Multiprocessing for CUDA App stuck on seconds request on Serverless worker.

Hi, I deployed an app in Runpod and run multiprocessing on serverless.
But from second request to the worker the request was stucking.
How can I resolve this issue?