Hugging Face's logo

Datasets:
aurora-m
/
aurora-m-dataset-partial

Modalities:
Text
Formats:
json
Size:
100M - 1B
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Aurora-m1 Dataset (Partial)

Open Source Continual Pre-training for Multilingual Language and Code

Description

This is part of the continued pretraining dataset used to train the Aurora-M model described in [Aurora-M: Open Source Continual Pre-training for Multilingual Language and Code, COLING 2025] (https://aclanthology.org/2025.coling-industry.56/).

We found that some of the non-English dataset especially the Vietnamese portion of the dataset was corrupted or missing when transfering from LUMI to huggingface so those are not included here. We recommend using the more modern HPLT dataset, which was used in part to construct the Aurora-M non-English (Vietnamese, Finnish and Hindi) specific portions: https://huggingface.co/datasets/HPLT/HPLT3.0
Downloads last month
191
Total file size:
999 GB

Models trained or fine-tuned on aurora-m/aurora-m-dataset-partial