README.md

[中文介绍] 数据集介绍：本数据集基于 GSM8K 数据集改编，训练数据集共包含约 4.5K 道数学题，测试数据集共包含784条。我们引入了大量超出常规范围的大整数，旨在测试大语言模型大数四则运算的能力，评估其在处理高精度运算时的稳定性和泛化能力 该数据集支持评估模型在以下方面的能力：

• 理解和解析包含大整数的自然语言问题；
• 执行高位数的加减乘除运算；
• 泛化到结构相似但数值范围显著扩展的问题

数据集结构

• data/train.parquet: 训练集，包含约4.5k行数据。
• data/test.parquet: 测试集，包含784行数据

数据样本结构说明 每个数据单元包含 6 个关键字段，具体定义如下：

字段名称	描述
new_question	包含大数运算的问题陈述
new_std	对应 new_question 的标准解答
general_question	通用问题模板，支持参数化替换（如 {a}、{b}、{c} 等）
general_std	解决该类问题的通用数学表达式（适用于 general_question 的所有变体）
question	原始问题文本（未参数化或未修改的初始问题）
answer_only	原始问题（question）的最终答案

[EN] Dataset Description. This dataset is adapted from the GSM8K dataset and contains approximately 4.5K math problems for training and 784 problems for testing. We introduce a large number of out-of-distribution big integers to evaluate the capabilities of large language models in performing arithmetic operations involving large numbers. The goal is to assess their stability and generalization when handling high-precision calculations. This dataset is designed to evaluate a model’s ability to:

• Understand and parse natural language problems involving large integers;
• Perform high-digit addition, subtraction, multiplication, and division;
• Generalize to problems with similar structure but significantly expanded numerical ranges.

Dataset Structure

• data/train.parquet: Training set containing approximately 4.5k entries.
• data/test.parquet: Test set containing 784 entries.

Dataset Structure Specification. Each data sample consists of 6 key fields , defined as follows:

Field Name	Description
new_question	Problem statement involving ​large-number operations
new_std	Standard solution for the new_question
general_question	General question template supporting parameterized substitution (e.g., {a}, {b}, {c}, etc.)
general_std	A general mathematical expression for solving this class of problems
question	The original question
answer_only	Standard solution for the question