CUDNN_STATUS_NOT_INITIALIZED when using benchmark.timeit

I use Benchmark.Timer with the official docker
Image Layer Details - pytorch/pytorch:1.13.1-cuda11.6-cudnn8-runtime | Docker Hub
and I ran into CUDNN_STATUS_NOT_INITIALIZED error. I found that the nvidia-smi shows cuda version 10.2 in the docker, so I wonder if this problem was caused by a version mismatch between cuda driver and runtime?

Could you post a minimal and executable code snippet reproducing the issue as well as the output of python -m torch.utils.collect_env, please?