dataseek/ptbr-gov-legal

PT-BR Legal & Government Documents 数据集是 MagTina350m 预训练语料库的一部分，由 Dataseek 发布。它包含 935,000 个巴西法律和政府文件，涵盖联邦/州法律、法院判决、法规和官方通信。该数据集通过合并 eduagarcia/LegalPT_dedup（来自 HuggingFace）和一个 Kaggle 巴西法律程序转储创建，并经过过滤（如 alpha_ratio≥0.65、文本长度≥200 字符、语言检测为葡萄牙语）和去重处理。数据模式包括来源标识、文本内容、字符数、单词数和元数据。数据集大小为 935,685 行，5.96 亿字符，估计 1.32 亿令牌，用于巴西葡萄牙语语言模型的预训练、领域适应和 NLP 研究。许可证主要为 CC0 1.0，但部分数据继承 CC-BY-SA 义务。PT-BR Legal & Government Documents is part of the MagTina350m pretrain corpus release by Dataseek. It contains 935 K Brazilian legal and government documents, including federal/state laws, court decisions, regulatory acts, and official communications. The dataset is a mixed corpus combining eduagarcia/LegalPT_dedup from HuggingFace with a Kaggle Brazilian-legal-proceedings dump. It has been filtered (alpha_ratio ≥ 0.65, text length ≥ 200 chars, FastText langid=pt) and deduplicated (SHA-1). The schema includes source identifier, text body, character count, word count, and metadata. The dataset has 935,685 rows, 5.96 billion characters, and an estimated 1.32 billion tokens, used for pre-training Brazilian Portuguese language models, domain adaptation, and NLP research. Licensed under CC0 1.0 with some CC-BY-SA obligations.