Generative Artificial Intelligence: Exploring the Power and Potential of Generative AI

This book explains the field of Generative Artificial Intelligence (AI), focusing on its potential and applications, and aims to provide you with an understanding of the underlying principles, techniques, and practical use cases of Generative AI models. The book begins with an introduction to the foundations of Generative AI, including an overview of the field, its evolution, and its significance in today’s AI landscape. It focuses on generative visual models, exploring the exciting field of transforming text into images and videos. A chapter covering text-to-video generation provides insights into synthesizing videos from textual descriptions, opening up new possibilities for creative content generation. A chapter covers generative audio models and prompt-to-audio synthesis using Text-to-Speech (TTS) techniques. Then the book switch gears to dive into generative text models, exploring the concepts of Large Language Models (LLMs), natural language generation (NLG), fine-tuning, prompt tuning, and reinforcement learning. The book explores techniques for fixing LLMs and making them grounded and indestructible, along with practical applications in enterprise-grade applications such as question answering, summarization, and knowledge-based generation. By the end of this book, you will understand Generative text, and audio and visual models, and have the knowledge and tools necessary to harness the creative and transformative capabilities of Generative AI. What You Will Learn What is Generative Artificial Intelligence? What are text-to-image synthesis techniques and conditional image generation? What is prompt-to-audio synthesis using Text-to-Speech (TTS) techniques? What are text-to-video models and how do you tune them? What are large language models, and how do you tune them? Table of Contents About the Authors About the Technical Reviewer Introduction Chapter 1: Introduction to Generative AI Unveiling the Magic of Generative AI The Genesis of Generative AI Milestones Along the Way Fundamentals of Generative Models Neural Networks: The Backbone of Generative AI Key Neural Network Architectures Relevant to Generative AI Convolutional Neural Networks Recurrent Neural Networks Generative Adversarial Networks Transformers Understanding the Difference: Generative vs. Discriminative Models Understanding the Core: Types and Techniques Diffusion Models Generative Adversarial Networks Variational Autoencoders Restricted Boltzmann Machines Pixel Recurrent Neural Networks Generative Models in Society and Technology Real-World Applications and Advantages of Generative AI Ethical and Technical Challenges of Generative AI DeepMind’s Approach to Data Privacy and Security Impact of Generative Models in Data Science The Diverse Domains of Generative AI Visuals: From Pixel to Palette Audio: Symphonies of AI Text: Weaving Words into Worlds The Future of Generative AI: A Symphony of Possibilities Setting Up the Development Environment Setting Up a Google Colab Environment Hugging Face Access and Token Key Generation OpenAI Access Account and Token Key Generation Troubleshooting Common Issues Summary Chapter 2: Text-to-Image Generation Introduction Bridging the Gap Between Text and Image Data Understanding the Fundamentals of Image Data Correlation Between Image and Text Data Using CLIP Model Architecture and Functioning CLIP Case Study Implementation of CLIP Step 1: Installing Libraries and Data Loading Step 2: Data Preprocessing Step 3: Model Inference Diffusion Model Implement Diffusion Model from Scratch Step 1: Installing Libraries Step 2: Data Preprocessing Step 3: Model Training Text-to-Image Generation Using a Pre-trained Model Step 1: Installing Libraries Step 2: Model Inference Fine-Tuning Text-to-Image Models Step 1: Installing Libraries and Data Loading Step 2: Model Training Step 3: Model Inference Common Challenges and Troubleshooting Tips Conclusion Chapter 3: From Script to Screen: Unveiling Text-to-Video Generation Introduction Understanding Video Data Challenges in Working with Video Data The Synergy of Video and Textual Data Tagging Videos with Semantic Metadata Hands-On: Demonstrating a Pre-Trained Model Step 1: Installing Libraries Step 2: Model Inference Fine-Tuning for Custom Applications Step 1: Installing Libraries Step 2: Data Loading and Preprocessing Step 3: Model Training (Fine-Tuning) Step 4: Model Inference Conclusion Chapter 4: Bridging Text and Audio in Generative AI Brief History Fundamentals and Challenges Understanding Audio Data Challenges in Working with Audio Data Mitigating Challenges in Audio Data Processing Bridging Text and Audio: The CLAP Model Implementation Step 1: Installing Libraries and Data Loading Step 2: Model Inference Understanding AI-Driven Text and Audio Conversion Models Understanding CTC Architectures Understanding Seq2Seq Architectures Implementation AI-Driven Text and Audio Conversion Models Speech to Text Using Pre-trained Model for STT Step 1: Installing Libraries and Data Loading Step 2: Model Inference Step 3: Model Evaluation Fine-Tuning a Speech-To-Text Model for a Specific Language Step 1: Installing Libraries and Data Loading Step 2: Data Preprocessing Step 3: Model Training (Fine-Tuning) Step 4: Model Evaluation and Inference Text to Speech Using a Pre-trained Model for TTS Step 1: Installing Libraries and Data Loading Step 2: Data Processing Step 3: Model Inference Fine-Tuning Text-To-Speech Model Step 1: Installing