IML4E – Model cards toolkit

As ML models are based on data rather than source code, many things are blackbox. Therefore, additional stakeholders, some of which can be non-technical, can have concerns related to a served model and its specific version. To elevate ML development to systematic engineering practice, traceable information must be captured.

Model cards toolkit works in existing workflows as a quality gate that ensures that relevant information is captured. Case-specific rules can be specified, and the idea is not have extensive rule base but rather ensure that “just enough” information is recorded.