README.md

---
license: apache-2.0
task_categories:
- text-generation
language:
- en
tags:
- web
- common crawl
size_categories:
- 10B<n<100B
---

# 📚 RedPajama-pro

<p align="center">
  <img src="prox-teaser.png">
</p>

[ArXiv](https://arxiv.org/abs/2409.17115) | [Models](https://huggingface.co/collections/gair-prox/prox-general-models-65f1674f0607712c4d6eec76) | [Code](https://github.com/GAIR-NLP/ProX)

RedPajama-pro is refined from [RedPajama-Data-V2](https://huggingface.co/datasets/togethercomputer/RedPajama-Data-V2) using the **ProX** refining framework.
It contains about 30B high quality tokens, ready for general language model pre-training.


## License
RedPajama-pro is based on RedPajama-Data-V2, which is made available under an apache-2.0 license; users should also abide by the CommonCrawl ToU: https://commoncrawl.org/terms-of-use/. We do not alter the license of any of the underlying data.


### Citation
```
@article{zhou2024programming,
  title={Programming Every Example: Lifting Pre-training Data Quality like Experts at Scale},
  author={Zhou, Fan and Wang, Zengzhi and Liu, Qian and Li, Junlong and Liu, Pengfei},
  journal={arXiv preprint arXiv:2409.17115},
  year={2024}
}
```