Slow inference speed

[maguro27]

Thank you for your great work!
Although I rewrite the code for TensorFlow ver. 2.5.0 and the code might work well, the inference code is very slow (in my case, when I parallelize the evaluation code to eight, the evaluation takes 2 days to 4 days).
Is this behavior correct?
My environment is as follows,

CUDA toolkit: 11.2
cuDNN: 8.1.0
GPU: NVIDIA RTX A6000
CPU: AMD EPYC 7642 (48-core)
TensorFlow: 2.5.0
DarkNet: The newest DarkNet

Currently, I set the GPU_ID="0" and DARKNET_GPU=1 in the constants.py.
Thank you in advance!