Generative AI for Everyone Deep learning, NLP, LLMs f creative practical applications (Sabesan, Karthikeyan Sivagamisundari Dutta etc.) (Z-Library)

📄 File Format: PDF

💾 File Size: 10.6 MB

📖 Read Online ⬇️ Download

Registered users can read the full content for free

Register as a Gaohf Library member to read the complete e-book online for free and enjoy a better reading experience.

📄 Page 1

(This page has no text content)

📄 Page 2

Generative AI for Everyone Deep learning, NLP, and LLMs for creative and practical applications Karthikeyan Sabesan Sivagamisundari Nilip Dutta www.bpbonline.com

📄 Page 3

First Edition 2025 Copyright © BPB Publications, India ISBN: 978-93-65897-388 All Rights Reserved. No part of this publication may be reproduced, distributed or transmitted in any form or by any means or stored in a database or retrieval system, without the prior written permission of the publisher with the exception to the program listings which may be entered, stored and executed in a computer system, but they can not be reproduced by the means of publication, photocopy, recording, or by any electronic and mechanical means. LIMITS OF LIABILITY AND DISCLAIMER OF WARRANTY The information contained in this book is true to correct and the best of author’s and publisher’s knowledge. The author has made every effort to ensure the accuracy of these publications, but publisher cannot be held responsible for any loss or damage arising from any information in this book. All trademarks referred to in the book are acknowledged as properties of their respective owners but BPB Publications cannot guarantee the accuracy of this information. www.bpbonline.com

📄 Page 4

About the Authors Karthikeyan Sabesan is a business consultant and innovation enthusiast with over 17 years of experience across various roles, including project management, business development, strategy consulting, sales, and leading a Center of Excellence. Karthikeyan specializes in innovation and business strategy building, and currently, Karthikeyan spearheads Edge computing and generative AI initiatives for enterprise businesses. He is passionate about sharing knowledge and exploring new frontiers in generative AI through his continuous blog writing. Additionally, he enjoys mentoring emerging talent and strives to make complex concepts accessible to everyone. Sivagamisundari has over 19 years of experience across multiple domains: banking, oil & gas, retail, and telecom. She handled multiple roles: cloud solution architect, Siebel consultant, test automation engineer, business analyst, and developer. She is involved in providing solutions and guiding the team to develop complete Cloud Solutions Design and deployment of Azure/AWS solutions considering sizing, infrastructure, data protection, disaster recovery, and application requirements for hybrid enterprise systems. Participated in Microsoft Hackathon around AI and was among the top 10 contestants by Techdig. Nilip Dutta is a seasoned technology professional with over 19 years of leadership experience, specializing in delivering cutting-edge solutions for clients ranging from small businesses to large enterprises. With experience in cloud computing, networking, telephony and a robust background in SRE (Site Reliability Engineering), virtualization, SAN storage, and Unix systems, Nilip has consistently demonstrated a proven ability to design innovative, scalable solutions tailored to business needs. His leadership extends across pre and post-sales processes, leveraging methodologies to meet evolving business requirements. A natural mentor and trainer, he excels at fostering collaboration among cross-functional teams and driving common goals, all while championing a customer-first approach. A thought leader and technology visionary, Nilip is passionate about exploring the forefront of emerging technologies like generative AI and quantum computing while also prioritizing sustainability and enterprise architecture.

📄 Page 5

About the Reviewers Rajiv Avacharmal is a leading expert in the field of AI/ML risk management, with a particular focus on generative AI. With a distinguished career spanning over 13 years, Rajiv has held senior leadership roles at several multinational banks and currently serves as the Corporate Vice President of AI and Model Risk at a leading Life Insurance Company. Rajiv’s research interests lie at the intersection of AI/ML, risk management, and explainable AI. Aqsa is an award-winning AI/ML product manager with expertise in the enterprise domain. An author, international speaker, and judge, she has been recognized on several forums for her product management outcomes and influence. She has 7+ years of experience and a proven track record in achieving business outcomes in products across Ads measurement, Recommendation systems, Content Management, Business intelligence platforms, and Data Analytics from her time across Google Ads, Cloud, and Assistant. She is passionate about building world-class innovative and AI product experiences and is energized working with motivated and humble teammates. Outside work, she enjoys beaches, brunches, and learning languages.

