Introduction⚓︎

The documentation for the MLOps building block is organized as follows…

• Introduction
  Introduction to the BB - including summary of purpose and capabilities.
• Design
  Description of the BB design - including its subcomponent architecture and interfaces.
• Administration
  Deployment guide, configuration and maintenance of the BB.
• Usage
  Tutorials, How-to guides, etc. to communicate usage of the BB.
• API
  Details of APIs provided by the BB - including endpoints, usage descriptions and examples etc.

About MLOps⚓︎

The MLOps BB provides support services for training of machine learning models within the cloud platform. It also integrates within the EOEPCA Building Block ecosystem, with the other Building Blocks such as Processing, Workspace, and Resource Discovery.

This building block is implemented with SharingHub, a service that enable the discovery of hosted repositories through an integrated STAC catalog answering STAC API. Each repository conforms to the definition of a STAC item, with some STAC extensions (notably ML Model). The content of this catalog is generated on the fly from the information stored in the GitLab projects and is depending on the authenticated user and his/her access rights. It is also integrated with an MLflow instance that can serve as a remote tracking server for model experiments, as well as expose a model registry.

Capabilities⚓︎

The MLOps BB orchestrates the training of ML models in a variety of popular machine learning frameworks, maintains a history of training runs with associated metrics used to assess model performance, and maintains the associated training data.

The Design section can give a good overview of the features expected for this Building Block.