Summary

In this chapter, we introduced the core concept of MLOps, especially in the context of vision and language. We discussed machine learning operations, including some of the technologies, people, and processes that make it work. We especially focused on the pipeline aspect, learning about technologies useful to build them, such as SageMaker Pipelines, Apache Airflow, and Step Functions. We looked at a handful of different types of pipelines relevant to machine learning, such as model deployment, model retraining, and environment promotion. We discussed core operations concepts, such as CI and CD. We learned about model monitoring and human-in-the-loop design patterns. We learned about some specific techniques for vision and language within MLOps, such as common development and deployment pipelines for large language models. We also looked at how the core methods that might work in language can be inherently less reliable in vision, due to the core differences in the modalities...