List your company

Mathematical Reasoning on GSM8K (Accuracy & Performance Gain)

91.28AccuracySDAR
85.757687.191388.62590.0587Mar 19, 2026

Evaluation results

MethodLinks
SDARBase Model=Qwen3-8B2026.0391.281.67
TraDOBase Model=Qwen3-8B2026.0391.051.44
Qwen3-8BBase Model=Qwen3-8B2026.0389.61—
Qwen2.5-7BBase Model=Qwen2.5-7B2026.0387.87—
dTRPOBase Model=Qwen2.5-7B2026.0385.97-1.9