Chapter 1. LLM Applications
A large language model (LLM) is a statistical model trained on large amounts of text data to emulate human speech for natural language processing tasks (Figure 1-1), such as information extraction, text classification, speech synthesis, summarization, and machine translation. LLMOps, thus, is a framework to automate and streamline large language model (also called a foundational or generative AI model) pipelines.
While task-specific models for natural language processing (NLP) have been used in practice for a while, recent advances in NLP have shifted public interest to more task-agnostic models that allow a single model to do all of the tasks listed in the preceding paragraph.
Figure 1-1. A Venn diagram explaining the correlation among AI, ML, and LLMs1
LLMs do this by using a large number of parameters (variables that store input-output patterns in the data to help the model make predictions): LLaMA, an LLM developed by Meta, contains 65 billion parameters; PaLM, by Google, has 540 billion; and GPT-4, developed by OpenAI, is estimated to have about 1.7 trillion. These parameters allow them to capture massive amounts of linguistic and contextual information about the world and perform well on a wide range of tasks.
As such, LLMs can be used in chatbots, coding assistance tools, and many other kinds of applications. Essentially, any information ...