Mathematical Reasoning on HMMT-F26 (avg@16, pass@16)
16.29Avg@16Rollout-level |Ai| replay
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Rollout-level |Ai| replayModel=Qwen3-4B2026.06 | 16.29 | 30.3 | |
| σgModel=Qwen3-4B2026.06 | 11.36 | 27.27 | |
| uniformModel=Qwen3-4B2026.06 | 10.44 | 26.38 | |
| GRPOModel=Qwen3-4B2026.06 | 10.23 | 21.21 | |
| GRPOModel=Qwen3-1.7B2026.06 | 3.22 | 9.09 | |
| σgModel=Qwen3-1.7B2026.06 | 3.03 | 12.12 | |
| Rollout-level |Ai| replayModel=Qwen3-1.7B2026.06 | 2.84 | 15.15 | |
| uniformModel=Qwen3-1.7B2026.06 | 1.78 | 9.09 | |
| Rollout-level |Ai| replayModel=Qwen3-0.6B2026.06 | 1.33 | 12.12 | |
| GRPOModel=Qwen3-0.6B2026.06 | 0.76 | 9.09 | |
| σgModel=Qwen3-0.6B2026.06 | 0.76 | 12.12 | |
| uniformModel=Qwen3-0.6B2026.06 | 0.19 | 3.03 |