Set up the conda environment by using:

$ conda env create -f environment.yml

Then download CuDNN 8.1.0 from https://developer.nvidia.com/rdp/cudnn-archive and add CUDA and CuDNN path to PATH and LD_LIBRARY_PATH. For example:

export PATH="PATH_TO_CUDNN/cudnn-11.2-linux-x64-v8.1.0.77/cuda/lib64:PATH_TO_ANACONDA/envs/tensorflow_gpu/lib:${PATH}"
export LD_LIBRARY_PATH="PATH_TO_CUDNN/cudnn-11.2-linux-x64-v8.1.0.77/cuda/lib64:PATH_TO_ANACONDA/envs/tensorflow_gpu/lib:${LD_LIBRARY_PATH}"

inspecting saved models

saved_model_cli show --dir saved_model --tag_set serve --signature_def serving_default
python /home/leo/anaconda3/envs/cuda_11_1/lib/python3.7/site-packages/tensorflow/python/tools/import_pb_to_tensorboard.py --model_dir ./saved_model --log_dir /tmp/tensorflow_logdir