📄 Page 6

Acknowledgements We would like to express our gratitude to our colleagues and co-workers in the tech industry, whose valuable contributions over the years have taught us so much and encouraged us to think beyond our daily routines. We would like to extend a special thank you to Lekha Menon for her active role in shaping the book’s content. We are deeply thankful to our families for their patience, motivation, and encouragement at every step of our journey, constantly pushing us to pursue our ambitions and goals. We appreciate our teamwork and the mutual support that helped us stay committed to completing this book. As one of the authors of this book, I, Karthikeyan Sabesan, would like to take a moment to convey my deepest gratitude to my family—Jayaraman Rukmani, Narayanaswamy Sabesan, T.S.Rajeshwari, Vaishnavi Sabesan, and my kids. Their unwavering support, love, and encouragement have been instrumental in the creation of this book. As one of the authors of this book, I, Nilip Dutta, would like to express my heartfelt gratitude to my family—my daughter Nimisha, wife Soumi, brother Nibir and parents Smt Rita and Mr Nisith Kumar Dutta. Their unwavering support, love, and encouragement have been invaluable in bringing this book to life. As one of the authors of this book, I, Sivagamisundari, want to extend my deepest gratitude to my family for their patience, encouragement, and understanding throughout this challenging process that involved countless revisions and late nights. I want to thank my co-authors for their expertise and collaborative spirit, their openness to explore ideas, and their dedication towards making this book a reality that I truly believe the readers will enjoy. We extend our gratitude to BPB Publications for their guidance and expertise in bringing this book to life. The process of revising the book was lengthy, but it was enriched by the valuable input and collaboration of reviewers, technical experts, and editors. Most importantly, we want to thank all the readers who have shown interest in our book and supported us in making it a reality. Your encouragement has been truly invaluable.

📄 Page 7

Preface Generative AI, a branch of artificial intelligence, has made remarkable strides in recent years. These advancements have enabled a variety of applications, including the creation of lifelike images, videos, text, and music. The many benefits of generative AI can help different industries improve their processes and boost efficiency. This technology has the potential to automate and enhance human and machine tasks, as well as autonomously execute business and IT processes. However, it is important to remain aware of the risks and limitations associated with generative AI. Therefore, understanding its evolution, techniques, architecture, and potential risks is essential. This book is designed for readers of all levels, from beginners to experienced professionals. It covers fundamental concepts, practical applications, advanced topics, and includes hands-on coding examples. Chapter 1: AI Fundamentals – The chapter introduces artificial intelligence (AI) technology, its terminologies, and its significance across industries. It explains the concepts of machine learning and various ML techniques, such as supervised, unsupervised, semi-supervised and reinforced learning, highlighting their benefits and challenges. The chapter also focuses on the concept of design patterns, explaining its importance in AI applications. Chapter 2: GenAI Foundation – The chapter is to provides a foundation for generative AI technology that is deep neural network and its significance. It aims to explain the concepts of deep learning and various deep learning (DL) techniques such as deep neural network elements, gradient descent, backpropagation, hyper-parameters, performance metrics and highlighting their significance and options. The chapter includes coded examples for supervised deep learning regression and classification problems . Chapter 3: GenAI for Images – This chapter provides an overview of computer vision, CNN architectures, and image processing techniques. It covers the basics of image processing, computer vision tasks, real-world applications, and the evolution of deep learning architectures. Additionally, it explores representation learning, including autoencoders and variational autoencoders, for computer vision tasks. The chapter discusses encoder- decoder architectures, encoding and decoding mechanisms, and coded examples such as U-Net for image segmentation and VAE for image generation Chapter 4: Transforming Images with GenAI – The chapter provides an overview of generative modeling, covering different types of explicit and implicit modeling methods. It delves into generative adversarial networks (GANs), discussing their architecture, variations, and applications including example for VQGAN and CLIP. Diffusion processes and diffusion models are explained, including their architecture, training process, and the latent diffusion model for image processing (MNST example). Chapter 5: GenAI for Text – The chapter provides an overview of NLP technology, its key concepts, and its applications in various industries. It explains text pre-processing and post-processing techniques, and discusses different deep learning architectures for NLP tasks, including traditional models like RNN and LSTM, and more recent transformer architectures. The chapter focuses on different types of transformer architectures, their components in details, and their applications in tasks like summarization, question answering, and speech-to-text recognition. The chapter includes coded examples to fine tune GPT model for new tasks and Text summarization task. Chapter 6: ChatGPT – In this chapter, we will look into ChatGPT essentials, its features, integrations and addons along with image generation and speech to text examples. Further, we will understand the importance of prompts, their types, frameworks, and examples, and delve into the prompt engineering process. Finally, we will create a prompt Playground application for enterprises. Chapter 7: Large Language Model Frameworks – The objective of this chapter is to provide a overview of LangChain framework, its components, and examples. We will build three prototypes 1.) use llms to chat with your excel documents 2.) Retrieve contextual information using Retrieval Augmented generation (RAG) from customer knowledge base 3.) create a chat application that can query medical research journals and extract property graph out of it

