.06 | 26.7 | 42.99 | |
| σgModel=Qwen3-4B2026.06 | 26.69 | 46.01 | |
| Rollout-level |Ai| replayModel=Qwen3-1.7B2026.06 | 16.58 | 32.35 | |
| σgModel=Qwen3-1.7B2026.06 | 16.12 | 28.33 | |
| uniformModel=Qwen3-1.7B2026.06 | 15.93 | 27.99 | |
| GRPOModel=Qwen3-1.7B2026.06 | 15.75 | 27.64 | |
| Rollout-level |Ai| replayModel=Qwen3-0.6B2026.06 | 11.58 | 23.24 | |
| σgModel=Qwen3-0.6B2026.06 | 11.31 | 22.57 | |
| GRPOModel=Qwen3-0.6B2026.06 | 11.2 | 21.32 | |
| uniformModel=Qwen3-0.6B2026.06 | 11.11 | 20.01 |