suaroman

Gregory J. Suarez · Engineering notebook

Notes on how language models work

I learn by building things and taking them apart. Then I write it down as films, hands-on labs and posts that explain how modern software and AI systems work, starting from first principles.

Watch: How an LLM works Read the writing
Next-token sampler context: “Notes on”
  1. how language models work 45.6%
  2. running models on my own hardware 24.4%
  3. Spark and data engineering 13.1%
  4. building with AI agents 10.2%
  5. tools worth knowing 5.4%
  6. why the moon looks bigger 1.4%
The headline is drawn from this distribution, the way a model picks its next token. Turn the temperature up and unlikely phrases get a real chance. Try the full Temperature lab.

Featured film

All films →
1:19:12

The Personal Learning Series

How a Large Language Model Works

A narrated, fully animated film that follows one sentence through a language model: tokens, vectors, attention, transformer layers, sampling, training, and what it all means.

  1. 01 Watch the film 1:19:12 · 59 chapters
  2. 02 Practice the labs 30 labs, in your browser
  3. 03 Review the workbook quizzes · glossary · explain-it-back
  4. 04 Read the notes why and how it was made

Latest writing

All posts →
About

Hi, I'm Greg.

I'm an engineer who learns by building. Lately that means running language models on my own hardware, taking them apart, and turning what I find into films, labs and notes that explain them from zero.

More about me →