Processing time is delayed when executing in multiple threads #222
Description
I have built and tested a model impl based on your code. I have problems when initializing model instance in multiple threads and running concurrently. The processing time will increase gradually when there are more threads running even though GPU, CPU, RAM are still redundant. If I run multiprocessing then the time to complete a process will not change or increase very little when creating other processes if the system has spare resources.