A free, interactive course
LLM Fundamentals
How a large language model is actually built — from a string of text to an aligned assistant — explained in three modules, every concept paired with a visualization you can poke. No frameworks, no maths wall, no hand-waving.
MODULE 01
Architecture
9 topics
- Introduction
- Tokenization
- The Embedding Layer
- Positional Encoding
- Attention
- Layers of Understanding
- Learning to Predict
- Instruction Tuning & RLHF
- Case Study — GPT-2 from Scratch
MODULE 02
Pre-Training
8 topics
- Overview
- Training Objectives & Details
- Scaling Laws & Optimization
- Training Data Engineering
- Infrastructure & Systems
- Advanced Pretraining Objectives
- Evaluation During Pretraining
- Case Study — Llama 3
MODULE 03
Post-Training
5 topics
- Overview
- Supervised Fine-Tuning
- Preference Optimization
- Tools & Safety Tuning
- Case Study — Tülu 3
How to use this
The three modules are meant to be read in order — Architecture builds the machine, Pre-Training fills it with knowledge, Post-Training turns it into an assistant — but each stands on its own. Inside every module, a sticky strip lets you jump between topics, and each topic ends in a small interactive demo: tokenize live text, scrub a training-loss curve, run a DPO step, rebuild GPT-2's parameter count. Everything runs in your browser with no dependencies. Toggle light/dark from the header.