david-thrower/cosmopedia-100k-simple-text

Cosmopedia是一个大型合成数据集，由HuggingFaceTB创建，包含超过3000万个文件和2500亿个标记。这个数据集由Mixtral-8x7B-Instruct-v0.1生成，覆盖了广泛的话题，并分为8个部分，包括网页样本、教育资源和指令调整数据集。它旨在支持合成数据领域的研究工作。HuggingFaceTB/cosmopedia-100k是Cosmopedia的一个微型子集，适合用于小规模测试和训练小型模型。这个数据集被重构为单轮对话模型训练的简单字符串列表。

Cosmopedia is a large-scale synthetic dataset created by HuggingFaceTB, containing over 30 million files and 25 billion tokens. Generated by Mixtral-8x7B-Instruct-v0.1, it covers a wide range of topics and is divided into 8 splits, including web samples, educational resources, and instruction-tuning datasets as seed samples. It aims to support research efforts in the field of synthetic data. HuggingFaceTB/cosmopedia-100k provides a miniature subset of the dataset suitable for small-scale testing and training smaller models. This fork of the dataset is reformatted as a simple iterable list of strings for single-turn chat model training.