Currently, MainStream's trainer only saves the accuracy of models, and not the models themselves. The models will be needed to deploy to Streamer.