There are certain behaviors in standard Python multiprocessing that we don't explicitly see in our library (although you could wrap launch() in such utilities).
For example: starting a "server" that runs in the background (#65), running launch() commands in parallel (e.g. evaluate 8 datasets in parallel on 8 GPUs) like a Pool, Pipes that would permit "yielding" objects from running processes (#54), and so forth.