am.
← All projects
Learning / Live

LLMs, from scratch

Understand a language model by taking it apart.

Interactive byte-pair encoding training exercise with vocabulary and merge steps

The idea

Explore tokenization, attention and training through browser experiments accompanying Sebastian Raschka’s book.

What I worked on

Built the interactive web presentation, browser algorithm ports, exercises and narrated chapter tours.

Context & attribution

An unofficial companion to Build a Large Language Model (From Scratch) by Sebastian Raschka, based on the book and its source-code repository.

Learn by changing things

Adjust an input, step through a calculation, and inspect the result. Seventeen interactive pages cover the book’s main chapters and selected extras, from tokenization to fine-tuning.