How to install Keras/TensorFlow on FAST.AI GPU machine


I have a GPU machine on Paperspace with template. Since it’s used for the Deep Learning course based on PyTorch environment, not Keras/TensorFlow which I want to test out, I created another environment “keras” and installed there TensorFlow-GPU and Keras using ‘pip install’. I tried to check TensorFlow version, but there were some errors. The same errors were seen when I tried to check Keras version in Jupyter notebook. Error is: “ cannot open shared object file: No such file or directory.” I checked online and people say it’s because CUDA 9.1 used instead of 9.0. Do you have any suggestions for installing TensorFlow for CUDA 9.1 that this machine has? I’m doing this just to try Keras environment since I already have a machine, instead of creating a new machine like ML-in-a-Box which all of these packages. If it’s too complicated and may affect my environment, I’ll try to create a new machine ML-in-a-Box and delete it when I don’t need it.

Thanks a lot in advance!


I think the problem seems from tensorflow which doesn’t support the latest version of cuda. I suggested this to solve the problem:
Install cuda-9.0:
sudo apt-get install cuda-9-0
Create a symbolic link to it:
sudo ln -sf /usr/local/cuda /usr/local/cuda-9.0