MU-NLPC/Calc-math_qa

Calc-math_qa数据集是基于math_qa数据集构建的，主要用于训练能够使用外部工具（如计算器）进行链式推理的模型。数据集包含数学问题的自然语言描述、逐步计算过程、正确答案等信息。数据集经过处理，将嵌套公式线性化，并替换了所有高级函数调用为基本操作。数据集的默认配置移除了测试和验证集，建议仅用于训练。The Calc-math_qa dataset is built upon the math_qa dataset, and is primarily designed for training models that can leverage external tools (e.g., calculators) to conduct chained reasoning. This dataset includes natural language descriptions of mathematical problems, step-by-step calculation processes, correct answers, and other relevant information. It has been processed to linearize nested formulas and replace all advanced function calls with basic arithmetic operations. The default configuration of this dataset removes the test and validation sets, and it is recommended to use it only for training purposes.

原始信息汇总

数据集概述

数据集名称

• 名称: Calc-math_qa

许可证

• 许可证: Apache 2.0

配置信息

• 默认配置:

  • 数据文件路径: data/train-*
  • 训练集:
    • 字节数: 25058735
    • 示例数: 20868
  • 下载大小: 11157481
  • 数据集大小: 25058735

• 原始分割配置:

  • 数据文件路径:
    • 训练集: original-splits/train-*
    • 验证集: original-splits/validation-*
    • 测试集: original-splits/test-*
  • 训练集:
    • 字节数: 25058735
    • 示例数: 20868
  • 验证集:
    • 字节数: 3722848
    • 示例数: 3102
  • 测试集:
    • 字节数: 2423833
    • 示例数: 2029
  • 下载大小: 13928430
  • 数据集大小: 31205416

特征信息

• 通用特征:
  • id: 字符串类型
  • question: 字符串类型
  • chain: 字符串类型
  • result: 字符串类型
  • result_float: 浮点数类型
  • question_without_options: 字符串类型
  • options: 结构体类型，包含选项A至E，均为字符串类型
  • annotated_formula: 字符串类型
  • linear_formula: 字符串类型
  • rationale: 字符串类型
  • category: 字符串类型

数据集使用建议

• 默认配置: 推荐仅用于训练
• 原始分割配置: 包含训练、验证和测试集，但建议谨慎使用原始的验证和测试集

搜集汇总

背景与挑战

背景概述

This dataset consists of mathematical problems converted into a structured format that includes questions, computational steps, and correct answers, aimed at training models to use external tools for arithmetic reasoning. It features 46,867 entries with attributes like question, solution chain, and result, supporting tasks that require precise calculations and reasoning.