large-traversaal/math500_urdu_cleaned

`math500_urdu_cleaned`是MATH-500基准的一个经过清理的双语（英语-乌尔都语）版本，包含500个具有挑战性的数学问题。该数据集专注于符号数学推理，需要多步逻辑和代数解决方案，而不仅仅是表面模式匹配。每个示例包括原始英语问题、详细的逐步解决方案和最终答案，以及问题、解决方案和答案的高质量乌尔都语翻译。这使得在低资源数学环境中评估和后训练乌尔都语和多语言推理能力语言模型成为可能。数据集名称：math500_urdu_cleaned，维护者：large-traversaal (Traversaal.ai)，原始来源：HuggingFaceH4/MATH-500，任务类型：数学推理和问题解决，领域：数学（代数、预微积分、中级代数等），语言：英语、乌尔都语，格式：Parquet，示例数量：500，主题：7个不同的数学主题类别。

`math500_urdu_cleaned` is a cleaned bilingual (English–Urdu) version of the **MATH-500** benchmark, a curated subset of 500 challenging math problems introduced in OpenAI’s *“Let’s Verify Step by Step”* work. The dataset focuses on **symbolic mathematical reasoning**, requiring multi-step logical and algebraic solutions rather than surface-level pattern matching. Each example includes the original English problem, a detailed step-by-step solution, and the final answer, along with high-quality Urdu translations of the problem, solution, and answer. This enables evaluation and post-training of **Urdu and multilingual reasoning-capable language models** in a low-resource mathematical setting. Dataset Name: math500_urdu_cleaned, Maintained by: large-traversaal (Traversaal.ai), Original Source: HuggingFaceH4/MATH-500, Task Type: Mathematical reasoning and problem solving, Domain: Mathematics (Algebra, Precalculus, Intermediate Algebra, etc.), Languages: English, Urdu, Format: Parquet, Number of Examples: 500, Subjects: 7 distinct math subject categories.