Docker container with Tensorflow GPU for Jetson Nano

Build docker image

At first, you need to install Jetpack 4.2.1 for Jetson Nano.

Next, set nvidia runtime as default for Your docker daemon.

In file /etc/docker/daemon.json add line

"default-runtime": "nvidia"

/etc/docker/daemon.json

{
"default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
}

When you set this, you can build the image using our Dockerfile with docker build command.

Verify

To check if it's working correctly run container and import tensorflow in python3. If it's working fine, the result will be:

$ docker run -it *<docker image name>* bash

root@3ff5d388cfc9:/# python3
Python 3.6.8 (default, Jan 14 2019, 11:02:34) 
[GCC 8.0.1 20180414 (experimental) [trunk revision 259383]] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as ts
>>> print(ts.__version__)
1.13.1