Skip to content

Learn · Interactive guide

How LLMs work

Five stages turn a prompt into Gemma 4's next token: tokenization, embeddings, attention, sampling, and the chat loop that ties them together. Walk through each one with a live demo.