I want to create a new dataset which I want to publish on e.g. Hugging Face. I plan to have multiple versions with different licenses, depending on which data I used.

I now wondered whether I can combine multiple licenses, i.e. having data which is licensed "less strictly" also in the dataset that is licensed "more strictly".

My question is: Is there a list, or even better, a hierarchy where I can look up which licenses are compatible?


1 Answer 1


There are several pages which offer a compatibility hierarchy or chart. Generally it's like in order of successively stricter terms:

  1. Public domain
  2. Permissive (MIT, BSD, CC-BY,...)
  3. Weak copyleft (LGPL, CC-BY-SA,...)
  4. Strong copyleft (GPL,...)
  5. Network protective (AGPL,...)

See (without particular order) e.g. on wikipedia here, mend.io, or on the GPL site.

  • 2
    Careful here: the real compatibility structure is a directed graph, which can probably be simplified to a tree. But a list suggests a linear ordering that's definitely wrong.
    – MSalters
    Mar 21 at 14:16
  • Can GPL really be applied to data? It seems to be designed specifically for software.
    – Barmar
    Mar 21 at 14:42
  • You can apply it to data - but it might not be the most suitable one as it was indeed designed with software in mind. However it might actually work... depending on what type of data we talk about. The question is, as I understand, not about the most suitable license, but about the compatibility breakdown - something which must come first. And then we or rather the OP can choose the best from those which remain available. Mar 21 at 15:30
  • 1
    If one creates a new dataset, then it's easy: one can think about the most suitable license first, consider the type of data, and publish it accordingly. However if you start to integrate further data which are handed to you with a license you cannot discuss, then you must remain compliant with those licenses - or reject integration of those data into your dataset. The latter is not necessarily always the worst choice. Mar 21 at 15:32
  • 1
    Not sure what is "weak copyleft" about CC-BY-SA. The LGPL allows combination with non-free work without triggering the redistribution requirement. CC-BY-SA does not (and is more comparable to the GPL, I think) Mar 21 at 17:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.