alea-institute/kl3m-data-dotgov

KL3M数据集是ALEA研究所提供的一个包含超过1.32亿个文档和数万亿个tokens的大规模数据集，这些文档来源于16个不同的经过版权和许可协议验证的数据源。该数据集旨在为大型语言模型提供无版权问题的训练资源，并以Parquet文件格式存储文档文本和元数据信息。数据集遵循CC BY 4.0许可协议，并且提供了适用于法律、金融和企业文档的kl3m-004-128k-cased tokenizer。

The KL3M dataset is a large-scale dataset provided by the ALEA Institute, containing over 132 million documents and trillions of tokens from 16 different sources verified for strict copyright and licensing protocols. The dataset aims to provide copyright-clean training resources for large language models and is stored in Parquet file format with document text and metadata. It follows the CC BY 4.0 license and includes the kl3m-004-128k-cased tokenizer optimized for legal, financial, and enterprise documents.