Datasets:

Multilinguality:
multilingual
Size Categories:
n>1T
Source Datasets:
original
ArXiv:
License:

Data hosting on Huggingface

#2
by hieuhocnlp - opened

Hi,

Thank you so much for the great work! I'm curious, are all datasets hosted on Huggingface only? Does Huggingface charge for this (since the amount of data is gigantic)? Also, is it ok if I ask how the datasets are pushed to the repository? And how long did it take for all the data to be completely pushed?

OSCAR org

Hello @hieuhocnlp ,

Thanks a lot for the message. To answer your questions, no, the dataset is also hosted on Huma-Num for distribution within the EU, but we only give access via Huma-Num for research and education in accordance with EU law.

Concerning Hugging Face, no they do not charge for this, this is all included on the free tier of their Hub at least for now. Currently, repositories are not limited in size but individual files are limited to 50GB.

We push the dataset with either git or the HF-Hub http client. Pushing took less than 6h last time, but know that our HPC has a 10Gbps connection. Sometimes the connection is cut, so for pushing I'd actually recommend using git and pushing everything in multiple commits.

Let me know if you have any further questions.

Best,
Pedro

uj changed discussion status to closed

Sign up or log in to comment