Mathematical Reasoning on AIME 2025 (P@1, P@32)
15.2P@1GRPO w/ KTAE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPO w/ KTAEBackbone=Qwen2.5-Math-7B2026.01 | 15.2 | 42.4 | |
| GRPO w/ OAR-PBackbone=Qwen2.5-Math-7B2026.01 | 14.5 | 40.8 | |
| GRPO w/ OAR-GBackbone=Qwen2.5-Math-7B2026.01 | 14.3 | 40.3 | |
| GRPO w/ Entropy AdvBackbone=Qwen2.5-Math-7B2026.01 | 13 | 39.6 | |
| GRPO w/ OAR-PBackbone=Qwen2.5-7B2026.01 | 12.2 | 33.9 | |
| GRPOBackbone=Qwen2.5-Math-7B2026.01 | 12.2 | 39.1 | |
| GRPO w/ Random AdvBackbone=Qwen2.5-Math-7B2026.01 | 11.9 | 39.4 | |
| GRPO w/ OAR-GBackbone=Qwen2.5-7B2026.01 | 11.8 | 33.2 | |
| GRPO w/ Entropy AdvBackbone=Qwen2.5-7B2026.01 | 10.3 | 32.4 | |
| GRPOBackbone=Qwen2.5-7B2026.01 | 9.4 | 31 | |
| GRPO w/ Random AdvBackbone=Qwen2.5-7B2026.01 | 9.2 | 31 | |
| Qwen2.5-Math-7BBackbone=Qwen2.5-Math-7B2026.01 | 5.7 | 33.5 | |
| Qwen2.5-7BBackbone=Qwen2.5-7B2026.01 | 2.5 | 26.4 |