TransformerFromScratch Tutorial Series

Binxu Wang ([email protected])

April. 17th, 2023

Tutorial website

Lecture slides (PDF)

Transformer Jupyter Notebook Tutorial Series

• Fundamentals
  • Understanding Attention & Transformer (no GPU required)
    • Tutorial on Einstein summation rules
  • Language modelling with transformer (CPU or GPU)
• Beyond Language
  • All the following notebooks include training transformer, shall be run with GPU runtime or the training takes too long.
  • Learn to do arithmetics by sequence modelling. (Simple, GPU Training ~ 10 min)
  • Image generation by sequence modelling. (Simple, GPU Training ~ 10 ~ 20 min)
  • ~~Audio signal classification (Medium, GPU Training ~ 20 min)~~ (WARNING: currently, there is a dependency install error, don't run on Colab.)
  • Image classification (Medium, GPU Training ~ 30 min)
  • Music generation by sequence modelling. (Difficult, GPU Training takes hrs)
• Large Language Model
  • OpenAI API and Chat with PDF (Simple, no GPU needed, ~5mins)