How would you handle deploying a deep learning model to an environment with severe compute constraints?
