Skip to content

Dataset · bigcode

Deduplicated code corpus The Stack v2

bigcode/the-stack-v2-dedup

Nearly deduplicated version of The Stack v2—a large code corpus for LLM training, grouped by repositories.

What's inside

This is a version of bigcode/the-stack-v2 with additional near-deduplication. Data is grouped by repositories. The Stack v2 family also includes the full version, this deduplicated version, filtered train-full-ids with 600+ programming languages, and the reduced train-smol. The data language is code.

What you can do with it

  • Pretraining code LLMs
  • Fine-tuning models on cleaner code
  • Evaluating deduplication quality for code corpora
Tasks
Text generation
Languages
Code
Size
1B–10B rows
Downloads this month
2.8K
Likes
148
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("bigcode/the-stack-v2-dedup")