The model registry component of the MLOps stack serves as a good connection between Data Scientists and ML Engineers.
DS can register production-ready models, and MLEs can access them and push them to production.
All the information related to the model is there.
Model weights, parameters, info about the dataset it was trained on, or about the person that registered it – basically, any metadata that might be needed.
Having this one source of truth for all the models makes the work much more organized and saves a lot of time. Plus, it gives you peace of mind when someone says “audit”.
In this video, Parth Tiwary presents Neptune’s model registry capabilities.
Register a model
Create model version
Associate metadata with model version
Resume a model version
Viewing models and model versions in the web app
Check Neptune’s model registry documentation to learn how to use it.