Yes, I train my models on my GPU cluster, and then transfer them to my iOS app, which does image recognition. Some of the models can be quite large for say, 3G

Yes, I train my models on my GPU cluster, and then transfer them to my iOS app, which does image recognition. Some of the models can be quite large for say, 3G (225mb), but I'm working on various compression techniques now too.