ethanker/math-reasoning-packed-1024
收藏Hugging Face2025-10-27 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/ethanker/math-reasoning-packed-1024
下载链接
链接失效反馈官方服务：
资源简介：
这是一个为GPT风格预训练优化的打包令牌数据集，包含预令牌化和打包的序列，适用于高效变压器训练。数据集的词汇量为100,277，序列长度为1024，由多个开源数学生成对话数据集组合而成，经过一系列处理流程，包括下载、标准化、去重、过滤、分割和打包。
This is a packed token dataset optimized for GPT-style pretraining, containing pre-tokenized and packed sequences for efficient transformer training. The dataset has a vocabulary size of 100,277 and a sequence length of 1024, compiled from multiple open-source math and conversational datasets, and processed through a pipeline including downloading, normalizing, deduplicating, filtering, splitting, and packing.
提供机构：
ethanker
