Installation

Here is the step-by-step guide to install MLCube library and run simple MLCube cubes.

Create a python environment

CondaVirtualEnv

conda create -n mlcube python=3.8
conda activate mlcube

virtualenv --python=3.8 .mlcube
source .mlcube/bin/activate

Install MLCube Runners

Reference MLCube runners are distributed in separate python packages.

DockerSingularityGCP (alpha)K8S (alpha)Kubeflow (alpha)SSH (alpha)

pip install mlcube-docker    

pip install mlcube-singularity    

pip install mlcube-gcp    

pip install mlcube-k8s    

pip install mlcube-kubeflow    

pip install mlcube-ssh    

Warning

GCP (Google Cloud Platform), K8S (Kubernetes), Kubeflow and SSH runners are in early stages of development.

Check that the docker runner has been installed.

mlcube config --get runners

Show MLCube system settings.

mlcube config --list

Information

This system settings file (~/mlcube.yaml) configures local MLCube runners. Documentation for MLCube runners describes each of these parameters in details. A typical first step for enterprise environments that are usually behind a firewall is to configure proxy servers.

platforms:
  docker:
    env_args:
      http_proxy: http://ADDRESS:PORT
      https_proxy: https://ADDRESS:PORT
    build_args:
      http_proxy: http://ADDRESS:PORT
      https_proxy: https://ADDRESS:PORT

Explore with examples

A great way to learn about MLCube is try out the example MLCube cubes located in the mlcube_examples GitHub repository.

git clone https://github.com/mlcommons/mlcube_examples.git 
cd ./mlcube_examples
mlcube describe --mlcube ./mnist