Ilyushin / datascience-docker-container-nvidia

Docker image for IBM supercomputers which contains GPUs.

Docker image for IBM Power8 which contain GPUs.

Contains widely used Python libraries such as:

• TensorFlow (r1.11)
• Keras
• numpy
• pandas
• sklearn
• sympy
• scipy
• matplotlib
• requests

And for interacting with the container Jupyter was pre-installed.

For using you need to perform following commands:

git clone https://github.com/Ilyushin/datascience-docker-container-nvidia.git
cd datascience-docker-container-nvidia
nvidia-docker build . -t ilyushin/datascience-container-nvidia:latest
nvidia-docker run -it -d -p 8888:8888 -p 7007:7007 ilyushin/datascience-container-nvidia bash

The current version of TensorFlow was built in the following environment:

• tensorflow-1.11.0-cp36-cp36m-linux_ppc64le.whl requires:
  • Python 3.6.7
  • GPU architectures Kepler, Maxwell, Pascal and Volta
  • CUDA 9.1
  • cuDNN 7

How to build your version of TensorFlow:

1. If you need another version of CUDA and cuDNN, you should change the first line in Dockerfile.build
2. Bild an image

   nvidia-docker build -f Dockerfile.build . -t ilyushin/datascience-container-nvidia-build:latest

3. Run the container and connect to it

   nvidia-docker run -it -d --name build-tensorflow ilyushin/datascience-container-nvidia-build:latest
   docker exec -it build-tensorflow bash 

4. Build TensorFlow

   cd ~/tensorflow
   ./configure
   bazel --bazelrc=/root/tensorflow/.tf_configure.bazelrc build -c opt //tensorflow/tools/pip_package:build_pip_package
   bazel-bin/tensorflow/tools/pip_package/build_pip_package ../tensorflow_pkg

5. Copy the assembled package to a local host

   docker cp build-tensorflow:/root/tensorflow_pkg/<package_name> ~/

6. Stop and remove the container and the image, if it needs

   docker stop build-tensorflow
   docker rm build-tensorflow
   docker rmi ilyushin/datascience-container-nvidia-build:latest