Salidas estructuradas

Artículo
12/18/2024

Las salidas estructuradas hacen que un modelo siga una definición de esquema JSON que usted proporciona como parte de la llamada API de inferencia. Esto contrasta con la característica del modo JSON anterior, que garantizaba que se generaría un JSON válido, pero no podía garantizar un cumplimiento estricto del esquema proporcionado. Las salidas estructuradas se recomiendan para las llamadas a funciones, la extracción de datos estructurados y la creación de flujos de trabajo complejos de varios pasos.

Nota:

Actualmente, las salidas estructuradas no se admiten en el escenario Traiga sus propios datos.

Modelos admitidos

o1 versión: 2024-12-17
gpt-4o-mini versión: 2024-07-18
gpt-4o versión: 2024-08-06

Compatibilidad con API

La compatibilidad con salidas estructuradas se agregó por primera vez en la API versión 2024-08-01-preview. Está disponible en las API de versión preliminar más recientes, así como en la API de disponibilidad general más reciente: 2024-10-21.

Introducción

Puede usar Pydantic para definir esquemas de objetos en Python. En función de la versión de OpenAI y las bibliotecas Pydantic que ejecute, es posible que tenga que actualizar a una versión más reciente. Estos ejemplos se probaron con openai 1.42.0 y pydantic 2.8.2.

pip install openai pydantic --upgrade

Si no está familiarizado con el uso de Microsoft Entra ID para la autenticación, consulte Configuración de Azure OpenAI Service con autenticación de Microsoft Entra ID.

from pydantic import BaseModel
from openai import AzureOpenAI
from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"
)

client = AzureOpenAI(
  azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
  azure_ad_token_provider=token_provider,
  api_version="2024-10-21"
)


class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]

completion = client.beta.chat.completions.parse(
    model="MODEL_DEPLOYMENT_NAME", # replace with the model deployment name of your gpt-4o 2024-08-06 deployment
    messages=[
        {"role": "system", "content": "Extract the event information."},
        {"role": "user", "content": "Alice and Bob are going to a science fair on Friday."},
    ],
    response_format=CalendarEvent,
)

event = completion.choices[0].message.parsed

print(event)
print(completion.model_dump_json(indent=2))

Output

name='Science Fair' date='Friday' participants=['Alice', 'Bob']
{
  "id": "chatcmpl-A1EUP2fAmL4SeB1lVMinwM7I2vcqG",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "message": {
        "content": "{\n  \"name\": \"Science Fair\",\n  \"date\": \"Friday\",\n  \"participants\": [\"Alice\", \"Bob\"]\n}",
        "refusal": null,
        "role": "assistant",
        "function_call": null,
        "tool_calls": [],
        "parsed": {
          "name": "Science Fair",
          "date": "Friday",
          "participants": [
            "Alice",
            "Bob"
          ]
        }
      }
    }
  ],
  "created": 1724857389,
  "model": "gpt-4o-2024-08-06",
  "object": "chat.completion",
  "service_tier": null,
  "system_fingerprint": "fp_1c2eaec9fe",
  "usage": {
    "completion_tokens": 27,
    "prompt_tokens": 32,
    "total_tokens": 59
  }
}

pip install openai pydantic --upgrade

from pydantic import BaseModel
from openai import AzureOpenAI

client = AzureOpenAI(
  azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
  api_key=os.getenv("AZURE_OPENAI_API_KEY"),  
  api_version="2024-10-21"
)


class CalendarEvent(BaseModel):
    name: str
    date: str
    participants: list[str]

completion = client.beta.chat.completions.parse(
    model="MODEL_DEPLOYMENT_NAME", # replace with the model deployment name of your gpt-4o 2024-08-06 deployment
    messages=[
        {"role": "system", "content": "Extract the event information."},
        {"role": "user", "content": "Alice and Bob are going to a science fair on Friday."},
    ],
    response_format=CalendarEvent,
)

event = completion.choices[0].message.parsed

print(event)
print(completion.model_dump_json(indent=2))

