دانلود کتاب علم داده در اولویت -استفاده از مدل‌های زبانی در برنامه‌های مبتنی بر هوش مصنوعی

Large Language Models (LLMs) are an unusual frontier in the development of artificial intelligence (AI). Thanks to commercial chatbots like ChatGPT, Claude, Gemini, Perplexity, and Grok, they have escaped from the bubble of technical discussion and become a familiar concept to almost every human on the planet in the span of a few short years. Machine learning engineers and data scientists are accustomed to working with techniques that very few people have heard of. So, it is something of an adjustment to have the whole world talking about your field with responses ranging from ecstatic enthusiasm to abject fear and doom mongering. In some sense, this excitement is warranted. We have broken through multiple barriers in AI simultaneously and made rapid progress on a broad range of tasks. But at the same time, the way we reliably solve problems with machine learning has not fundamentally changed. The difficult task for many technical people at present is to determine the right way to incorporate language models into their ways of working and the solutions they build, while they keep at bay the extreme opinions of their non-technical managers. This book aims to help you solve those problems. It explains the potential applications of language models, discusses many ways to evaluate them, and describes appropriate design patterns for your data science applications. To achieve the goals of this book, we need to convey several important ideas. We start the book with three chapters that lay the groundwork. The first explains what we mean by the term language model and reviews the history of developments that brought us to modern large language models. This chapter introduces lots of core terminology, and through a discussion of model architectures and training processes, it aims to build some intuitions about what language models of different kinds can be used for. What makes this chapter difficult is that the advancements of language models are likely not yet complete. It will inevitably contain some omissions of models or techniques that have emerged since I wrote these words or simply escaped my attention at the time. However, the core innovations behind language models are already well developed enough that we can derive patterns of application which will likely remain relevant. Chapter 2 overviews a wide range of specific libraries, packages, and models that are used throughout the book. These include open-source and proprietary approaches to using language models and will build up to frameworks designed for people building applications from a base layer of different language models. This chapter again introduces important terminology and programming concepts that are used throughout the chapters of the book where we start looking at specific applications. The final of the three introductory chapters delves into some core data science principles that you should keep in mind when building with language models. The idea of this chapter is to introduce, or refresh, these data science ideas to emphasize what is important about using this new technology. Key to this is the observation that these models produce output that is fundamentally stochastic and require evaluation techniques that seek to mitigate the impact of this randomness on your application users. The remainder of the book then focuses on different general strategies for using language models in data science work, starting with the most primitive idea of extracting and using semantic vectors from raw text. Each of these strategy chapters builds on the ideas of previous chapters, until we are looking at sophisticated methods for automating parts of your data science workflow or building custom data science applications. Readers who are more familiar with the landscape of language models and data science work should feel free to move between the chapters as interest takes them. Those readers new to working with language models programmatically, or to data science concepts, are encouraged to start with the initial chapters, as they will greatly aid in comprehension of the later chapters. All examples in this book were created and executed using either an Apple Mac or executed as scripts on cloud-based Linux machines. The code examples have been developed using different versions of Python 3, ranging from 3.10 to 3.13. We recommend that you create a virtual environment for the execution and testing of all examples to ensure that the required libraries do not interfere with your existing setup. The management of Python packages and environments is discussed using the standard tools of pip and venv. In the process of writing the book, I have personally switched to using uv for both these tasks. For the sake of consistency, I have omitted that change from the book. I do encourage the reader to consider adopting uv and make the effort to adapt the examples where necessary. Through the course of the book’s development, some of the proprietary models used in the examples have been deprecated. This means you will not be able to use those exact models and will need to modify the scripts to use a more recent model. In our testing, this indicated some changes in performance, both improvements and degradations. This highlights one of the problems with choosing proprietary models for building robust systems—reliability requires using models that are under your control. Finally, we have used many different packages and frameworks in the examples of this book. The intention is not to encourage use of one platform over another, but to quickly demonstrate key ideas and patterns as simply as possible. The application design principles are more important that the specific libraries we are demonstrating.__