ML//training//dataset//The Stack

The Stack is the largest open code dataset, built by BigCode (HuggingFace and ServiceNow) as the code companion to The Pile. StarCoder and other open code models train on it, which makes it the foundation of open-source code generation: it is to code what Common Crawl is to text.

The Stack is the largest open code dataset, built by BigCode (HuggingFace and ServiceNow) as the code companion to The Pile. StarCoder and other open code models train on it, which makes it the foundation of open-source code generation: it is to code what Common Crawl is to text.

Version 1 holds 6.4 TB of permissively licensed code in 358 languages; version 2 expanded it, with better deduplication.