Deduplicated Pile
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated
下载链接
链接失效反馈官方服务：
资源简介：
该数据集用于预训练Pythia-410m模型，包含了2000亿去重文本标记。该数据集在各类语言建模任务中发挥着关键作用，对于本论文中的实验研究至关重要。其所涉及的任务是语言建模。
This dataset is utilized for pre-training the Pythia-410m model and contains 200 billion deduplicated text tokens. It plays a critical role across various language modeling tasks and is essential for the experimental research conducted in this paper. The task associated with this dataset is language modeling.
提供机构：
EleutherAI
