mlops
Glossary ↗Model Registry
A model registry is a versioned catalog for trained and fine-tuned models — the equivalent of a Git repository, but for model artifacts and their metadata. Each entry records the model version, the data and code that produced it, evaluation scores, and a stage tag such as staging or production. Instead of a mystery model_final_v3.bin sitting on someone's laptop, the registry becomes the single source of truth for which model is live and how it got there. For SaaS builders this matters most once you start fine-tuning your own models or juggling several open-weight checkpoints. The registry lets you promote a model to production with one action, roll back instantly when quality drops, and prove to auditors exactly what shipped and when. Popular tools include MLflow Model Registry, Weights & Biases, and the Hugging Face Hub. Practical note: always store eval results next to each version so promotion is a data-driven decision, not a guess.
Related terms