name: TensorFlow Vocabulary
description: >-
  Core vocabulary and taxonomy for TensorFlow, the open source machine learning
  platform. Covers ML model lifecycle concepts, inference types, tensor data
  formats, and operational dimensions.
created: '2026-05-03'
modified: '2026-05-03'
terms:
  - term: Tensor
    definition: >-
      A multi-dimensional array that is the primary data structure in TensorFlow.
      Tensors represent model inputs, outputs, weights, and intermediate
      computations.
    tags:
      - Core Concepts
      - Data Types

  - term: Model
    definition: >-
      A trained machine learning artifact consisting of a computational graph and
      weights that maps inputs to outputs. TensorFlow models are typically saved
      in SavedModel format.
    tags:
      - Core Concepts
      - Model Lifecycle

  - term: SavedModel
    definition: >-
      TensorFlow's serialization format for models. A SavedModel contains a
      complete TensorFlow program including weights, computation graph, and
      metadata. The standard format for deployment.
    tags:
      - Model Lifecycle
      - Serialization

  - term: Serving Signature
    definition: >-
      A named entry point for a TensorFlow model that defines the input and
      output tensor specifications. A model may have multiple signatures for
      different inference tasks.
    tags:
      - TensorFlow Serving
      - Model Lifecycle

  - term: Inference
    definition: >-
      The process of running a trained model on new input data to produce
      predictions, classifications, or other outputs. TensorFlow Serving
      supports three inference modes: predict, classify, and regress.
    tags:
      - Inference
      - Core Concepts

  - term: Prediction
    definition: >-
      Generic inference operation that maps input tensors to output tensors.
      Uses row format (instances array) or column format (inputs object). The
      most flexible inference type.
    tags:
      - Inference
      - REST API

  - term: Classification
    definition: >-
      Inference operation that maps input examples to label-score pairs.
      Follows the PredictionService.Classify gRPC API semantics. Returns
      class labels and associated confidence scores.
    tags:
      - Inference
      - REST API

  - term: Regression
    definition: >-
      Inference operation that maps input examples to continuous numeric
      values. Follows the PredictionService.Regress gRPC API semantics.
    tags:
      - Inference
      - REST API

  - term: TensorFlow Serving
    definition: >-
      A high-performance serving system for TensorFlow models designed for
      production environments. Supports REST and gRPC APIs, model versioning,
      A/B testing, and request batching.
    tags:
      - TensorFlow Serving
      - Production

  - term: ModelServer
    definition: >-
      The TensorFlow Serving executable that hosts one or more models and
      exposes REST and gRPC endpoints for inference. Configured via command
      line or config file.
    tags:
      - TensorFlow Serving
      - Infrastructure

  - term: Model Version
    definition: >-
      A specific numbered iteration of a trained model. TensorFlow Serving
      manages multiple versions simultaneously and supports canary rollouts
      and version-specific routing.
    tags:
      - Model Lifecycle
      - Versioning

  - term: Model Label
    definition: >-
      A human-readable alias (e.g., "stable", "canary") assigned to a model
      version. Labels provide stable routing references without requiring
      version numbers in client code.
    tags:
      - Model Lifecycle
      - Versioning

  - term: Row Format
    definition: >-
      Prediction request format using an 'instances' array where each element
      is one input example. All inputs must share the same 0th tensor dimension.
    tags:
      - REST API
      - Data Formats

  - term: Column Format
    definition: >-
      Prediction request format using an 'inputs' object where each key is a
      named input tensor and values are arrays across all instances. Used when
      tensor dimensions differ.
    tags:
      - REST API
      - Data Formats

  - term: TFJob
    definition: >-
      A Kubernetes Custom Resource Definition (CRD) from the Kubeflow Training
      Operator that enables distributed TensorFlow training on Kubernetes clusters.
    tags:
      - Kubernetes
      - Training

  - term: TensorFlow Hub
    definition: >-
      A repository of pre-trained TensorFlow modules for transfer learning.
      Models are available for image, text, video, and audio tasks.
    tags:
      - Pre-Trained Models
      - Transfer Learning

  - term: TensorFlow Lite
    definition: >-
      A lightweight version of TensorFlow optimized for mobile and embedded
      devices. Supports on-device inference with quantized models.
    tags:
      - Mobile
      - Edge Computing

  - term: TensorFlow.js
    definition: >-
      A JavaScript library for training and deploying TensorFlow models in
      web browsers and Node.js environments.
    tags:
      - JavaScript
      - Browser

  - term: gRPC
    definition: >-
      A high-performance RPC framework used by TensorFlow Serving alongside
      REST. Provides lower latency for high-throughput inference scenarios.
    tags:
      - Protocols
      - TensorFlow Serving