x0root/Math-HQ-20k

Math-HQ-20k是一个精心策划的合成数据集，包含20,000个英语数学推理示例，专为监督微调、推理评估和错误分析工作流设计。每个记录包含问题陈述、一个或多个解决路径、一致性核对、可能的错误目录和简短的概念总结。数据集涵盖了120个主题，难度级别从1到10，格式为JSONL，语言为英语，许可证为MIT。文件大小约为72.2 MB。数据集的技术规格包括总示例数、唯一主题数、难度范围、难度分布、JSONL有效性、字段完整性等。此外，还提供了数据集的预期用途、注意事项和限制。

Math-HQ-20k is a curated synthetic dataset of 20,000 English-language math reasoning examples designed for supervised fine-tuning, reasoning evaluation, and error-analysis workflows. Each record contains a problem statement, one or more solution paths, a consistency reconciliation, a catalogue of plausible mistakes, and a short conceptual takeaway. The dataset covers 120 topics, with difficulty levels ranging from 1 to 10, in JSONL format, in English, under the MIT license. The file size is approximately 72.2 MB. Technical specifications include the total number of examples, unique topics, difficulty range, difficulty distribution, JSONL validity, field completeness, etc. Additionally, the intended uses, notes, and limitations of the dataset are provided.