llm2024

Winter 2024 course on LLMs at University of Washington, Seattle

View the Project on GitHub bytesizeml/llm2024

LLMs: From Transformers to GPT| Winter 2024

Professional Masters Program | University of Washington, Seattle

Instructor - Dr. Karthik Mohan

Home Lectures Assessments References Demos About Team

Pre-requisites

The course assumes that you have the basics of machine learning and are comfortable coding in Python. If this is not the case, please reach out to the TAs or the instructor asap. Set up is also key for success in the course. Assignment 1 partly takes care of the setup (coding environment, APIs, etc).

Flavor of the Course

The course will be a modern introduction to deep learning, transformers, GPT and beyond. We will have a mix of concepts, examples, applications in the industry, theory, algorithms, code and demos. Expect a lot of hands-on coding assignments and mini-projects. Towards the later half of the course, we will start to touch on the latest trends in Generative AI.

Lecture Dates

  1. Tuesday, 4-6 pm In-person
  2. Thursday, 4-6 pm In-person

Course Syllabus (we may leave out some topics depending on the available time)

  1. Introduction and Motivation for LLMs
  2. When even DL became a thing of the past! Deep Learning and its evolution
  3. What started the AI transformation? Transformer Architecture
  4. How do you concisely express data? Embeddings
  5. How does one make search smarter? Similarity Search with Transformers and Embeddings
  6. Discriminative vs Generative Transformers
  7. It’s just the stream of consciousness! Purely Generative Tranformers: GPT, GPT-2, GPT-3
  8. What made ChatGPT so popular? Fine-tuning and RLHF: GPT-3.5 and GPT-4
  9. LLMs vs APIs
  10. If only you had prompted me! Prompt Engineering
  11. Use of APIs
  12. To open or not to open? Closed vs Open-source LLMs
  13. Privacy, cost and other issues? Open-source LLMs: LLama, MixTral, Phi-1.5 and Phi-2
  14. Fine-tuning LLMs
  15. Can you make my fine-tuning easy for me? Tricks to fine-tune LLMs
  16. I can only handle smaller models!! Distillation and its use-cases
  17. What do I do if I don’t have enough data :-/ LLMs for Data Augmentation
  18. When LLM becomes your annotator Using LLMs to label data and train smaller models
  19. How can you trust an LLM? Evaluating LLMs
  20. Show me the cool stuff!: Question-Answering, Sentiment Analysis and more
  21. Showcasing LLMs over web demos: Use of StreamLit to build web-apps with innovative uses of LLMs and smaller models
  22. Is my data safe? Privacy and building in-house LLMs with privacy constraints
  23. LLMs -> LVMs: Moving from Language to Images and Videos
  24. How the heck can you generate an image from just text? Stable Diffusion
  25. Can you remove the photo-bomber from the pic? In-painting using image segmentation and stable diffusion