Ever wondered what’s actually happening inside ChatGPT when it predicts the next word? This brilliant tool from Georgia Tech runs a real GPT-2 model directly in your browser and visualizes every step of the transformer process in real-time. Type anything and watch attention heads light up, see how embeddings flow through layers, and observe the exact mechanics of next-token prediction.
What sets this apart from static diagrams or theoretical explanations is the interactivity – you can experiment with your own text and immediately see how different inputs affect the model’s internal computations. The visualization breaks down complex concepts like multi-head attention, positional encoding, and layer normalization into digestible, animated components. It’s educational gold that makes transformer architecture finally click for developers who learn by doing.
Perfect for ML engineers wanting to debug their intuition, students tackling NLP courses, or anyone curious about how LLMs actually work under the hood. The fact that it requires zero setup (just open the webpage) and comes with solid academic backing (published at IEEE VIS 2024) makes this an instant bookmark for the AI development community.
⭐ Stars: 6776
💻 Language: JavaScript
🔗 Repository: poloclub/transformer-explainer