Possible reasons: CUDA version is wrong

1. View the next version information

cat /usr/local/cuda/version.txt  

2. According to the relationship table of CUDA and tensorflow versions, check whether their versions match

3. If it doesn’t match

Please install the corresponding tensorflow or CUDA version

Install the GPU version of the corresponding version tensorflow

pip install tensorflow-gpu==version number

Install the corresponding version of CUDA

4. If it matches, an error is still reported

That’s because your CUDA environment variable configuration is wrong. Please execute:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/[CUDA Version]/lib64

For example

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-9.0/lib64