List your company

Training Steps to Target Accuracy on AMC23

100Training StepsSAERLD
73.6251.8430608.2May 26, 2026

Evaluation results

MethodLinks
SAERLDModel=Qwen2.5-Math-1.5B2026.05100
SAERLGModel=Qwen2.5-Math-7B2026.05120
DIFFModel=Qwen2.5-Math-7B2026.05140
GAINRLModel=Qwen2.5-Math-7B2026.05240
GRPOModel=Qwen2.5-Math-7B2026.05320
DIFFModel=Qwen2.5-Math-1.5B2026.05340
ADARFTModel=Qwen2.5-Math-1.5B2026.05540
DAPOModel=Qwen2.5-Math-1.5B2026.05580
SAERLGModel=Qwen2.5-Math-1.5B2026.05580
GRPOModel=Qwen2.5-Math-1.5B2026.05680
ADARFTModel=Qwen2.5-Math-7B2026.05740
GAINRLModel=Qwen2.5-Math-1.5B2026.05760