TemplateMath Part I: TemplateGSM
收藏资源简介:
TemplateMath Part I: TemplateGSM是由清华大学IIIS实验室创建的一个大规模小学数学问题数据集。该数据集包含超过700万条合成生成的小学数学问题,每个问题都配有代码和自然语言的解决方案。数据集的创建过程利用了GPT-4生成的参数化模板,通过自动生成和验证确保了数据的高质量和多样性。该数据集主要用于预训练、微调和评估大型语言模型在数学推理任务中的表现,旨在解决现有数学数据集规模和多样性不足的问题。
TemplateMath Part I: TemplateGSM is a large-scale primary school mathematics problem dataset developed by the Institute for Interdisciplinary Information Sciences (IIIS) at Tsinghua University. This dataset contains over 7 million synthetically generated primary school math problems, each paired with both code-based and natural language solutions. The dataset construction process leverages parameterized templates generated by GPT-4, and ensures high data quality and diversity through automated generation and validation procedures. Primarily intended for pre-training, fine-tuning, and evaluating large language models (LLMs) on mathematical reasoning tasks, this dataset aims to address the limitations of existing math datasets in terms of scale and diversity.