Output

name='Science Fair' date='Friday' participants=['Alice', 'Bob']
{
  "id": "chatcmpl-A1EUP2fAmL4SeB1lVMinwM7I2vcqG",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null,
      "message": {
        "content": "{\n  \"name\": \"Science Fair\",\n  \"date\": \"Friday\",\n  \"participants\": [\"Alice\", \"Bob\"]\n}",
        "refusal": null,
        "role": "assistant",
        "function_call": null,
        "tool_calls": [],
        "parsed": {
          "name": "Science Fair",
          "date": "Friday",
          "participants": [
            "Alice",
            "Bob"
          ]
        }
      }
    }
  ],
  "created": 1724857389,
  "model": "gpt-4o-2024-08-06",
  "object": "chat.completion",
  "service_tier": null,
  "system_fingerprint": "fp_1c2eaec9fe",
  "usage": {
    "completion_tokens": 27,
    "prompt_tokens": 32,
    "total_tokens": 59
  }
}

response_format se establece en json_schema con strict: true establecido.

curl -X POST  https://YOUR_RESOURCE_NAME.openai.azure.com/openai/deployments/YOUR_MODEL_DEPLOYMENT_NAME/chat/completions?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
    -d '{
        "messages": [
                {"role": "system", "content": "Extract the event information."},
                {"role": "user", "content": "Alice and Bob are going to a science fair on Friday."}
            ],
            "response_format": {
                "type": "json_schema",
                "json_schema": {
                    "name": "CalendarEventResponse",
                    "strict": true,
                    "schema": {
                        "type": "object",
                        "properties": {
                            "name": {
                              "type": "string"
                            },
                            "date": {
                                "type": "string"
                            },
                            "participants": {
                                "type": "array",
                                "items": {
                                    "type": "string"
                                }
                            }
                        },
                        "required": [
                            "name",
                            "date",
                            "participants"
                        ],
                        "additionalProperties": false
                    }
                }
          }
  }'

Salida:

{
  "id": "chatcmpl-A1HKsHAe2hH9MEooYslRn9UmEwsag",
  "object": "chat.completion",
  "created": 1724868330,
  "model": "gpt-4o-2024-08-06",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "{\n  \"name\": \"Science Fair\",\n  \"date\": \"Friday\",\n  \"participants\": [\"Alice\", \"Bob\"]\n}"
      },
      "logprobs": null,
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 33,
    "completion_tokens": 27,
    "total_tokens": 60
  },
  "system_fingerprint": "fp_1c2eaec9fe"
}

Llamada a funciones con salidas estructuradas

Las salidas estructuradas para las llamadas a funciones se pueden habilitar con un único parámetro, proporcionando strict: true.

Nota:

Las salidas estructuradas no se admiten con llamadas a funciones paralelas. Cuando se usan salidas estructuradas, establezca parallel_tool_calls en false.

from enum import Enum
from typing import Union
from pydantic import BaseModel
import openai
from openai import AzureOpenAI

client = AzureOpenAI(
  azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
  api_key=os.getenv("AZURE_OPENAI_API_KEY"),  
  api_version="2024-10-21"
)


class GetDeliveryDate(BaseModel):
    order_id: str

tools = [openai.pydantic_function_tool(GetDeliveryDate)]

messages = []
messages.append({"role": "system", "content": "You are a helpful customer support assistant. Use the supplied tools to assist the user."})
messages.append({"role": "user", "content": "Hi, can you tell me the delivery date for my order #12345?"}) 

response = client.chat.completions.create(
    model="MODEL_DEPLOYMENT_NAME", # replace with the model deployment name of your gpt-4o 2024-08-06 deployment
    messages=messages,
    tools=tools
)

print(response.choices[0].message.tool_calls[0].function)
print(response.model_dump_json(indent=2))

from enum import Enum
from typing import Union
from pydantic import BaseModel
import openai
from openai import AzureOpenAI

client = AzureOpenAI(
  azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
  api_key=os.getenv("AZURE_OPENAI_API_KEY"),  
  api_version="2024-10-21"
)

class GetDeliveryDate(BaseModel):
    order_id: str

tools = [openai.pydantic_function_tool(GetDeliveryDate)]

