Training Steps to Target Accuracy on AMC23
100Training StepsSAERLD
Evaluation Results
| Method | Links | |
|---|---|---|
| SAERLDModel=Qwen2.5-Math-1.5B2026.05 | 100 | |
| SAERLGModel=Qwen2.5-Math-7B2026.05 | 120 | |
| DIFFModel=Qwen2.5-Math-7B2026.05 | 140 | |
| GAINRLModel=Qwen2.5-Math-7B2026.05 | 240 | |
| GRPOModel=Qwen2.5-Math-7B2026.05 | 320 | |
| DIFFModel=Qwen2.5-Math-1.5B2026.05 | 340 | |
| ADARFTModel=Qwen2.5-Math-1.5B2026.05 | 540 | |
| DAPOModel=Qwen2.5-Math-1.5B2026.05 | 580 | |
| SAERLGModel=Qwen2.5-Math-1.5B2026.05 | 580 | |
| GRPOModel=Qwen2.5-Math-1.5B2026.05 | 680 | |
| ADARFTModel=Qwen2.5-Math-7B2026.05 | 740 | |
| GAINRLModel=Qwen2.5-Math-1.5B2026.05 | 760 |