License of model weights
Are the model weights really under the MIT license? And how does that work when the dataset or parts of it are licensed under CC non commercial?
Yes they are under the MIT license. We just followed the licensing of prior works, but we are reviewing it again internally.
Please maintain MIT. It is sorely needed in this space, and the release of public models helps to commodify the offerings of tech giants like OpenAI, which in turn helps to ensure that this generation's arms race (the race to AGI) doesn't simply go to those who were funded earliest, delaying the benefit of "network effects". Since AGI is roughly equivalent in importance to the nuclear bomb, that is akin to delaying WWIII.
Please maintain MIT.
This is sadly not a simple decision you can beg for. The code can definitely stay MIT but the weights are another problem. If the datasets are all permissively licensed there is no problem. But there is at least one dataset that has a non commercial clause. It's a general question about how the licenses of the datasets effect the trained weights.
Are there any updates regarding this issue?
Sorry for the delay in updates. After some discussions with both our team internally and our data providers, we've decided the license for XEUS will be changed to non-commercial use (CC-BY-NC-SA).