Mathematical Reasoning on MinervaMath (Training Steps to Target Accuracy)
220Steps to Target AccuracySAERLD
Evaluation Results
| Method | Links | |
|---|---|---|
| SAERLDModel=Qwen2.5-Math-1.5B2026.05 | 220 | |
| GRPOModel=Qwen2.5-Math-7B2026.05 | 220 | |
| GAINRLModel=Qwen2.5-Math-7B2026.05 | 220 | |
| DAPOModel=Qwen2.5-Math-1.5B2026.05 | 260 | |
| SAERLGModel=Qwen2.5-Math-7B2026.05 | 280 | |
| DIFFModel=Qwen2.5-Math-7B2026.05 | 420 | |
| SAERLGModel=Qwen2.5-Math-1.5B2026.05 | 520 | |
| DIFFModel=Qwen2.5-Math-1.5B2026.05 | 540 | |
| GRPOModel=Qwen2.5-Math-1.5B2026.05 | 560 | |
| GAINRLModel=Qwen2.5-Math-1.5B2026.05 | 600 | |
| ADARFTModel=Qwen2.5-Math-1.5B2026.05 | 860 | |
| ADARFTModel=Qwen2.5-Math-7B2026.05 | 900 |