Using "CUDA_VISIBLE_DEVICES=0" to accelerate my python program has no effect?

I want to accelerate my common python program which only imports cv2,numpy,math and scipy by using os.environ[“CUDA_VISIBLE_DEVICES”]=“0” at the front of my program,but the time the program used was same as the one without using GPU. Have some friends met this problem and have solved? I use the Jetson nano orin NX T801

There is no reason to expect that using CUDA_VISIBLE_DEVICES=0 will make a program run faster. Declaring the CUDA_VISIBLE_DEVICES variable is not necessary to allow an application to use the GPU.

When i run !nvidia-msi it shows the GPU and CUDA version as 12.3 but the latest Pythorch is for 12.1. when i search for GPU in PyTorch it shows 0 devices. How to mitigate this issue?