[CameronNemo]

They should really have trained models based on the license, so a GPL-2.0-only model, 2+, 3+, 3 only LGPL 2.1(+), CDDL, MIT, et cetera.

As it stands, the combined inputs leaves the model in the most murky of gray areas.

[remram]

Knowing which license applies is not sufficient if that license requires attribution. Or would you list the authors of every single input?