LLMs Onboarding Guide

The following resources will thoroughly introduce large language models (LLMs) and how to use them. You can either pick one of the resources listed in All-In-One Resources or follow chapters 1 to 6 in order. Chapter 7 offers more context on generative AI in general.

The essential content for each chapter is marked with a bold font. Additional resources offer a more in-depth view of the topic.

I'd encourage you to come back to Chip Huyen's blog post Building LLM applications for production from time to time. The article covers a wide range of topics, and as you learn more about LLMs, you will be able to explore these topics further.

Table of Contents

0. All-In-One Resources
1. High-Level Overview
2. Deep-Dive into Language and Transformers
3. Training LLMs
4. Fine-Tuning LLMs
5. Using LLMs
6. Building Applications with LLMs
7. More on Generative AI

0. All-In-One Resources

Gain a comprehensive insight into LLMs and their applications. These resources offer both theoretical knowledge and practical applications, enabling you to explore and utilize large language models effectively.

• The ultimate LLM Learning Mind Map
• Hands-on course, explaining Natural Language Processing, Language Models and how to bring those together with code: Hugging Face NLP Course
• Intro to LLMs and building applications with LangChain and VectorDBs: LangChain & Vector Databases in Production - 40 hours course

1. High-Level Overview

Embark on your journey into the world of large language models, understanding their basic functioning, potential and use cases.

• How can we use traditional ML for language generation: How do LLMs work? Next Word Prediction with the Transformer Architecture Explained
• A mellow introduction to language models and generative AI: How GPT models work
• LLMs from a practical perspective: Building LLM applications for production

2. Deep-Dive into Language and Transformers

Explore the intricacies of the transformer architecture, understanding key concepts like attention mechanisms, position embeddings, and the foundational principles that enable LLMs to understand and generate human-like text.

• Practical introduction to transformers: How do transformers work?
• How language models make sense of language: What are Word and Sentence Embeddings?
• Video series giving a nice starting point to understand language and transformers: Visual Guide to Transformers
  • Part 0: The Neuroscience of “Attention”
  • Part 1: Position Embeddings
  • Part 2: Multi-Head & Self-Attention
  • Part 3: Decoder’s Masked Attention
• Illustrated guides to Attention and Transformers (also available as video)
• Technical deep-dive into the underlying transformer architecture: Transformers from Scratch
• The OG paper: Attention is all you need

3. Training LLMs

Learn how language models trained and how you can train your own model.

• Step-by-step guides to train a language model from scratch with the Hugging Face transformers library:
  • How to train a new language model from scratch using Transformers and Tokenizers
  • Training a language model with 🤗 Transformers using TensorFlow and TPUs
• Putting the golden layer on a language model and make for a human like chat experience: RLHF: Reinforcement Learning from Human Feedback
• More practical explanation of RLHF: Illustrating Reinforcement Learning from Human Feedback (RLHF)

4. Fine-Tuning LLMs

Unlock the true potential of pre-trained LLMs by learning about fine-tuning processes, exploring various strategies to adapt existing models to your specific use-cases and understanding how to incorporate domain-specific knowledge.

• Introduction to fine-tuning: LLMs Unleashed: The Power of Fine-Tuning
• Incorportate domain knowledge without fine-tuning: Retrieval Augmented Generation
• When to apply fine-tuning, how to prepare the data and how to execute it: Fine-tuning LLMs, Deeplerning.AI - 1 hour course
• Fine-tune without retraining and little data: SetFit: Efficient Few-Shot Learning Without Prompts
• Examples of fine-tuning models:
  • LLaMa 2 model, including code, data, and hardware requirements
  • LLaMa model with reinforcement Learning with humand feedback (RLHF)

5. Using LLMs

Understand how to effectively communicate with LLMs, learning about prompt engineering and discovering strategies to extract desired outputs and answers from models.

• Interact with a model and get the answer you need: Prompt Engineering Guide
• Learn how to chain prompts effectively: Building Systems with the ChatGPT API, Deeplerning.AI - 1 hour course
• Introduction to Prompt Engineering, Deeplerning.AI - 1 hour course

6. Building Applications with LLMs

Learn to develop applications utilizing LLMs to transition from theoretical understanding to actual implementations.

• Learn the basics of LangChain and how to use them: LangChain for LLM Application Development, Deeplerning.AI - 1 hour course
• Understanding VectorDBs: Explaining Vector Databases in 3 Levels of Difficulty
• In-depth course on VectorDBs: Master Vector Database with Python for AI & LLM Use Cases - 7 hour course

7. More on Generative AI

Explore the broader sphere of generative AI, learning about its historical development, underlying technologies like Generative Adversarial Networks (GANs), and gaining insights into the broader implications and applications of generative models in the AI field.

• OpenAI giving an intro to GenAI back in 2016: Generative Models
• The technique powering good generative output: Generative Adversarial Nets (GANs)
• That same technique explained in an understandable way: A Gentle Introduction to Generative Adversarial Networks (GANs)