Skip to content

Latest commit

 

History

History
49 lines (29 loc) · 2.71 KB

README.md

File metadata and controls

49 lines (29 loc) · 2.71 KB
Raw

GlotCC HomePage

arXiv

GlotCC is a multilingual corpus built by the GlotLID language identification and cisnlp/Ungoliant pipeline from CommonCrawl.

Lastest version supports more than 1000 languages and is filtered based on adopted filters from C4, CCNet, MADLAD-400, RedPajama-Data-v2, OSCAR, Gopher, RefinedWeb, FineWeb, Datatrove, Dolma, Pile-CC, Pretrainer's Guide, and GlotScript. ™ The logo features a llama with the style of C.C. from the Code Geass anime reading a book.

Dataset

GlotCC Dataset, Version 1: https://huggingface.co/datasets/cis-lmu/GlotCC-V1

Running the pipeline

We forked oscar-project/ungoliant to cisnlp/ungoliant and made the necessary changes to integrate it with the GlotLID language identification model.

For detailed instructions on running the pipeline, refer to the cisnlp/ungoliant repository. The README is up-to-date.

Acknowledgements

  • We appreciate the collaborators who are collectively advancing the frontier of open datasets and LLM models.
  • Thanks to the community and friends who enable the auditing of this dataset with higher quality. Also, to everyone contributing to the GlotCC dataset.
  • Our gratitude extends to the OSCAR team for pioneering the development of open pipelines and datasets from CommonCrawl, as well as to the CommonCrawl team.

License

  • GlotCC data is released under the following licensing scheme: We do not own any of the text from which this data has been extracted. The data is licensed under the terms of the CommonCrawl Terms of Use. We license the actual packaging, metadata, and annotations of this data under the Creative Commons CC0 license.
  • Ungoliant license remains unchanged as the Apache License 2.0.
  • GlotLID license remains unchanged as the Apache License 2.0.

Citation

If you find our repo and data useful for your research, please cite:

@article{kargaran2024glotcc,
  title     = {Glot{CC}: An Open Broad-Coverage CommonCrawl Corpus and Pipeline for Minority Languages},
  author    = {Kargaran, Amir Hossein and Yvon, Fran{\c{c}}ois and Sch{\"u}tze, Hinrich},
  journal   = {Advances in Neural Information Processing Systems},
  year      = {2024},
  url       = {https://arxiv.org/abs/2410.23825}
}