📄 Page 8

Chapter 8: Large Language Model Operations – The objective of this chapter is to go through complete LLM system lifecycle, starting from data preparation, pre-training, benchmarking, experimentation, AI alignment, model experimentation, model serving, validation, security and monitoring. This chapter aims to explain the concepts of LLMOps, including various optimization techniques to improve the performance and efficiency of the LLM system as a whole. Chapter 9: Generative AI for Enterprise – The chapter provides an overview of how GenAI technology is leveraged by enterprises for their business improvements. This chapter covers various enterprise use cases, GenAI products and their capabilities. It also focusses on risks related to AI models, GenAI model vulnerabilities and factors that enterprise must consider while implementing GenAI projects. It also gives foundation to responsible AI, Its applicability in enterprise systems and the examples. Chapter 10: Advances and Sustainability in Generative AI – The chapter helps to understand the advancements happening in generative AI in terms of business, technology and regulation. We will also look into the types of business strategy that an enterprise must adopt to leverage GenAI’s potential to their competitive advantage. Further we will look into the environmental impact due to training & inferencing of GenAI models, way to measure and mitigate the same.

📄 Page 9

Code Bundle and Coloured Images Please follow the link to download the Code Bundle and the Coloured Images of the book: https://rebrand.ly/1fdc32 The code bundle for the book is also hosted on GitHub at https://github.com/bpbpublications/Generative-AI-for- Everyone. In case there’s an update to the code, it will be updated on the existing GitHub repository. We have code bundles from our rich catalogue of books and videos available at https://github.com/bpbpublications. Check them out! Errata We take immense pride in our work at BPB Publications and follow best practices to ensure the accuracy of our content to provide with an indulging reading experience to our subscribers. Our readers are our mirrors, and we use their inputs to reflect and improve upon human errors, if any, that may have occurred during the publishing processes involved. To let us maintain the quality and help us reach out to any readers who might be having difficulties due to any unforeseen errors, please write to us at : errata@bpbonline.com Your support, suggestions and feedbacks are highly appreciated by the BPB Publications’ Family. Did you know that BPB offers eBook versions of every book published, with PDF and ePub files available? You can upgrade to the eBook version at www.bpbonline.com and as a print book customer, you are entitled to a discount on the eBook copy. Get in touch with us at : business@bpbonline.com for more details. At www.bpbonline.com, you can also read a collection of free technical articles, sign up for a range of free newsletters, and receive exclusive discounts and offers on BPB books and eBooks. Piracy If you come across any illegal copies of our works in any form on the internet, we would be grateful if you would provide us with the location address or website name. Please contact us at business@bpbonline.com with a link to the material. If you are interested in becoming an author If there is a topic that you have expertise in, and you are interested in either writing or contributing to a book, please visit www.bpbonline.com. We have worked with thousands of developers and tech professionals, just like you, to help them share their insights with the global tech community. You can make a general application, apply for a specific hot topic that we are recruiting an author for, or submit your own idea. Reviews Please leave a review. Once you have read and used this book, why not leave a review on the site that you purchased it from? Potential readers can then see and use your unbiased opinion to make purchase decisions. We at BPB can understand what you think about our products, and our authors can see your feedback on their book. Thank you! For more information about BPB, please visit www.bpbonline.com. Join our book’s Discord space Join the book’s Discord Workspace for Latest updates, Offers, Tech happenings around the world, New Release and Sessions with the Authors: https://discord.bpbonline.com

