TokenTown –> Learn how LLM's work in a SimCity-like world (laurentiugabriel.github.io)

TokenTown is an innovative isometric city simulation that visually demonstrates how transformer language models operate. Each district represents a stage in the processing pipeline of the model, where a convoy transports data through various stages, including token splitting, embedding, attention mechanisms, and sampling. The simulation effectively translates complex machine learning processes into an engaging visual narrative, allowing users to understand the architecture of large language models (LLMs) in real time within their browser. This project is significant for the AI/ML community as it offers an accessible educational tool that simplifies the mechanics of LLMs for learners and practitioners alike. By scaling down technical aspects—such as reducing dimensions and attention heads—TokenTown lowers the barrier to understanding while preserving key functionalities of larger models. Though the model's weights remain random and artificial, the simulation artfully integrates realistic visual cues and explanations to illustrate how trained models generate text patterns. As a result, TokenTown not only serves as a unique learning platform but also enhances engagement in understanding sophisticated AI concepts.