A hypermarket wants to construct a computer vision model to classify different product images into 40 unique products in order to organize their warehouse using robots. An image dataset was already collected by the hypermarket and a machine learning specialist was hired for this role. The specialist trained the model on Amazon Sagemaker using its ready-made Pytorch container. The model should be deployed on a Sagemaker endpoint in which the hypermarket could request it anytime from any of its branches. The model does not use intensive GPU, however, sometimes it needs a moderate GPU processing power. What is the cheapest and most appropriate instance type that the specialist should use to deploy the model?