First thing I tried to find on that webpage is whether I can run Gemma models on the app with inference on-device. Apparently you can't.