I have windows 10 64 bit installed and when I executed train.py I get the following error: Failed to load the native TensorFlow runtime.

[aliNader]

this is the error details:

[ghost]

Have you installed Tensorflow using PIP?

I found that installing Tensorflow on Windows 10 using PIP results in DLL errors, especially when being configured with GPU acceleration. This gave me headaches for days on end, I was told to build Tensorflow from source and whole slew of other suggestions, until I found out Anaconda is a much better way of handling this.

I'm currently running Ubuntu 18.04, however this guide should point you in the right direction even if you're running Windows 10.

Installing Tensorflow with GPU Support - Puget Systems

Be sure to specify the Python version while creating a virtual environment.

conda create --name tf-gpu python=3.6

NOTE - If you do not want to run Tensorflow with GPU acceleration then use this command

conda install tensorflow

instead of the original command posted in the article.

[Chadhendrixs]

it's because Paperspace's tensorflow-gpu version is too new. Do this to fix it: pip install --upgrade tensorflow-gpu==1.4

[kaljitism]

Its the Cuda error, nothing can be done apart from modifying code base's tensorflow version migration to 2.x to match with current Cuda version.