📄 Page 10

Table of Contents 1. AI Fundamentals Introduction Structure Objectives Introduction to AI Defining AI Artificial Narrow Intelligence Artificial General Intelligence Artificial Super-intelligence Exploring Artificial Narrow Intelligence Machine learning Understanding different learning approaches in machine learning Supervised and unsupervised learning Supervised learning Supervised learning methods Unsupervised learning Unsupervised learning methods Semi-supervised learning and reinforcement learning Semi-supervised learning Semi-supervised algorithms Pseudo labeling methods Consistency regularization method Hybrid method Reinforcement learning Reinforcement learning components Algorithms Design patterns Real world data representation Encoding Embedding Image embedding Audio embedding Text embedding Training strategies Transfer learning Multi-task learning Distribution strategy Performance improvement strategies Conclusion Key terms

📄 Page 11

Questions 2. GenAI Foundation Introduction Structure Objectives Basic mathematics of artificial intelligence Scalar, vectors and matrices Linear and non-linear functions Differentiable function Derivatives, gradient and Jacobian matrix Chain rule Deep learning Single and multi-layer perceptron Multi-layer perceptron Activation function (σ) Choosing activation function Bias term Weight and bias initialization Cost function Binary and multi-class classification loss functions Cross-entropy and Kullback–Leibler divergence Gradient descent Gradient descent procedure Types of gradient descent algorithm Backpropagation Generalization L1/L2 regularization Dropouts Batch normalization Gradient centralization Learning rate Adaptive learning rates Model performance metrics Classification accuracy Deep learning guide Train and test split Model building Regression model training Regression model testing Artificial neural network: Classification problem Introduction to generative AI Generative artificial intelligence Foundation model Prompt engineering Impact on industries and societies Security, ethics and responsibilities

📄 Page 12

Sustainability Limitations Generative AI use cases Business/Enterprises Choosing the right use cases Conclusion Key terms Questions 3. GenAI for Images Introduction Structure Objectives Introduction to computer vision Computer vision architectures Convolution neural networks Convolution layer Padding Pooling layer Fully connected layer Examples: U-Net Image segmentation Representation learning Autoencoders and variational autoencoders Examples: Generate new faces VAE variants Conclusion Key terms Questions 4. Transforming Images with GenAI Introduction Structure Objectives Generative modeling Generative adversarial networks Example: Using GAN for improving image resolution Generative adversarial networks variations Example: Image generation from text Diffusion model Denoising diffusion probabilistic models The reverse process Model training Neural architecture for diffusion model Latent diffusion models Flow model Evaluation metrics for image generative model

📄 Page 13

Conclusion Key terms Questions 5. GenAI for Text Introduction Structure Objectives Natural language processing Natural language processing tasks Sequence to sequence models Recurrent neural network Long-short term memory networks Transformer architecture Types of architecture Transformer components Tokenization Token IDs Embeddings Positional embeddings Encoder layer Multi-head self-attention Layer normalization Residual connections Feedforward layer Decoder Cross attention Masked multi-head attention mechanism Language modelling head Decoder sampling techniques Top-k sampling Top-P sampling Temperature sampling Next sentence prediction task Text summarization task Question and answering task Machine translation task NLP examples Tune GPT-2 model for new tasks Encoder-decoder text summarization task Conclusion Key terms Questions 6. ChatGPT Introduction

📄 Page 14

Structure Objectives Introduction to OpenAI OpenAI image models OpenAI Whisper OpenAI Playground OpenAI Cookbook OpenAI fine-tuning Format ChatGPT plugins ChatGPT Enterprise Pricing Limitations of ChatGPT ChatGPT future Experience the ChatGPT add-on feature Prompt engineering Importance of prompt Prompt engineering process Setting objectives Prompt design Evaluation of responses Prompt refining Prompt templating Build your prompt playground Metrics Jailbreaking Conclusion Key terms Questions 7. Large Language Model Frameworks Introduction Structure Objectives Introduction to LangChain Chain Prompt template ChatPromptTemplate Memory Document loaders Splitters or chunking Vector stores Agents Prototype Chat with Excel files Retrieval augmented generation

