Teaching LLMs According to Their Aptitude: Adaptive Reasoning for Mathematical Problem Solving

TATA enables language models to adapt between chain-of-thought and tool-integrated reasoning based on their intrinsic aptitude.