TensorFlow · Schema

TensorFlow Serving Prediction Request

Schema for a prediction inference request to TensorFlow Serving. Supports row format (instances array) and column format (inputs object).

AIDeep LearningJavaScriptMachine LearningModel ServingNeural NetworksOpen SourcePython

Properties

Name Type Description
signature_name string Optional name of the signature to use. Defaults to the model's default serving signature.
instances object Row format input array. Each element is a single prediction input - can be a scalar, list, or named-input object.
inputs object Column format input. Map of input tensor names to arrays of values across all instances.
View JSON Schema on GitHub

JSON Schema

tensorflow-serving-prediction-request-schema.json Raw ↑
{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "$id": "https://github.com/api-evangelist/tensorflow/blob/main/json-schema/tensorflow-serving-prediction-request-schema.json",
  "title": "TensorFlow Serving Prediction Request",
  "description": "Schema for a prediction inference request to TensorFlow Serving. Supports row format (instances array) and column format (inputs object).",
  "type": "object",
  "properties": {
    "signature_name": {
      "type": "string",
      "description": "Optional name of the signature to use. Defaults to the model's default serving signature."
    },
    "instances": {
      "description": "Row format input array. Each element is a single prediction input - can be a scalar, list, or named-input object.",
      "oneOf": [
        {
          "type": "array",
          "items": { "type": "string" },
          "description": "Array of string inputs"
        },
        {
          "type": "array",
          "items": { "type": "number" },
          "description": "Array of numeric inputs"
        },
        {
          "type": "array",
          "items": {
            "type": "object",
            "description": "Named input object with multiple tensor values",
            "additionalProperties": true
          },
          "description": "Array of named-input objects"
        }
      ]
    },
    "inputs": {
      "type": "object",
      "description": "Column format input. Map of input tensor names to arrays of values across all instances.",
      "additionalProperties": {
        "oneOf": [
          { "type": "array" },
          { "type": "number" },
          { "type": "string" }
        ]
      }
    }
  },
  "examples": [
    {
      "instances": ["hello", "world"]
    },
    {
      "signature_name": "serving_default",
      "instances": [
        {
          "tag": "foo",
          "signal": [1, 2, 3, 4, 5]
        }
      ]
    },
    {
      "inputs": {
        "tag": ["foo", "bar"],
        "signal": [[1, 2, 3, 4, 5], [3, 4, 1, 2, 5]]
      }
    }
  ]
}