📄 Page 15

Small-to-big retrieval Create vector embedding for the document chunks Retriever Retrieval query engine Property graph creation of medical research journal Conclusion Key terms Questions 8. Large Language Model Operations Introduction Structure Objectives Introduction to LLMOps LLMOps Phase 1 Data preparation Model training and fine-tuning Pre-training Data formats for generative AI training Data quality evaluation Synthetic data generation Memory optimization Parallel computation methods Performance benchmarking Evaluation metrics Benchmarking dataset Leaderboard LLM benchmarking tools and libraries AI alignment Continued pre-training Fine-tuning Low-rank adaptation Merging of model Experimentation RAG experimentation example LLMOps Phase 2 Model serving Strategies followed in deployment Types of inferencing setups Types of serving environments Model serving components needed in production environment Inference optimization Quantization Pruning Knowledge distillation Request batching Large language model security

📄 Page 16

Tools and libraries for vulnerability identification Monitoring and feedback Monitoring Metrics Drift detection Alerting and feedback systems LLM application tracing Other Deployment practices Hugging Face Conclusion Key terms Questions 9. Generative AI for the Enterprise Introduction Structure Objectives GenAI for enterprise Rise of GenAI Enterprise use cases of GenAI GenAI real-world examples GenAI vendors AI risks and challenges Main AI concerns GenAI vulnerabilities Enterprise level considerations Responsible AI The relevance of responsible AI Responsible AI principles Responsible AI examples Conclusion Key Terms Questions 10. Advances and Sustainability in Generative AI Introduction Structure Objectives Advances of GenAI Business Economic potential of GenAI Job impact Industries Technology Data selection Architectures beyond transformers Sophia optimization

📄 Page 17

Diffusion alignment: Direct preference optimization Federated learning Split learning Confidential AI Regulations New AI regulatory acts across regions Regulatory challenges GenAI business strategy Strategy for a new revenue stream Strategy for augmenting product and services Hyperbolic time cone Enterprise strategy using hyperbolic time cone GenAI sustainability Environmental impact due to GenAI Tools to measure the carbon intensity and total energy consumption Mitigations Conclusion Key terms Questions Index

📄 Page 18

CHAPTER 1 AI Fundamentals Introduction Generative AI (GenAI) has become a catalyst for any business. Though it does not seem to be disrupting industries immediately, the advances in GenAI is making it move from current state of collaborator to disruptor. AI has been a while since 1950’s and machine learning (ML) sub-domain of AI has been successful in many industry domains. So, understanding its evolution, techniques, and underlying architecture is a must. Before delving into GenAI, this chapter gives an introduction to AI; it is methods and design features. In upcoming chapters, we will detail GenAI concepts of deep learning, generative modeling techniques, various GenAI examples, ChatGPT essentials, and GenAI use cases, and conclude with enterprise risks and responsibility. Throughout the book, we will understand the AI concepts, specific techniques, mathematical briefings, and examples. Though we were not able to provide examples of all techniques, we have carefully chosen the ones that are needed to understand GenAI concepts holistically. Structure In this chapter, we will learn the following topics: • Introduction to AI • Supervised and unsupervised learning • Semi-supervised learning and reinforcement learning • Design patterns Objectives The objective of this chapter is to introduce artificial intelligence (AI) technology, its terminologies, and its significance across industries. It aims to explain the concepts of machine learning (ML) and various ML techniques, such as supervised and unsupervised learning, highlighting their benefits and challenges. The content also focuses on the advanced ML methods of semi-supervised and reinforcement learning (RL). Furthermore, the content introduces the concept of design patterns, explaining its importance in AI applications. Introduction to AI Welcome to the world of AI, a term that might sound complex but holds a simple and very interesting idea at its core. AI is all about infusing intelligence into things that are not alive, making them smart in a way that even surpasses human intelligence. This concept envisions achieving super-intelligence, where non-living entities become smarter than the brightest human minds. Now, you might wonder, should we be worried? The answer is no. History has shown that with every technological leap, there has been disruption, but it has been a disruption for the better. Think about the transition from calculators to computers or from steam engines to electric trains—each change has transformed our lives, opening new opportunities and enhancing our lifestyles. What exactly is AI? Why do we believe AI is necessary? Are there any challenges or concerns hidden in the AI landscape? In this chapter, we will delve into all these questions, exploring the depths of AI and uncovering the

