Explore the Transformer architecture, the foundation of state-of-the-art AI/ML models like BERT and GPT, in this 30-minute visual presentation. Delve into the components of Transformer language models, including feed-forward neural networks and self-attention mechanisms. Learn about tokenization, embedding, and output projection processes. Gain insights into model training and probability visualization. Suitable for viewers with various levels of machine learning experience, this accessible video provides a comprehensive overview of the Transformer model's structure and applications in natural language processing.
Overview
Syllabus
Intro
The Architecture of the Transformer
Model Training
Transformer LM Component 1: FFNN
Transformer LM Component 2: Self-Attention
Tokenization: Words to Token Ids
Embedding: Breathe meaning into tokens
Projecting the Output: Turning Computation into Language
Final Note: Visualizing Probabilities
Taught by
Jay Alammar
