Skip to content

Dataset · ZF5761

The Stack v3: giant GitHub source code corpus

ZF5761/stack-v3-train

Hundreds of millions of source code files with full repository context for pre-training code language models.

What's inside

The largest open source code dataset ranging from 100 million to 1 billion records in text and tabular formats. Data was collected directly from GitHub as a continuation of The Stack v2 series and organized preserving full repository context. The dataset is intended for pre-training programming language models.

What you can do with it

  • Pre-training long-context code LLMs from scratch
  • Fine-tuning code autocompletion and analysis models at the full-repository level
Tasks
Text generation
Languages
Code
Size
100M–1B rows
Downloads this month
22.9K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("ZF5761/stack-v3-train")