Datasets:

Modalities:
Tabular
Text
Formats:
parquet
Languages:
code
ArXiv:
Libraries:
Datasets
Dask
License:

🚩 Report: Legal issue(s)

#12
by arborelia - opened

This dataset is made of infringing copies of copyrighted code. It contains code with open-source licenses whose terms they have not followed, as well as code with no license ("all rights reserved").

The documentation at https://huggingface.co/datasets/bigcode/the-stack-v2 acknowledges that they have copied code with "no license", which is an obvious copyright violation.

The Stack also violates the terms of most permissive open-source licenses, because those require the code to be correctly attributed, and often require preserving some amount of license text with copies of the code. In the "Licensing information" at https://huggingface.co/datasets/bigcode/the-stack-v2#licensing-information, the Stack acknowledges that they do not preserve this information, and that The Stack is under no valid license. They claim that it is the user's responsibility to find out all the sources of copied code and put the licenses back.

They cannot delegate responsibility for following the terms of a license to others while not following it themselves. Permissive open-source licenses grant rights to people under the condition that they follow the terms of the license. People who don't follow the terms of the licenses, such as the Software Heritage Archive creating The Stack v2, are not granted any rights to copy code.

Having a supposed opt-out process is not a defense of copyright infringement.

Sign up or log in to comment