About this course

Learn LLM Inference is a free course by fignix on serving large language models: what an inference server does, why it is slow and expensive by default, and the techniques that make it fast and cheap.

It is written for engineers who are new to language models. The theory teaches concepts and keeps the arithmetic to what you can do on a napkin from a model's config file. The labs are where the numbers come from, and every number on a lab page was measured on the machine the lab names, on the date it says.

Reading needs a Fignix account, which is free. Labs run on a free Kaggle GPU or on Modal, and the labs section remembers which path you chose.