See a transformer turn a prompt into next-token probabilities

Transformer Explainer is an interactive browser lesson built around GPT-2 small. It connects tokenization, embeddings and attention to the probabilities used to choose the next token. Readers can change a prompt and explore how temperature, top-k and top-p sampling affect the available choices, making abstract model diagrams easier to connect to actual generation.

Polo Club of Data Science

Why it matters

A practical learning resource for engineers who want to understand model behavior beneath an API.

More context

Try the browser demo with a short prompt and vary one sampling control at a time. The project uses a 124-million-parameter GPT-2 model with ONNX Runtime. This is an older educational discovery with no verified original publication date.

Keep in mind

GPT-2 is much smaller and simpler than current frontier systems. The visualization does not explain every modern architecture; 1min reviewed the project text, not an interactive test session.

Original read