Operating machine learning systems for standard models is established practice. However, large language models (LLMs), often containing billions or trillions of parameters (e.g., a model with $P$ parameters where $P \gg 10^9$), introduce distinct operational requirements that go beyond traditional MLOps. Their sheer size, computational demands for training and inference, and unique failure modes necessitate specialized approaches.

This chapter establishes the foundational concepts of LLMOps. We will examine how established MLOps principles adapt, and where they fall short, when applied to LLMs. You will learn about:

• The key differences distinguishing LLMOps from conventional MLOps.
• The specific operational difficulties presented by LLMs, including scale, cost, and performance management.
• The necessary hardware and software infrastructure for supporting the LLM lifecycle.
• An overview of the distinct stages involved in developing, deploying, and maintaining LLMs.
• Considerations for selecting appropriate tools and platforms for an LLMOps stack.

By the end of this chapter, you will understand the unique context and requirements for managing large models operationally, preparing you for the specific techniques covered in subsequent chapters.