Learning From Mistakes Makes LLM Better Reasoner

Large language models (LLMs) recently exhibited remarkable reasoning capabilities on solving math problems. To further improve their reasoning capabilities, this work explores whether LLMs can LEarn from MistAkes (LEMA), akin to the human learning process. Consider a human student who failed to solve a math problem, he will learn from what mistake he has made and how to correct it. Mimicking this error-driven learning process, LEMA incorporates mistake-correction data pairs during fine-tuning LLMs. Specifically, we first collect inaccurate reasoning paths from various LLMs, and then employ GPT-4 as a "corrector" to identify the mistake step, explain the reason for the mistake, correct the mistake and generate the final answer. In addition, we apply a correction-centric evolution strategy that effectively expands the question set for generating correction data. Experiments across various LLMs and reasoning tasks show that \textsc{LeMa} consistently improves CoT-alone fine-tuning. Our further analysis sheds light on the non-homogeneous effectiveness between CoT data and correction data, and the contribution from different correction information. These results suggest a significant potential for LLMs to improve through learning from their mistakes. Our code and models are publicly available at https://github.com/microsoft/LEMA.