📄 Page 19

mechanisms that power it. We will navigate through the various categories of AI, observing its remarkable progress. You might be surprised to learn that AI has already found its way into our lives, subtly weaving its benefits into our routines. You might be using AI and its applications without even realizing it. Ignorance causes fear, but knowledge and awareness help us to be more creative by utilizing the powerful tool. Acquiring knowledge about this tool and understanding how to hold and use it properly is crucial, especially when the tool possesses intelligence. Extreme caution is required. As we learn more about AI, we must approach it with responsibility, guided by appropriate governance, policies, and guardrails. Moving on, let us explore a specific example that is familiar to many: self-driving cars. The reason why we do not see self-driving cars dominating the roads as expected is not just about technology; it is about regulations, rules, and the allocation of responsibility in case of mishaps. From these scenarios, we draw a key message: leveraging AI tools and technologies demands not only the right resources but also a deep understanding of their application and a well-crafted governance plan. This combination ensures that we can fully harness AI’s potential, reaping its benefits while minimizing risks. Now, let us further understand the world of AI. Defining AI Imagine teaching computers to be smart, just like we teach our friends. AI is like giving computers brains to think, learn, solve problems, and talk like humans do. It is like having a new friend who is good at understanding and helping us. AI is already reshaping various fields like education, research, healthcare, and transportation. However, the recent goal is to develop artificial general intelligence (AGI), which would empower computers to perform any cognitive task that a human can. This concept might sound like science fiction, but it is becoming a reality. AI’s primary aim is to enhance computer intelligence significantly, making them exceptionally intelligent and helpful. In the future, these machines could surpass human intelligence, a state known as super-intelligence. This journey in AI is all about creating incredibly advanced and friendly technology. Take a look at the following figure: Figure 1.1: Current and future of AI Artificial Narrow Intelligence Just like the name suggests, Narrow AI has a limited scope. It is designed for a particular job within a specific field or area. Think of it as a specialist AI that is trained for a particular industry or situation, and it cannot handle tasks beyond its specialized training. That is why it is also called Weak AI. For instance, imagine a chatbot working as a customer service representative for a bank. This chatbot is only trained to handle questions related to that specific bank’s customer service. It would not be able to chat about anything else

📄 Page 20

or perform tasks outside of banking inquiries. Similarly, let us consider a predictive maintenance model used to prevent failures in a system. However, this model is limited to predicting issues within the specific system it is designed for. So, if it is created for a car engine, it would not be able to predict problems in an entirely different type of machinery. In a nutshell, Narrow AI or Weak AI is like a highly focused expert—it is great at what it is trained for, but it cannot handle tasks that fall outside its narrow area of expertise. Artificial General Intelligence Strong AI’s purpose is to build human-like intelligence to think and give general answers like a human. Strong AI, or AGI, aims to create intelligence in machines that is like human thinking. Imagine a computer that can understand and respond to things just like a human can. An example could be the future multi-modal Generative Pre-trained Transformer (GPT). The goal, arguably, is to have super-intelligence. Artificial Super-intelligence In AI, the idea of super-intelligence sparks varied opinions due to different levels of understanding. Some believe that achieving super-intelligence is both possible and beneficial, while others hold the view that it is neither attainable nor desirable. In his enlightening book SUPERINTELLIGENCE, Nick Bostrom puts forth a definition for super-intelligence. He describes it as any intellect that substantially exceeds the cognitive performance of humans in practically all categories of interest. Bostrom categorizes super-intelligence into three types, each with its own unique characteristics, as follows: • Speed super-intelligence: A system that can do all that a human intellect can do, but much faster. • Quality super-intelligence: A system that is at least as fast as a human mind and vastly qualitatively smarter. • Collective super-intelligence: A system composed of a large number of smaller intellects such that the system’s overall performance across many very general domains vastly outscripts that of any current cognitive system. Exploring Artificial Narrow Intelligence Well, AI has different ways to work its magic, but two of the coolest ANI ones are machine learning (ML) and deep learning. These are like training programs for computers. You know how we learn from books, experiences, and practice? Computers learn from lots of information too, just like we teach them. Deep learning uses something special called neural networks (NN). These networks are like the computer’s brain cells that help them learn from a ton of information. Then, using what they have learned, computers can make clever choices, almost like how you make smart decisions after learning from your experiences. Take a look at the following figure that showcases the artificial narrow intelligence sub-domains: