Dockerfiles for both CPU and GPU builds are available in standalone folder. To use the CPU only version use the commands given. A GPU version of docker can be run using the command nvidia-docker instead of docker using the standalone/gpu folder.

Dockerfiles for CPU build is provided in standalone/cpu folder. The image can be built by running:

docker build -t caffeonspark:cpu standalone/cpu

After the image is built, use docker images to validate.

Hadoop and Spark are essential requirements for CaffeOnSpark. To ensure that both process runs flawless, we have included standalone/cpu/config/bootstrap.sh script which must be run everytime the container is started.

To launch a container running CaffeOnSpark please use:

docker run -it caffeonspark:cpu /etc/bootstrap.sh -bash

Now you have a working environment with CaffeOnSpark.

To verify installation, please follow GetStarted_yarn guide from Step 7.