Tensorflow Import Error on Amazon EC2 Instance

Question

I am trying to run a convolutional neural network in my p3.16xlarge Amazon EC2 instance. I installed tensorflow and all the other requisite libraries for my program using pip install, however, when I run this program, I receive the error message:

"ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory" "Failed to load the native TensorFlow runtime."

These are my instance details: Python 2.7.14 (default, May 2 2018, 18:31:34) Amazon Linux AMI 2018.03.0.20180811 x86_64 HVM GP2

Upon looking into the issue, it seems like I need CUDA 9.0 whereas most linux machines use CUDA 9.1. How exactly can I downgrade CUDA/change/configure my instance so that tensorflow runs? I do not know how to do this.

Thank you

Answer 1

a lot of customers find that the Deep Learning AMI is the simplest way to run their workloads on EC2. It comes pre-installed with all the popular libraries.

https://aws.amazon.com/machine-learning/amis/

Alternatively, you could consider using Amazon SageMaker Store your data in S3, train and deploy on fully-managed infrastructure. Quite easier than managing their own EC2 instances :)

https://aws.amazon.com/sagemaker/

I'd love to hear your feedback and answer any questions.