HuggingFaceTB/finemath

FineMath数据集包含从CommonCrawl中提取的数学教育内容，分为FineMath-3+和FineMath-4+两个版本，分别包含34B和9.6B的token，以及InfiMM-WebMath-3+和InfiMM-WebMath-4+两个版本，分别包含20.5B和8.5B的token。数据集通过LLama-3.1-70B-Instruct生成的注释进行过滤，保留了高质量的数学内容，特别是清晰的解释和逐步的问题解决方法。数据集的使用受到Open Data Commons Attribution License (ODC-By) v1.0许可。

The FineMath dataset consists of mathematical educational content filtered from CommonCrawl, released in two versions: FineMath-3+ with 34B tokens and FineMath-4+ with 9.6B tokens. Both versions focus on mathematical reasoning and problem solving, with FineMath-4+ offering higher quality content. Additionally, the dataset includes a filtered English text-only portion of the InfiMM-WebMath-40B dataset, classified using the same approach as FineMath. The README describes the dataset curation process, including initial content extraction, classification, and refined quality assessment. The dataset schema is provided, detailing the features and their data types. The README discusses the social impact of the dataset, potential biases, and known limitations. The dataset is licensed under the Open Data Commons Attribution License (ODC-By) v1.0.