Libraries and Data Loading Step 2: Data Preprocessing Step 3: Model Training (Fine-Tuning) Step 4: Model Inference Troubleshooting Common Issues Conclusion Chapter 5: Large Language Models Introduction Phases of Training and Adoption of Large Language Models Pre-training of LLMs: Fine-Tuning and Adaptation of LLMs Why Large Language Models? Types of Language Transformers Models Encoder Models BERT Fine-Tuning BERT Fine-Tuning for Sentence Classification Step 1: Installing Libraries and Data Loading Step 2: Data Preprocessing Step 3: Model Training Step 4: Model Evaluation Conclusion Fine-Tuning for Named Entity Recognition Step 1. Installing Libraries and Data Loading Step 2: Data Pre-processing Step 3: Model Training Step 4: Model Evaluation Conclusion Decoder-Only Models (Generative Pre-trained Transformer) Autoregressive Modeling: The Technical Backbone The Mechanics of Sequence Generation Real-World Implications and Technical Applications The Road Ahead: Challenges and Opportunities Encoder-Decoder Models The Genesis and Evolution The Breakthrough: Transformer Models Real-World Applications Training and Fine-Tuning Challenges and Future Directions A Glimpse into the LLM Horizon: Where Do We Go from Here? Summary Chapter 6: Generative Large Language Models Introduction NLP Tasks Using LLMs Sentiment Analysis Best Practices for Prompt Engineering in Sentiment Analysis Entity Extraction Limitations of LLMs in Entity Extraction Topic Modeling Comparative Insights: LLM-Based Topic Modeling vs. Traditional Method Natural Language Generation Tasks Using LLMs Creative Writing Fine-Tuning on Creative Writing Tasks for Genre and Style Text Summarization Navigating the Challenges in LLM-Generated Summaries Dialogue Generation Integrating LLM-Generated Dialogues into Virtual Assistants Technical Integration User Experience Considerations Practical Deployment Advanced Prompting Techniques Few-Shot Prompting Industrial Applications of Few-Shot Prompting Chain-of-Thought Enhancing Interpretability with Chain-of-Thought Prompting Application in Complex Reasoning Advantages of Interpretability Prompting vs. Fine-Tuning Weighing the Trade-Offs: Prompting vs. Fine-Tuning Prompt Engineering Fine-Tuning Example Application Fine-Tuning LLMs Case Study: Fine-Tuning an LLM for Sentiment Analysis Fine-Tuning Steps Common Pitfalls and How to Overcome Them Parameter Efficient Fine-Tuning Fine-Tuning LLM for Question Answering Step 1: Setting Up the Development Environment Step 2: Data Preprocessing Step 2.1: Load the Dataset Step 2.2: Train-Test Split Step 2.3: Transform the Dataset Step 2.4: Tokenize the Dataset Step 2.5: Preprocess the Dataset Step 3: Model Training/Fine-Tuning Step 3.1: Load the Model Step 3.2: Prepare the Model Step 3.3: Create a Data Collator Step 3.4: Define Training Hyperparameters Step 3.5: Training the Model Step 3.6: Saving the Model Step 4: Model Evaluation Step 4.1: Load the Fine-Tuned Model Step 4.2: Test the Fine-Tuned Model Step 4.3: Evaluate the Fine-Tuned Model on the Test Dataset Step 4.4: Create an Evaluation DataFrame Summary Chapter 7: Advanced Techniques for Large Language Models Introduction Fine-Tuning LLMs for Abstractive Summarization Fine-Tuning an Encoder-Decoder Model Step 1: Installing Libraries and Data Loading Step 2: Data Pre-processing Step 2.1: Evaluation and Benchmark Setup Step 3: Model Training Step 4: Model Evaluation Abstractive Summarization Using a Decoder-Only Model Step 1: Installing Libraries and Data Loading Step 2: Data Pre-processing Step 3: Model Training Step 4: Model Inference and Evaluation Guidelines on Fine-Tuning a Large Language Model Types of SFT (Supervised Fine-Tuning) Memory Consumption During SFT Reinforcement Learning from Human Feedback What Is RLHF? How Does RLHF Work? Reward Model Implementation Controlled Review Generation Setting Up the Environment Import Dependencies Configure Environment Loading the Dataset Loading the Models Load the Pre-trained GPT-2 Models Initialize PPOTrainer Load BERT Classifier Text Generation Settings Optimizing the Model for Positive Review Generation Training Model Evaluation Save Model Evaluating and Ensuring the Fairness of the Reward Model in RLHF RLHF Summary Summary Chapter 8: Building Demo Applications Using LLMs Making Sense of Website Content Data Scraping Question-answering Summarization User Interface/Application Uncovering Insights and Gaining a Quick Understanding of PDF Documents Question-Answering for PDF PDF Summarization Extracting Insights from Video Transcripts Video Caption Summarization and Q&A Video Transcript Analysis Using Langchain and OpenAPI Summary Chapter 9: Building Enterprise-Grade Applications Using LLMs Retrieval-Augmented Question-Answering Chatbot Real-World Use Cases of Retrieval Augmentation Generation RAG Architecture Creating a Knowledge Base Step 1: Collecting Data Step 2: Setting Up a Retriever Step 2.1: Downloading a Solr Binary File Step 2.2: Install Solr Step 2.3: Start the Solr Cloud Service Locally Step 2.4: Index the Documents Scaling Up Indexing for Enterprise-Grade Knowledge Bases Setting Up a Retrieval System Block 1: URL Construction and Request Block 2: Document Information Extraction Block 3: Search Result Processing Block 4: Results Display Block 5: Output and Return Neural Reranker DrDecr Reranker Generative LLM Mitigating Hallucination Risks in In-Context Learning User Interface Suggested Improvements in the RAG Pipeline for Generative Q&A Summary Conclusion: Generative AI Journey References Index