BAAI/IndustryCorpus_technology

该数据集是通过对超过100TB的开源数据集进行清洗和过滤，最终得到3.4TB的高质量多行业分类的中英文预训练数据集。数据集包含1TB的中文数据和2.4TB的英文数据，并标注了12种类型的中文数据标签。数据集在医疗行业示范模型上进行了预训练、SFT和DPO训练，结果显示客观性能提升了20%，主观胜率为82%。数据集涵盖了18个行业类别，并进行了基于规则和模型的过滤、数据去重等处理。

The dataset contains 3.4TB of high-quality multi-industry classified Chinese and English pre-training data, which has been cleaned and filtered through 22 industry data processing operators, resulting in 1TB of Chinese data and 2.4TB of English data. The dataset covers 18 industry categories and includes 12 types of labels for Chinese data, such as alphanumeric ratio, average line length, language confidence score, maximum line length, and perplexity. The performance of the dataset was validated through continued pre-training, SFT, and DPO training on a medical industry demonstration model, showing a 20% improvement in objective performance and an 82% subjective win rate.