Pre-trained Large Language Models (LLMs) offer broad capabilities, yet specific applications demand tailored behavior. This chapter establishes the groundwork for adapting these models effectively. We will begin by briefly reviewing the concepts of pre-trained models and the transformer architecture that underpins them.

You will learn why generic pre-trained models often require further adaptation for specific tasks or domains. We will examine the connection between LLM adaptation and established transfer learning concepts within Natural Language Processing. Furthermore, we will analyze how the underlying architecture of an LLM influences the choice and effectiveness of adaptation strategies. Finally, this chapter provides an overview of the different methods available for fine-tuning, ranging from modifying all model parameters to more resource-efficient techniques, setting the stage for the detailed discussions in subsequent chapters.