What is this?
This is a subset of the aurora-m training data. The data has been clustered into samples by languages and for code. The license to the underlying data is based on the original data source, which is the HPLT dataset, the Pile, Refined Web, Red Pajama 1, and the Stack.
We make no claim to any copyrights under the original works, and have only compiled these into clusters for fair use machine learning research.
PLEASE USE AT YOUR OWN RISK. WE MAKE NO WARRANTIES AS TO NON-INFRINGEMENT.
Please note that some multi-lingual wikipedia articles from red pajama 1 are in the "en" folder. We will need to move those out to their individual language folders.
Usage:
We will write a more full fledged data card, but the user can use the files under each language as a cluster of data for training in context or otherwise.
- Downloads last month
- 1