tl;dr: An interactive visualization tool (Transformer Explainer) runs a live GPT-2 small model in the browser to demonstrate how Transformers work, walking through embedding, multi-head self-attention, MLP layers, and output sampling. Built by Georgia Tech researchers using Svelte and D3.js with ONNX Runtime, it lets users input text, tweak temperature/top-k/top-p parameters, and inspect attention weights and intermediate computations in real time.