q·k The annotated LLM

Loading model…

Inside Qwen3

The diagram shows how Qwen3 processes tokens, with links to the Python implementation.

Reading the model…

Matrix sizes are schematic; see labels for actual dimensions.

01 / MODEL OVERVIEW
An operation or a whole block
Learned weights · fixed
Activations · depend on input
Click to look inside · Pinch inward or Esc to return
THE WEIGHT MICROSCOPE

Learned weights