How LLMs Work

A ground-up technical course on how large language models are built and trained. Starting from the map of the whole system, it works through tokenization, self-attention, the transformer block, the loss function, backpropagation, optimizers, fine-tuning, and inference — with full derivations at each step.

Audience: technically comfortable, new to ML.
Prerequisites: linear algebra, multivariate calculus, basic probability.

Appendices