vocabulary: name: Seldon Vocabulary description: > Domain vocabulary for Seldon MLOps platform covering inference, model serving, drift detection, explainability, and enterprise deployment management. version: "1.0" created: "2026-06-12" source: https://docs.seldon.ai terms: - term: SeldonDeployment definition: > A Kubernetes custom resource that describes one or more inference graphs consisting of predictors, transformers, combiners, and routers deployed as microservices. tags: [deployment, kubernetes, CRD] - term: Predictor definition: > A component within a SeldonDeployment that wraps a machine learning model and exposes it as an inference service via REST or gRPC endpoints. tags: [inference, model-serving] - term: InferenceServer definition: > A pre-packaged, optimised server for running models from popular ML frameworks such as TensorFlow, PyTorch, scikit-learn, XGBoost, and MLflow. tags: [inference, framework] - term: OpenInferenceProtocol definition: > A standardised REST and gRPC API contract (also called V2 protocol) for model serving interoperability between different inference servers and orchestration layers. tags: [protocol, standard, inference] - term: Pipeline definition: > A directed acyclic graph of models and transformers through which an inference request flows, enabling chaining, routing, and combining of multiple models. tags: [pipeline, graph, routing] - term: CanaryDeployment definition: > A gradual traffic-shifting deployment strategy that routes a configurable percentage of inference requests to a new model version while the stable version handles the rest. tags: [deployment, strategy, canary] - term: ABTest definition: > A deployment strategy that splits inference traffic between two or more model variants to compare predictive performance under production conditions. tags: [deployment, strategy, a/b-testing] - term: ShadowDeployment definition: > A deployment mode in which a shadow model receives a copy of live inference traffic but its responses are not returned to clients, enabling safe offline evaluation. tags: [deployment, strategy, shadow] - term: DriftDetector definition: > A model deployed via Alibi-Detect that monitors incoming request feature distributions against a reference dataset to detect statistical data drift. tags: [monitoring, drift, alibi-detect] - term: Explainer definition: > A service deployed alongside a prediction model that uses the Alibi library to generate human-interpretable explanations (e.g., SHAP, Anchors, Integrated Gradients) for individual model predictions. tags: [explainability, alibi, shap] - term: ModelMetadata definition: > Structured information about a deployed model including its name, version, inputs, outputs, and framework, returned by the /v2/models/{model_name} endpoint. tags: [metadata, model-info] - term: InferenceRequest definition: > A structured payload sent to an inference endpoint containing one or more named input tensors with their shapes, datatypes, and data values. tags: [inference, request, tensor] - term: InferenceResponse definition: > A structured payload returned from an inference endpoint containing one or more named output tensors with their shapes, datatypes, and inferred data values. tags: [inference, response, tensor] - term: Namespace definition: > A Kubernetes namespace used to scope and isolate SeldonDeployment resources and their associated services within a cluster. tags: [kubernetes, isolation] - term: OIDC definition: > OpenID Connect — the authentication protocol used by the Seldon Enterprise Platform REST API to authenticate programmatic clients. tags: [security, authentication, oidc] - term: AlibiDetect definition: > An open-source Python library for outlier, adversarial, and concept drift detection, integrated into Seldon Core as a model-type deployment for monitoring inference traffic. tags: [monitoring, drift, library] - term: Alibi definition: > An open-source Python library providing model explainability algorithms including SHAP, Integrated Gradients, Counterfactuals, and Anchors, integrated into Seldon as an explainer service. tags: [explainability, library] - term: SeldonCore definition: > The open-source Kubernetes operator and microservice framework that enables packaging, deploying, and orchestrating machine learning models as inference services. tags: [core, open-source, kubernetes] - term: SeldonCore2 definition: > The second-generation Seldon Core platform built on a Kafka-based data plane for decoupled, streaming-capable inference pipelines with improved scalability. tags: [core2, kafka, streaming] - term: SeldonEnterprisePlatform definition: > The commercial Seldon product that extends Seldon Core with a management REST API, UI dashboard, advanced monitoring, RBAC, and enterprise support features. tags: [enterprise, commercial, management] - term: Feedback definition: > An endpoint (/api/v1.0/feedback) that accepts ground-truth labels for previous predictions, enabling reward-based routing for multi-armed bandit deployments. tags: [feedback, routing, bandit]