Programs compiled with GCC are not GPL by extension. Arguably the net weights are a kind of program themselves.
If you have a non-GPL license to use and redistribute the data you use to train the neural net, I would imagine not.
There are long running disagreements about what this means under GPLv2 in various cases.
Under GPLv3, the answer is explicitly clear: "The output from running a covered work is covered by this License only if the output, given its content, constitutes a covered work."
Re:Our lawsuit against ChessBase
by subroutine on Tuesday July 20, 2021 @04:03PM
(#27898547)
Sorry, but could you clarify whether the NN weights constitute a covered work? If so, why? If not, what sort of output would constitute a covered work?
An example of an output constituting a covered work might be that if you use a GPLv3-licensed library to output a timezone database shipped with the library in a JSON format, because the content of the JSON output is itself a derivative of a GPLv3 work, not only an output, it would be covered by the GPLv3.
In order for NN weights to be covered, the weights would need to be in themselves a derivative of a GPLv3-licensed work.
Training a NN on GPLv3 licensed training sets would (IMHO) make it a derivative work, that's what a lot of the Copilot uproar is about.
Should I be legally compelled to share the net weights of my trained NN?