messages = []
messages.append({"role": "system", "content": "You are a helpful customer support assistant. Use the supplied tools to assist the user."})
messages.append({"role": "user", "content": "Hi, can you tell me the delivery date for my order #12345?"}) 

response = client.chat.completions.create(
    model="MODEL_DEPLOYMENT_NAME", # replace with the model deployment name of your gpt-4o 2024-08-06 deployment
    messages=messages,
    tools=tools
)

print(response.choices[0].message.tool_calls[0].function)
print(response.model_dump_json(indent=2))

curl -X POST  https://YOUR_RESOURCE_NAME.openai.azure.com/openai/deployments/YOUR_MODEL_DEPLOYMENT_NAME/chat/completions?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant. The current date is August 6, 2024. You help users query for the data they are looking for by calling the query function."
    },
    {
      "role": "user",
      "content": "look up all my orders in may of last year that were fulfilled but not delivered on time"
    }
  ],
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "query",
        "description": "Execute a query.",
        "strict": true,
        "parameters": {
          "type": "object",
          "properties": {
            "table_name": {
              "type": "string",
              "enum": ["orders"]
            },
            "columns": {
              "type": "array",
              "items": {
                "type": "string",
                "enum": [
                  "id",
                  "status",
                  "expected_delivery_date",
                  "delivered_at",
                  "shipped_at",
                  "ordered_at",
                  "canceled_at"
                ]
              }
            },
            "conditions": {
              "type": "array",
              "items": {
                "type": "object",
                "properties": {
                  "column": {
                    "type": "string"
                  },
                  "operator": {
                    "type": "string",
                    "enum": ["=", ">", "<", ">=", "<=", "!="]
                  },
                  "value": {
                    "anyOf": [
                      {
                        "type": "string"
                      },
                      {
                        "type": "number"
                      },
                      {
                        "type": "object",
                        "properties": {
                          "column_name": {
                            "type": "string"
                          }
                        },
                        "required": ["column_name"],
                        "additionalProperties": false
                      }
                    ]
                  }
                },
                "required": ["column", "operator", "value"],
                "additionalProperties": false
              }
            },
            "order_by": {
              "type": "string",
              "enum": ["asc", "desc"]
            }
          },
          "required": ["table_name", "columns", "conditions", "order_by"],
          "additionalProperties": false
        }
      }
    }
  ]
}'

Esquemas y limitaciones admitidas

Las salidas estructuradas de Azure OpenAI admiten el mismo subconjunto del esquema JSONque OpenAI.

Tipos admitidos

Cadena
Number
Booleano
Entero
Object
Array
Enum
anyOf

Nota:

Los objetos raíz no pueden ser de tipo anyOf.

Todos los campos deben ser obligatorios

Todos los campos o parámetros de función deben incluirse según sea necesario. En el ejemplo siguiente, location y unit se especifican en "required": ["location", "unit"].

{
    "name": "get_weather",
    "description": "Fetches the weather in the given location",
    "strict": true,
    "parameters": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The location to get the weather for"
            },
            "unit": {
                "type": "string",
                "description": "The unit to return the temperature in",
                "enum": ["F", "C"]
            }
        },
        "additionalProperties": false,
        "required": ["location", "unit"]
    }

Si es necesario, es posible emular un parámetro opcional mediante un tipo de unión con null. En este ejemplo, esto se logra con la línea "type": ["string", "null"],.

{
    "name": "get_weather",
    "description": "Fetches the weather in the given location",
    "strict": true,
    "parameters": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The location to get the weather for"
            },
            "unit": {
                "type": ["string", "null"],
                "description": "The unit to return the temperature in",
                "enum": ["F", "C"]
            }
        },
        "additionalProperties": false,
        "required": [
            "location", "unit"
        ]
    }
}

Profundidad de anidamiento

Un esquema puede tener hasta 100 propiedades de objeto en total, con hasta cinco niveles de anidamiento

additionalProperties: false debe establecerse siempre en objetos

Esta propiedad controla si un objeto puede tener pares de valor de clave adicionales que no se definieron en el esquema JSON. Para usar salidas estructuradas, debe establecer este valor en false.

