Public | Automated Build

Last pushed: a month ago
Short Description
Minimal Keras + CNTK/TensorFlow for Deep Learning
Full Description


Docker container for keras + cntk intended for nvidia-docker. Based off of
Durgesh Mankekar's minimal keras/jupyter container + corresponding blog post.

This container contains CNTK, TensorFlow, and Keras. When Keras is run in the container, CNTK will be the selected backend by default. Only recommended to run on a GPU instance.

Container on Docker Hub


To set up the CUDA+container drivers from scratch on the GPU instance, run these commands:

curl -O -s
sudo sh

You may need to run the script again after a few days (due to a GPU update on host hardware)

To run a deep learning script in the container:

sudo docker run --runtime=nvidia run -it --rm -v $(pwd)/:/keras --name keras minimaxir/keras-cntk python3 <x>.py

Where <x> is the Python script on the host server.

To run TensorFlow on Keras in the container instead of CNTK, add a -e KERAS_BACKEND='tensorflow' flag:

sudo docker run --runtime=nvidia run -it --rm -v $(pwd)/:/keras --name keras -e KERAS_BACKEND='tensorflow' minimaxir/keras-cntk python3 <x>.py

To run a Jupyter Notebook in the container (where invoking Keras in a notebook will use the CNTK backend):

sudo docker run --runtime=nvidia -it --rm -p 8888:8888 -v $(pwd):/keras --name jupyter minimaxir/keras-cntk jupyter notebook --allow-root


Max Woolf (@minimaxir)

Max's open-source projects are supported by his Patreon. If you found this project helpful, any monetary contributions to the Patreon are appreciated and will be put to good creative use.


Install script sourced from Durgesh Mankekar's blog post noted above. (reproduced here for postertity)

Docker Pull Command
Source Repository