In Generative AI Using Large Language Models (LLMs), you'll learn the fundamentals of how generative AI works, and how it can be deployed in real-world applications.
By taking this course, you will learn to:
- In-depth understanding of generative AI, describing the key steps in a typical LLM-based generative AI life cycle, from data collection and model selection, to performance evaluation and deployment
- Describe in detail the transformer architectures that power LLMs, how they are trained, and how fine-tuning enables LLMs to adapt to various specific use cases
- Use empirical scaling laws to optimize a model's objective function across dataset size, computational budget, and inference requirements
- Apply state-of-the-art training, tuning, inference, tooling, and deployment methods to maximize model performance within project-specific constraints
- Discuss the challenges and opportunities that generative AI brings to business after hearing stories from industry researchers and practitioners
Developers who have a good foundational understanding of how LLMs work, as well as the best practices behind training and deploying them, will be able to make informed decisions for their companies and build working prototypes faster. This course will help learners develop practical intuition on how to best utilize this exciting new technology.
This is an intermediate course, so you should have some experience coding in Python to get the most out of it. You should also be familiar with the fundamentals of machine learning, such as supervised and unsupervised learning, loss functions, and splitting data into training, validation, and test sets. If you've taken DeepLearning.AI's Machine Learning Specialization or Deep Learning Specialization, you'll be ready to take this course and dive into the fundamentals of generative AI.
We discuss large language models, their use cases, how models work, hint engineering, how to generate creative text output, and outline a project lifecycle for generative AI projects.
Given your interest in this course, it is safe to say that you have already had the opportunity to try out a generative AI tool or wish to do so. Whether it's chatbots,
generating images from text,
or use plugins to help you develop your code,
What you see in these tools is a machine capable of creating content that mimics or approaches human capabilities.
Generative AI is a subset of traditional machine learning. The machine learning models underpinning generative AI learn these abilities by finding statistical patterns in vast content datasets originally generated by humans. Large language models are trained on trillions of words over weeks and months and use a lot of computing power. These models, which we call fundamental models, have billions of parameters and exhibit emergent properties beyond language itself, and researchers are unlocking their ability to decompose complex tasks, reason and solve problems.
Here is a collection of base models, sometimes called base models, and their relative sizes in terms of parameters. You'll look at these parameters in more detail later, but for now, consider them the model's memory. The more parameters a model has, the more memory it has and, it turns out, the more complex tasks it can perform. Throughout the course, we will use these purple circles to represent LLMs, and in the lab, you will use a specific open source model, flan-T5, to perform language tasks. By using these models directly or applying fine-tuning techniques to adapt them to your specific use case, you can rapidly build custom solutions without training new models from scratch.
Now, while generative AI models are created for a variety of modalities, including image, video, audio, and speech, in this course you will focus on large-scale language models and their use in natural language generation. You'll learn how they're built and trained, and how to interact with them through text, known as prompts. And how to fine-tune models for your use case and data, and how to deploy them with applications to solve your business and social tasks.
The way you interact with language models is very different from other machine learning and programming paradigms. In those cases, you use a formalized syntax to write computer code to interact with libraries and APIs. In contrast, large language models are able to take instructions in natural language or human-written and perform tasks like a human. The text you pass to LLM is called a prompt. The space or memory available for hints is called the context window, which is usually enough for a few thousand words, but varies from model to model.
In this example, you ask the model to determine Ganymede's location in the solar system. The prompt is passed to the model, which then predicts the next word, and since your prompt contains a question, the model generates an answer. The output of the model is called completion, and the act of using the model to generate text is called inference. Completion includes the text contained in the original prompt followed by the generated text. You can see that this model answers your question very well. It correctly identified Ganymede as a moon of Jupiter and generated a plausible answer to your question stating that the moon is within Jupiter's orbit.
Throughout the course, you'll see many hints and finished examples of this style.
reference
- https://www.coursera.org/programs/hsbc-finance-people-and-personal-development-dnger/learn/generative-ai-with-llms
- https://www.coursera.org/learn/generative-ai-with-llms/lecture/IrsEw/generative-ai-llms