--- license: apache-2.0 task_categories: - text-generation language: - en - zh tags: - language model pretty_name: Matrix size_categories: - n>1T --- # Matrix An open-source pretraining dataset containing 4690 billion tokens, this bilingual dataset with both English and Chinese texts is used for training neo models. ## Dataset Composition The dataset consists of several components, each originating from different sources and serving various purposes in language modeling and processing. Below is a brief overview of each component:

Common Crawl
Extracts from the Common Crawl project, featuring a rich diversity of internet text including websites, blogs, news articles, and more.
Code
A collection of coding-related data. Paper
Consists of academic and research papers covering a broad spectrum of disciplines, offering technical and domain-specific language.
Book
Comprises texts from a range of published books, encompassing literature, non-fiction, textbooks, and more.
Instruction
Features a collection of texts primarily in a Q&A format. Exam
Contains various educational materials and assessments used in academic examinations. News
A collection of texts from various journalistic sources, reporting on current events and news stories.
Wiki
Articles from various encyclopedic sources, not limited to Wikipedia, covering a wide array of topics and information.
Patent
Includes texts from patent documents, providing detailed descriptions of inventions and their applications.