Modelforge

Modelforge is a foundation for sharing trained machine learning models. It is a set of command line tools and a Python library. Modelforge maintains model files in a third-party remote storage service ("cloud") using the backend mechanism. Model metadata (download links, names, descriptions, versions, etc.) resides in a Git repository called the "Index", and documentation is automatically generated there. Modelforge does no assumptions about the models: they can be of any origin, such as TensorFlow, scikit-learn, or your custom. The underlying model storage format - Advanced Scientific Data Format - can wrap any data easily and efficiently, but it's the developer's responsibility to convert.

Learn more about:

• Why? - what problem Modelforge tries to solve.
• Modelforge model - what is a model in Modelforge context.
• Model storage format - low-level serialization details.
• Backends - extension system to upload and download models from clouds.
• Git Index - how documentation about the models is generated from the structured metadata.
• Command line tools - how to perform typical operations.
• API - Modelforge API for developers.

Who uses Modelforge?

• source{d}, in src-d/ml and src-d/lookout-sdk-ml; the public index is src-d/models.

Install

You can run Modelforge through Docker:

docker run -it --rm srcd/modelforge --help

or install it using the Python package manager:

pip3 install modelforge

Usage

The project exposes two interfaces: command line and API.

Contributions

Contributions are pretty much welcome! Please follow the contributions guide and the code of conduct.

If you wish to make your MLonCode model available in src-d/models, please fork that repository and run modelforge publish to upload your model on your fork. Then create a pull request. You should provide read access to the model file for everybody. If you consider using our Google Cloud Storage bucket, feel free to contact us through GitHub issues.

License

Apache 2.0.