What I cannot execute, I do not understand: Training and Evaluating LLMs on Program Execution Traces

Armengol-Estapé, Jordi, Carbonneaux, Quentin, Zhang, Tianjun · arXiv · 2025