Tensorflow Serving Prediction Request Structure

{
  "name": "TensorFlow Serving Prediction Request",
  "description": "Structure documentation for a prediction inference request to TensorFlow Serving REST API.",
  "fields": [
    {
      "name": "signature_name",
      "type": "string",
      "required": false,
      "description": "Optional name of the serving signature to use. Omit to use the default serving signature."
    },
    {
      "name": "instances",
      "type": "array",
      "required": false,
      "description": "Row format input. An array of individual inputs. Use this when all inputs share the same first (0th) tensor dimension.",
      "items": {
        "type": "string | number | object",
        "description": "Single input - scalar, list, or named-tensor object"
      }
    },
    {
      "name": "inputs",
      "type": "object",
      "required": false,
      "description": "Column format input. A map of input tensor names to arrays. Use this when inputs have different 0th dimensions.",
      "additionalProperties": {
        "type": "array",
        "description": "Array of values for a named input tensor"
      }
    }
  ],
  "notes": [
    "Use 'instances' (row format) when all tensor inputs share the same batch dimension",
    "Use 'inputs' (column format) when tensor inputs have different sizes",
    "Only one of 'instances' or 'inputs' should be provided per request",
    "Binary data should be encoded as base64 JSON strings"
  ]
}