MLCube Project Template

This repository provides a cookiecutter-based template for MLCommons MLCube®. A detailed description of MLCubes can be found here.

Prerequisites

Currently, the MLCube supports docker- and singularity-based cubes. The cookiecutter template comes with docker configuration files.

Initialize python environment

• Install cookiecutter for creating MLCubes from templates.
• Install mlcube-docker for running docker-based cubes.

virtualenv -p python3 ./env && source ./env/bin/activate
python -m pip install --upgrade pip && pip install cookiecutter mlcube-docker

Creating MLCubes from cookiecutter template

Run the following commands to create a directory for MLCube projects and create a simple project from cookiecutter template. The cookiecutter will ask several questions, such as MLCube name and author.

Do not use ~/mlcube directory. It is used by the MLCube runtime as a default cache directory and storage location. For instance, by default, MLCube stores singularity containers here.

$ mkdir -p ~/mlcube_projects && cd ~/mlcube_projects 
$ cookiecutter https://github.com/mlcommons/mlcube_cookiecutter.git

mlcube_name [my_mlcube]:
mlcube_description [my_mlcube_description]:
author_name [mlcube_author_name]:
author_email [mlcube_author_email]:
author_org [mlcube_author_org]:

Running MLCubes

The MLCube template project that has just been created provides an example implementation of a task that MLCube runners can run. Assuming your MLCube name is my_mlcube, configure and run it using docker container runtime:

$ cd ./my_mlcube && ls ./workspace
parameters_file.yaml

$ mlcube configure --mlcube=. --platform=docker -Prunner.build_strategy=auto
$ mlcube run --mlcube=. --platform=docker --task=my_mlcube

$ ls ./workspace
output.txt           parameters_file.yaml

$ cat ./workspace/output.txt 
Running my_mlcube task with parameters '{'abc': 123}'.

Distributing ML workloads as MLCubes.

Start updating your MLCube by studying the mlcube.yaml file in the root directory of your MLCube. Every configuration file provides a description and what needs to be updated next. Mode details are available at MLCube wiki.

Contribution guidelines

Thank you for your interest in contributing to MLCube! The best way to get started is to make sure you signed the Google CLA (https://cla.developers.google.com/clas), soon to be replaced by a MLCommons/MLCube CLA.

• If you are making a trivial change (such as a typo fix), please feel free create a PR once you have signed the CLA.
• For larger changes, please join our mailing list https://groups.google.com/g/mlcommons-best-practices and/or slack http://mlperf.slack.com/ to discuss.

Resources

• MLCommons
• MLCube
• MLCube Wiki
• MLCube Examples

License

Apache License 2.0