kp7742/YALM-pretrain5-60M
收藏Hugging Face2025-06-29 更新2025-10-25 收录
下载链接:
https://hf-mirror.com/datasets/kp7742/YALM-pretrain5-60M
下载链接
链接失效反馈官方服务:
资源简介:
YALM预训练数据-5是一个包含英语(70%)、印地语(20%)、数学(5%)和Python代码(5%)的数据集,这些数据来自不同的来源,用于语言建模任务和YALM(Yet Another Language Model)的开发。总样本数为59M,大约有36B个token,使用了2048上下文的样本打包。测试集包含6k个样本。
The YALM Pretraining Data - 5 is a mix of English (70%), Hindi (20%), Math (5%), and Python Code (5%) sourced from various origins for the language modeling task and the development of YALM (Yet Another Language Model). It contains a total of 59M samples, approximately 36B tokens with sample packing at 2048 Context. The test split includes 6k samples.
提供机构:
kp7742



