List your company

Mathematical Reasoning on MATH 500 (Accuracy, Token Count, ReL)

90.4AccuracyMixed Attention Module
87.2888.0988.989.71Feb 1, 2026

Evaluation results

MethodLinks
Mixed Attention ModuleSeries=Qwen2.5-7B, beta=0.4, alpha=0.32026.0290.44,244142.2
Qwen2.5-7B (Base)Series=Qwen2.5-7B, beta=0.4, alpha=0.32026.0287.44,253161.6