Error when installing Keras on Nano

I followed the instructions at https://elinux.org/Jetson_Zoo and the install went without any errors. But then when I try importing Keras I get an error. Searching this forum suggests I need to install Cuda 10.0. When I go to the Cuda download page I don’t see it for the Jetson Nano. Could somebody please help? Thanks. Here’s the error I get:
t@j:~$ python3
Python 3.6.9 (default, Apr 18 2020, 01:56:04)
[GCC 8.4.0] on linux
Type “help”, “copyright”, “credits” or “license” for more information.

import keras
Using TensorFlow backend.
2020-05-01 10:08:41.120204: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libcudart.so.10.0’; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory
2020-05-01 10:08:41.121096: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2020-05-01 10:08:41.122149: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library ‘libcudart.so.10.0’; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory
2020-05-01 10:08:41.122665: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
Segmentation fault (core dumped)
t@j:~$

The nano SD image is included Cuda 10.0 already.
Could you check /usr/local/cuda has the correct cuda or not?

Thank you for the help. I checked as you requested and the version is 10.2.89. So it seems the keras install wants 10.0. I’m using the SD image disk, so I guess it does not come with 10.0. It is also possible that I did “apt-get upgrade,” which may have messed it up. I will try a fresh install. Could you please tell me how to install an old version? Thanks. – Tim

can you try to use tf.keras instead of keras?

There is no easy way to downgrade cuda version. Also it will not be the correct solution.
Your log shows tensorflow version not match cuda version.

Are you using the TF version for JP4.4 DP as https://elinux.org/Jetson_Zoo mentioned?

Maybe you should use the gpu api to check whether tf match cuda first.
https://www.tensorflow.org/api_docs/python/tf/test/is_gpu_available