Experiment Tracking & Model Registry¶

Coming soon.

• Versioning models, datasets, and hyperparameters so any past run is reproducible
• Choosing what to log (metrics, artifacts, lineage) and what's noise
• Promoting a registered model from experiment to staging to production
• Rollback: reverting a production model to a known-good prior version

Related¶

• Feature Store — the reproducible inputs a tracked experiment depends on
• Model Training Pipelines — where tracked experiments come from
• AI Model — Fine-Tuning — a fine-tuning run is exactly the kind of experiment this topic will cover tracking for

Part of MLOps → AI Engineering.