> New SSoT: https://gitlab.com/groups/gitlab-org/-/epics/8560 Overview: https://www.youtube.com/watch?v=V4hos3VFeC4 GitLab issue: https://gitlab.com/gitlab-org/gitlab/-/issues/364648 ## Opportunity By adding an option for a user to add a url to an MLFlow instance, we can surface the information on different parts of the DevOPS lifecycle, and create pre-defined flows (for example MR that automatically pushes a built model to MLFlow). Make GitLab work by default with MLFlow, and in the future expand this capability to other Model Registry. ## Rice+ Justification - Reach: **High**, MLFlow is one of the most used Model Registries out there - Impact: **High**, as it enables integrations with other Model Registries in the future - Confidence: **Moderate**, will require UX research to understand what information to surface and where - Effort: **Moderate**, each potential feature will require a bit of work - Colab: **Low**, part of Package/Monitor/Deploy but can be developed independently - MLOps Branding: **High**, this feature targets DS users alone ## Additional Context ### What is a Model Registry? As the name suggests, similar to other registries Model Registry is a catalogue for ML models. It is used to store different versions of a model and it's associated metadata: distribution of input data, version, hyperparameters, training performance, etc. MLFlow is an example, but there are many other vendors/OSS tools: neptune.ai, comet.ml, mlem.ai, vertex.ai model registry (https://cloud.google.com/vertex-ai/docs/model-registry), etc ### How can this work be expanded? We can create this in a way to pull info from any other model registry, not only MFLow