Summary

In this chapter, we laid down the foundations with a theoretical section on DevOps. Then, we moved on to MLOps and its core components and principles. Finally, we presented how LLMOps differs from MLOps by introducing strategies such as prompt monitoring, guardrails, and human-in-the-loop feedback. Also, we briefly discussed why most companies would avoid training LLMs from scratch but choose to optimize them for their use case through prompt engineering or fine-tuning. At the end of the theoretical portion of the chapter, we learned what a CI/CD/CT pipeline is, the three core dimensions of an ML application (code, data, model), and that, after deployment, it is more critical than ever to implement a monitoring and alerting layer due to model degradation.

Next, we learned how to deploy the LLM Twin’s pipeline to the cloud. We understood the infrastructure and went step by step through deploying MongoDB, Qdrant, the ZenML cloud, and all the necessary AWS resources...