Oh look, another Debian discussion that will end up nowhere.
> Things in Debian main [should] be buildable from source using Debian main. In the case of a pretrained neural network, the source code is the training data.
Technically correct, completely missing the point
Neural Network weights are not source code. Models are, but they are only useful after training and plugging weights into new models go from impossible to very hard.
No, nobody is going to ship the training data again so some open source bureaucracy is happy. Not to mention it goes completely against the objective of sharing model weights.
If an open source software includes a picture of a landscape should it include a free ticket and a camera to whoever downloads the software so they can recreate the picture? No
3 comments
[ 3.1 ms ] story [ 17.5 ms ] thread> Things in Debian main [should] be buildable from source using Debian main. In the case of a pretrained neural network, the source code is the training data.
Technically correct, completely missing the point
Neural Network weights are not source code. Models are, but they are only useful after training and plugging weights into new models go from impossible to very hard.
No, nobody is going to ship the training data again so some open source bureaucracy is happy. Not to mention it goes completely against the objective of sharing model weights.
If an open source software includes a picture of a landscape should it include a free ticket and a camera to whoever downloads the software so they can recreate the picture? No