transformer-explainer

Transformer Explainer is an educational visualization tool from the Polo Club at Georgia Tech that helps people understand how Transformer-based text-generation models such as GPT work. Rather than only describing the architecture in prose, it runs a live GPT-2 model directly in the browser so users can enter their own text and observe, in real time, how internal components and operations combine to predict the next tokens. Key points from the repository: - Interactive learning: users experiment with their own input text and see the model's internal flow as it generates predictions. - Runs in the browser: a live GPT-2 model executes client-side, so no server-side inference setup is required for the hosted demo. - Public demo: available at poloclub.github.io/transformer-explainer, with a linked YouTube demo video. - Research-backed: described in a paper, "Transformer Explainer: Learning LLM Transformers with Interactive Visual Explanation and Experimentations," published in the Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems, with an arXiv entry (2408.04619). - Local development: requires Node.js v20+ and NPM v10+; clone the repository, run npm install, then npm run dev, and open http://localhost:5173. - License: MIT. - Related explainers from the same group include Diffusion Explainer, CNN Explainer, and GAN Lab. The project is aimed at students, educators, and anyone curious about how large language models turn text into next-token predictions. It is a visualization and teaching aid rather than a production inference library or model-serving framework.