Reviewing Andrej Karpathy’s MicroGPT: Training a GPT in ~200 Lines of Python

​In this session we will review a blog post by Andrej Karpathy on MicroGPT, a minimal implementation of a GPT-style language model written in roughly 200 lines of pure Python with zero external dependencies. The project demonstrates the core algorithmic components required to train and run inference for a GPT model—including tokenization, a simple autograd engine, a GPT-like architecture, an optimizer, and the training loop all in a single file.

​

​https://karpathy.github.io/2026/02/12/microgpt/

​

​Will meet online at:

​https://discord.gg/wE3ufNvQN8