Downloading and compiling the model in your browser via WebGPU. This can take a while on first load — it is cached afterwards.