How a language model writes, one token at a time

From the text you type, to tokens, to vectors, through 64 layers of a real model, to a probability for every possible next token. Then we pick one and do it all again.

Every number about the model is real and comes from Qwen3.8-27B's own configuration. Example values inside vectors are illustrative. Pick a chapter below, or start from the beginning.

Start with chapter 1 →