Mathematical Reasoning on MATH500 0-shot (test)
40Pass@1 (Greedy)SPG w/ mixture
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SPG w/ mixtureMode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 40 | 36.5 | 46 | 51.2 | 55.6 | |
| WD1Mode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 38.6 | 36 | 44.9 | 49.9 | 53.6 | |
| UniGRPOMode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 38.4 | 34.7 | 43.9 | 49.5 | 53.2 | |
| SPG w/ EUBOMode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 38 | 34.4 | 44.3 | 49.9 | 54 | |
| D1Mode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 37.8 | 34.3 | 43.1 | 48 | 52 | |
| LLaDA-8B-InstructMode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 32.4 | 31.5 | 40.9 | 45.7 | 48.8 | |
| LLaDA-1.5Mode=0-shot, Temperature=0.9, Generation length=256, Checkpoint selection=best2025.10 | 32.2 | 32.6 | 42.2 | 47.4 | 50.4 |