Mathematical Reasoning on MATH 500 (Pass@k and BoN)
74.7Pass@1SimpleRL-Zoo (GRPO)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SimpleRL-Zoo (GRPO)Backbone=7B, Training Params=100%2026.02 | 74.7 | — | — | — | — | — | |
| Qwen-2.5-7B (Base)Backbone=7B2026.02 | 57.7 | 63.6 | 68.2 | 71.5 | 73.9 | — | |
| Base + Pairwise RMBackbone=1.5B, Training Params=<0.3%2026.02 | 57.7 | 63.6 | 66.2 | 67 | 67.4 | 31.3 | |
| Base + Pointwise PRMBackbone=1.5B, Training Params=<0.3%2026.02 | 57.7 | 67.1 | 71.6 | 73.5 | 73.8 | 52.1 | |
| Base + ARTS (Ours)Backbone=1.5B, Training Params=<0.3%2026.02 | 57.7 | 66 | 70.1 | 72.2 | 74.6 | 54.7 |