Transformers & LLMs
Build a real understanding of the transformer architecture, then train and fine-tune LLMs — theory and practice.
Chapter 01
The Transformer Architecture
- The Big Picture: What a Transformer DoesComing soon
- Tokenization: Text → IntegersComing soon
- Embeddings: Tokens → VectorsComing soon
- Positional Information: Order MattersComing soon
- Attention I: Query, Key, ValueComing soon
- Attention II: Scale, Mask, Softmax → WeightsComing soon
- Multi-Head AttentionComing soon
- The Feed-Forward Network (MLP)Coming soon
- Residual Connections & NormalizationComing soon
- The Full Transformer BlockComing soon
- The Output Head: Logits → Next TokenComing soon
- Putting It Together: A Forward Pass, End to EndComing soon
- Chapter examComing soon
Chapter 02
Building & Training a Small LLM
- Next-Token Prediction as ClassificationComing soon
- The Data Pipeline: Text → BatchesComing soon
- Backprop Through the TransformerComing soon
- The Optimizer: AdamW in PracticeComing soon
- Initialization & Numerical StabilityComing soon
- The Training LoopComing soon
- Watching a Model LearnComing soon
- Sampling & Generation RevisitedComing soon
- Scaling Laws: Predicting Loss Before You TrainComing soon
- Capstone: Your Tiny LLMComing soon
- Chapter examComing soon