Learn how generative AI works by building your very own models that can write coherent text, create realistic images, and even make lifelike music.
Learn Generative AI with PyTorch teaches the underlying mechanics of generative AI by building working AI models from scratch. Throughout, youâll use the intuitive PyTorch framework thatâs instantly familiar to anyone whoâs worked with Python data tools. Along the way, youâll master the fundamentals of General Adversarial Networks (GANs), Transformers, Large Language Models (LLMs), variational autoencoders, diffusion models, LangChain, and more!
In Learn Generative AI with PyTorch youâll build these amazing models:
âą A simple English-to-French translator
âą A text-generating model as powerful as GPT-2
âą A diffusion model that produces realistic flower images
âą Music generators using GANs and Transformers
âą An image style transfer model
âą A zero-shot know-it-all agent
The generative AI projects you create use the same underlying techniques and technologies as full-scale models like GPT-4 and Stable Diffusion. You donât need to be a machine learning expertâyou can get started with just some basic Python programming skills.
About the technology
Transformers, Generative Adversarial Networks (GANs), diffusion models, LLMs, and other powerful deep learning patterns have radically changed the way we manipulate text, images, and sound. Generative AI may seem like magic at first, but with a little Python, the PyTorch framework, and some practice, you can build interesting and useful models that will train and run on your laptop. This book shows you how.
About the book
Learn Generative AI with PyTorch introduces the underlying mechanics of generative AI by helping you build your own working AI models. Youâll begin by creating simple images using a GAN, and then progress to writing a language translation transformer line-by-line. As you work through the fun and fascinating projects, youâll train models to create anime images, write like Hemingway, make music like Mozart, and more. You just need Python and a few machine learning basics to get started. Youâll learn the rest as you go!
What's inside
âą Build an English-to-French translator
âą Create a text-generation LLM
âą Train a diffusion model to produce high-resolution images
âą Music generators using GANs and Transformers
About the reader
Examples use simple Python. No deep learning experience required.
About the author
Mark Liu is the founding director of the Master of Science in Finance program at the University of Kentucky.
The technical editor on this book was Emmanuel Maggiori.
Table of Contents
Part 1
1 What is generative AI and why PyTorch?
2 Deep learning with PyTorch
3 Generative adversarial networks: Shape and number generation
Part 2
4 Image generation with generative adversarial networks
5 Selecting characteristics in generated images
6 CycleGAN: Converting blond hair to black hair
7 Image generation with variational autoencoders
Part 3
8 Text generation with recurrent neural networks
9 A line-by-line implementation of attention and Transformer
10 Training a Transformer to translate English to French
11 Building a generative pretrained Transformer from scratch
12 Training a Transformer to generate text
Part 4
13 Music generation with MuseGAN
14 Building and training a music Transformer
15 Diffusion models and text-to-image Transformers
16 Pretrained large language models and the LangChain library
Appendixes
A Installing Python, Jupyter Notebook, and PyTorch
B Minimally qualified readers and deep learning basics