Running Tensor Flow and Keras in Jupyter notebook via Docker
Assuming that you have NVIDIA GPU in your machine and NVIDIA Drivers installed. Please read the belowbefore you install NVIDIA drivers
The easiest way to use TF is via the docker container; which uses the nvidia docker container.
If you do not need Keras you can use like below
Tip: If you are working on a remote GPU machine you can use the port forwarding option so that you can browse the Jupyter notebook from your local machine
ssh -L 8888:127.0.0.1:8888 root@<remotemachine ip>
----------docker run --entrypoint=/bin/bash --env http_proxy<if you're behing one> --env https_proxy=<xyz> --runtime=nvidia -p 8888:8888 -it --rm -v /usr/alex/:/coding --net=host tensorflow/tensorflow:1.13.0rc1-gpu-jupyter
----------
cd /coding
jupyter notebook --alow-root
(do note the token)
-----------
And since port 8888 remote is mapped to local, you can fire up the webpage from your machine
http://127.0.0.1:8888/?token=<pass in the token copied from remote console>You should be able to see jupyter notebook at http://127.0.0.1:8888
I have installed Keras and some other python libs on top of the official tensorflow_tfserving:latest-gpu Docker image and saved it as alexcpn/tfserving-dev-gpu.
Happy coding!
