phizaz/tensorflow repository overview

⁠Non-root Tensorflow based on original tensorflow (python 3) image

This Dockerfile extends the tensorflow/tensorflow:latest-gpu-py3 image by adding the ability to run in "non-root" mode meaning the output the jupyter notebook will be totally owned by a specific user (given a user id).

This Dockerfile use fixuid (https://github.com/boxboat/fixuid⁠) to leverage the aforementioned ability.

This Dockerfile is based on tensorflow/tensorflow:latest-gpu-py3 than means it supports Python 3.

⁠What's in the Dockerfile

• tensorflow/tensorflow:latest-gpu-py3
  • jupyter
  • python 3
  • tensorflow
• jupyter nbextensions
• apt-fast (https://github.com/ilikenwf/apt-fast⁠)
• fixuid (https://github.com/boxboat/fixuid⁠)

⁠Build

nvidia-docker build -t phizaz/tensorflow:gpu -f Dockerfile.gpu .

⁠Build :keras-gpu

nvidia-docker build -t phizaz/tensorflow:keras-gpu -f Dockerfile.gpu.keras .

⁠Usage

NB_PORT=10010
nvidia-docker run -it --rm \
    --user=$UID:$UID \
    -p $NB_PORT:$NB_PORT \
    -v <src>:/notebooks \
    phizaz/tensorflow:gpu \
    jupyter notebook --ip="*" --port=$NB_PORT

With --user you can specify the runner of this image which in this case is the current user $UID.

Or with :keras-gpu instead.

You can quick start by seeing the test directory within this repo. There is a notebook

⁠Tag Variations

⁠phizaz/tensorflow:keras-gpu - with Keras (only with tensorflow backend) with GPU support

⁠phizaz/tensorflow:dcgan-gpu - A Dockerfile for carpedm20's DCGAN Tensorflow

Visit: https://github.com/phizaz/dockerfiles/tree/master/tensorflow/DCGAN_tensorflow⁠