Hi Ludo,
Ludovic Courtès <l...@gnu.org> writes:
Ian Eure <i...@retrospec.tv> skribis:
Guix sends archive requests to SWH. SWH gives that source code
to
HuggingFace. HuggingFace demonstrably violates the licenses.
Which licenses? As has been said previously, and you can verify
for
yourself, it does not ingest code under copyleft licenses.
While this is what their paper claims[1], it doesn’t appear to be
true, since I can see my own GPL’d code in the training set. I’ve
since moved nearly all of my code off GitHub, but if you visit
their "Am I in The Stack?" page[2] and enter my old username
("ieure"), you will see pretty much every repository I ever hosted
there, including both unlicensed and GPL’d code. Some examples
are hyperspace-el, nssh-el, tl1-mode, etc. While there aren’t
LICENSE files in those repos, the file headers of all clearly
indicate that they’re GPL’d.
Unfortunately, there is no way to check for the presence of code
in the training set except by GitHub username.
What I don’t know for certain is whether these are in the training
set because they came from SWH, or because HuggingFace obtained
them through other means. Given that all the links for my GitHub
username on that "Am I in The Stack" link back to SWH, it seems
very likely that it came from them.
Thanks,
— Ian
[1]: https://arxiv.org/pdf/2402.19173 "We also exclude
copyleft-licensed code..."
[2]: https://huggingface.co/spaces/bigcode/in-the-stack
