Although new tools enable data and research scientists to create models more effectively and efficiently, deploying models to production is still tedious and challenging. In this blogpost Luis Ceze explains why deploying machine learning models is hard, for example by clarifying the differences between ML models and software code and explaining why ML models are difficult to port. He explains how Apache TVM makes models portable by applying machine learning to optimize code generation. Finally, he introduces the Octomizer platform, which automates efficient MLOps for its users.