I would like to quit the GPU process from inside a python session (that uses cupy). I would like to release all memory (some 270 MB always remain in use even when cupy functions have completed). Furthermore, a library seems to lead to memory misallocation, throwing an error, if executed several times in a row within the session. I would like to keep the python session running since it has preprocessed (large) data in memory which I would like to examine using the GPU in multiple different ways in a sequential manner.
Is there something like a quit() or exit() function in cupy to only terminate the GPU process and would allow the re-execution of a new “Cuda session” within the existing python session? Any hint would be highly appreciated.