Stevedore: a generic lightweight ML build setup in Docker

Wrapper class and generic API plus Podman/Docker build automation for Python ML models. A specific use case is a set of machine learning models that may be composed, so that they should be tested and packaged together. This by no means excludes use on single model.

Wrapper class and generic API plus Podman/Docker build automation for Python ML models. A containerized model suitable for local runner, Kubernetes, or a serverless container runner behind a well-thought HTTP API that forces you to second thoughts about monitoring i.e. requires that you provide some model monitoring and automatically builds some more for you, and whose monitoring endpoints are named uniformly, with JSON payloads validated against the schemas you supply and automated build-time testing of the API endpoints before container finalisation.

HTTP-framework that serves the API is FastAPI. See details in https://github.com/reaktor/ml-py-stevedore

MIT license