MLflow
MLflow is an open-source framework for experiment tracking, model packaging, and model registry that Azure Machine Learning uses natively. AI-300 requires tracking experiments and registering MLflow models.
All AI-300 Terms
MLOps
MLOps (machine learning operations) is the practice of automating and operationalizing the ML lifecycle — training, deployment, monitoring, and retraining.
GenAIOps
GenAIOps (generative AI operations) is the operational discipline for deploying, evaluating, monitoring, and optimizing generative AI applications and agents.
Azure Machine Learning
Azure Machine Learning is the Azure service for training, deploying, and managing traditional ML models with workspaces, compute, and pipelines.
MLflow
MLflow is an open-source framework for experiment tracking, model packaging, and model registry that Azure Machine Learning uses natively.
Model registry
A model registry is a versioned catalog of trained models and their metadata used to manage promotion and deployment.
Data drift
Data drift is the change over time in the distribution of production input data relative to training data, which can degrade model accuracy.
Managed endpoint
A managed endpoint is an Azure Machine Learning hosted service that serves a model for real-time or batch inference with managed infrastructure.
Progressive rollout
Progressive rollout is a deployment strategy that shifts traffic to a new model version gradually so it can be validated and safely rolled back.
Bicep
Bicep is a domain-specific language for declaratively deploying Azure resources as infrastructure as code.
Infrastructure as code
Infrastructure as code (IaC) is the practice of provisioning and managing infrastructure through version-controlled definition files rather than manual steps.
GitHub Actions
GitHub Actions is a CI/CD automation platform that runs workflows on repository events.
Provisioned throughput
Provisioned throughput units reserve dedicated model capacity for predictable, high-volume generative AI workloads.
Groundedness
Groundedness is an AI quality metric that measures how well a generated response is supported by the provided source context.
Evaluation
Evaluation is the systematic measurement of generative AI quality and safety using test datasets and metrics.
RAG
RAG (retrieval-augmented generation) grounds a model's output in retrieved data to improve relevance and accuracy.