metadata

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

📚 RedPajama-pro

ArXiv | Models | Code

RedPajama-pro is refined from 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}
}