What considerations would you take into account when scaling a machine learning model for production at Deepmind?
