List your company

Mathematical Reasoning on AMC 2023 (Avg@10 accuracy)

83.25Avg@10 AccuracyAsymGRPO w/ Clip-higher
62.9768.23573.578.765Apr 6, 2026

Evaluation results

MethodLinks
AsymGRPO w/ Clip-higherBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0483.25
AsymGRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0481
GRPO w/ Clip-higherBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0478.5
AsymGRPO (βpos = βneg)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0478.5
GRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0478.25
Dr.GRPOBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0478.25
Neg-Only Modulation (§ 3.1)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0478
GRPO w/ Entro.AdvBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0477.25
Pos-Only Modulation (§ 3.1)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0476.75
GRPO w/ Entro.RegularizationBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0475.5
Pass@K TrainingBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0474
REINFORCEBase Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0473.75
EntIncrease (§ 3.2)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0471.75
EntDecrease (§ 3.2)Base Model=Qwen3-4B, Temperature=0.4, Top-p=1.02026.0471.5
Qwen3-4BTemperature=0.4, Top-p=1.02026.0463.75