Skip to content

Latest commit

 

History

History

docker

Docker image for MetaDL competition

The Docker image used for MetaDL challenge is created with Dockerfile in this directory.

Here is the link to Docker Hub.

Build Docker image for GPU/CPU

For GPU, use

docker build -t ebadrian/metadl:gpu .

For CPU, use

docker build -t ebadrian/metadl:cpu -f Dockerfile.cpu .

If you have push access, you can do

docker push ebadrian/metadl:gpu

or

docker push ebadrian/metadl:cpu

For simplicity, two scripts are written to automate the build

./build.sh

and the release

./release.sh

Creating a compute worker for the CodaLab competition

The worker_setup.sh is a shell script created to install some modules and packages in an Azure VM. More specifically the VM should have Unbuntu 18.04 (x86_64). Once created, you can clone the GitHub repository with the following command :

git clone https://github.com/ebadrian/metadl.git

Then you can run the worker_setup.sh script to install all the required softwares/packages to create a compute worker for this competition.

cd metadl/docker/
bash ./worker_setup.sh

The process may take several minutes and user input is required to accept the Here are the softwares/packages installed during the process:

  • Gcc compiler and associated packages
  • Nvidia cuda
  • Docker
  • Nvidia docker

Note: If you want to use such a worker for another CodaLab workers queue, you must change the environment variable BROKER_URL with the corresponding CodaLab queue identifier. More detailled informations about this can be found in the dedicated wiki.