docker-cuda-app.md

Building the docker container to run CUDA-enabled Java apps

For the sake of this project you won't need to do any of the below steps in this section, as we already have a pre-baked CUDA-enabled docker image for your use. The docker image contains everything mentioned in the pre-requites in the Advanced installation and setup section.

But if you like to explore further please dive into it.

Build docker image

Builds a CUDA-enabled docker image based on Nvidia's CUDA docker image: nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04 and a GraalVM 19.2.1 image from findepi

$ cd dl4j-nlp-cuda-example/docker
$ ./docker-runner --buildImage   ### DOCKER_USER_NAME is set to your Docker Hub username 

or

$ ./docker-runner --dockerUserName [your Docker Hub username] \
                  --buildImage

Push the docker image to docker hub

Push the created image to the Docker hub, you need an account on Docker Hub.

$ cd dl4j-nlp-cuda-example/docker
$ ./docker-runner --pushImageToHub   ### DOCKER_USER_NAME is set to your Docker Hub username 

or

$ ./docker-runner --dockerUserName [your Docker Hub username] \
                  --pushImageToHub

You will be prompted for a password to your account when you try to push this to the hub, unless you are already logged in.

Run the docker container

In this specific case, running the container would make more sense if you have a machine with a GPU and Nvidia drivers installed and working. See Resources to find out how to go about with that.

$ cd dl4j-nlp-cuda-example/docker
$ ./docker-runner --runContainer   ### neomatrix369 or DOCKER_USER_NAME is set to your Docker Hub username 

or

$ ./docker-runner --dockerUserName [neomatrix369 or your Docker Hub username] \
                  --runContainer

Although if you look at the valohai.yaml file, it's already done on the Valohai platform.

---

Return to main README.md