Overview

Learn the nuts and bolts of LLMs and the revolutionary transformer architecture they are based on!

Large Language Models (LLMs) have become pivotal tools driving some of the most stunning advancements and applications in today's AI landscape. This hands-on course will equip you with the practical knowledge and skills needed to understand, build, and harness the power of LLMs for solving complex language tasks such as translation, language generation, question-answering, and more. Through iteractive coding exercises, you will learn to work with transformers, the most popular deep-learning architecture for constructing LLMs and NLP systems. You'll also explore pre-trained language models and datasets from Hugging Face: a vast repository of resources for seamlessly integrating LLMs into your projects. By the end of this course, you will be able to build LLMs using various transformer architectures and configure, fine-tune, and evaluate pre-trained LLMs using specialized metrics. You will also gain insights into advanced concepts like Reinforcement Learning from Human Feedback (RLHF) and understand the key challenges and ethical considerations of enabling real-world LLM applications.

Syllabus

The Large Language Models (LLMs) Landscape

Large Language Models (LLMs) represent the current pinnacle of AI technology, driving remarkable advancements in Natural Language Processing and Understanding. This chapter serves as your gateway to comprehending LLMs: what they are, their remarkable capabilities, and the wide array of language tasks they excel at. You'll gain practical experience in loading and harnessing various LLMs for both language understanding and generation tasks. Along the way, you'll be introduced to the successful catalyst at the heart of most LLMs: the transformer architecture. Ready to start this journey into the world of LLMs?

Building a Transformer Architecture

In this chapter, you'll uncover the secrets and practical intricacies of transformers, the most popular deep learning architecture used to create today's most successful Language Models. Step by step, and aided by the PyTorch library, you'll learn how to manually design and configure different types of transformer architectures. You'll develop a strong understanding of their core elements, including self-attention mechanisms, encoder and decoder layers, and specialized model heads designed for specific language tasks and use cases.

Harnessing Pre-trained LLMs

This chapter unveils the transformative potential of harnessing pre-trained Large Language Models (LLMs). Throughout the chapter, you'll discover effective tips and tricks for mastering intricate language use cases and gain practical insights into leveraging pre-trained LLMs and datasets from Hugging Face. Along the way, you will learn the ins and outs of several common language problems, including sentiment classification to summarization to question-answering, and explore how LLMs are adaptively trained to solve them.

Evaluating and Leveraging LLMs in the Real World

Our exciting LLMs learning journey is approaching its end! You'll delve into different metrics and methods to assess how well your model is performing, whether it's a pre-trained one, a fine-tuned version, or something you've built from the ground up! You'll learn about the crucial aspects and challenges of applying Language Models in real-world scenarios, including optimizing a model with feedback from humans (RLHF), tackling biased language outputs, and more.