This allows the inference code to start up with a prompt, instead of
waiting for user input from stdin. Allows easier scripting, useful for
batch generation, benchmarking etc.