Skip to content

Developer setup

Installation

This section covers the installation of a development environment for contributing to MoZuMa. If you want to use MoZuMa in your project, see installation instructions instead.

We are developing on Python 3.7 and PyTorch 1.9.

This method is the recommended method as it will install a complete environment with PyTorch. It requires conda and tox to be installed.

Create a development environment:

# CPU development
tox --devenv venv -e py37-pt19
# or with GPU support
tox --devenv venv -e py37-pt19-cuda111

The environment can be activated with:

conda activate ./venv

Using pip

This method requires you to install PyTorch and TorchVision (see PyTorch documentation).

pip install -r requirements.txt
# To install MoZuMa in development mode with all dependencies
pip install -e .

Testing

Testing can be done using tox:

# CPU testing
tox -e py37-pt19
# or with GPU support
tox -e py37-pt19-cuda111

or with directly using pytest on an environment with all dependencies installed

pytest

Code quality

We use black as formatter. Install pre commit hooks for systematic styling on commits:

pip install pre-commit
pre-commit install

Publish a new version

  • Push the new version to the master branch
  • Test the release on all supported Python / PyTorch versions with the command. 
    tox -e "{py37,py38,py39}{-pt19,-pt110}{-cuda111,}"
  • Create a GitHUB release on the branch master with format vX.Y.Z. For instance, v0.1.1.

Upload new model weights

Note

You will need to set the GH_TOKEN environment variable to the token given by gh auth status -t of the GitHUB CLI.

  • Add the model and provider stores to the scripts/update_public_store.py. They should be added in a function that returns a list of tuple with model and provider store.
  • Update the get_all_models_stores to iterate over your new function.
  • Run the update_public_store scripts. This script accepts a --dry-run argument to see changes before actually uploading models.

Update documentation

Install requirements to build the documentation:

pip install -r docs/requirements.txt

We are using mkdocs-material. It allows for previewing the changes with hot-reload:

mkdocs serve

The documentation can then be deployed to GitHUB pages manually with:

mkdocs gh-deploy

Or automatically when merging a Pull Request into the main branch with GitHUB actions.

Adding notebooks to the documentation

When adding new notebooks in the documentation, the bash script docs/pair-notebooks.sh should be executed to create the notebooks in Markdown version.

Once the notebooks are paired to a markdown version, they can be updated with the docs/sync-notebooks.sh script.