Ordenación de claves

Las salidas estructuradas se ordenan igual que el esquema proporcionado. Para cambiar el orden de salida, modifique el orden del esquema que envía como parte de la solicitud de inferencia.

Palabras clave específicas del tipo no admitidas

Tipo	Palabra clave no admitida
Cadena	minlength maxLength patrón format
Number	minimum Máximo multipleOf
Objetos	patternProperties unevaluatedProperties propertyNames minProperties maxProperties
Matrices	unevaluatedItems contains minContains maxContains minItems maxItems uniqueItems

Los esquemas anidados que usan anyOf deben cumplir con el subconjunto de esquema JSON general

Esquema anyOf admitido de ejemplo:

{
	"type": "object",
	"properties": {
		"item": {
			"anyOf": [
				{
					"type": "object",
					"description": "The user object to insert into the database",
					"properties": {
						"name": {
							"type": "string",
							"description": "The name of the user"
						},
						"age": {
							"type": "number",
							"description": "The age of the user"
						}
					},
					"additionalProperties": false,
					"required": [
						"name",
						"age"
					]
				},
				{
					"type": "object",
					"description": "The address object to insert into the database",
					"properties": {
						"number": {
							"type": "string",
							"description": "The number of the address. Eg. for 123 main st, this would be 123"
						},
						"street": {
							"type": "string",
							"description": "The street name. Eg. for 123 main st, this would be main st"
						},
						"city": {
							"type": "string",
							"description": "The city of the address"
						}
					},
					"additionalProperties": false,
					"required": [
						"number",
						"street",
						"city"
					]
				}
			]
		}
	},
	"additionalProperties": false,
	"required": [
		"item"
	]
}

Se admiten definiciones

Ejemplo admitido:

{
	"type": "object",
	"properties": {
		"steps": {
			"type": "array",
			"items": {
				"$ref": "#/$defs/step"
			}
		},
		"final_answer": {
			"type": "string"
		}
	},
	"$defs": {
		"step": {
			"type": "object",
			"properties": {
				"explanation": {
					"type": "string"
				},
				"output": {
					"type": "string"
				}
			},
			"required": [
				"explanation",
				"output"
			],
			"additionalProperties": false
		}
	},
	"required": [
		"steps",
		"final_answer"
	],
	"additionalProperties": false
}

Se admiten esquemas recursivos

Ejemplo de uso de # para recursividad raíz:

{
        "name": "ui",
        "description": "Dynamically generated UI",
        "strict": true,
        "schema": {
            "type": "object",
            "properties": {
                "type": {
                    "type": "string",
                    "description": "The type of the UI component",
                    "enum": ["div", "button", "header", "section", "field", "form"]
                },
                "label": {
                    "type": "string",
                    "description": "The label of the UI component, used for buttons or form fields"
                },
                "children": {
                    "type": "array",
                    "description": "Nested UI components",
                    "items": {
                        "$ref": "#"
                    }
                },
                "attributes": {
                    "type": "array",
                    "description": "Arbitrary attributes for the UI component, suitable for any element",
                    "items": {
                        "type": "object",
                        "properties": {
                            "name": {
                                "type": "string",
                                "description": "The name of the attribute, for example onClick or className"
                            },
                            "value": {
                                "type": "string",
                                "description": "The value of the attribute"
                            }
                        },
                      "additionalProperties": false,
                      "required": ["name", "value"]
                    }
                }
            },
            "required": ["type", "label", "children", "attributes"],
            "additionalProperties": false
        }
    }

Ejemplo de recursividad explícita:

{
	"type": "object",
	"properties": {
		"linked_list": {
			"$ref": "#/$defs/linked_list_node"
		}
	},
	"$defs": {
		"linked_list_node": {
			"type": "object",
			"properties": {
				"value": {
					"type": "number"
				},
				"next": {
					"anyOf": [
						{
							"$ref": "#/$defs/linked_list_node"
						},
						{
							"type": "null"
						}
					]
				}
			},
			"additionalProperties": false,
			"required": [
				"next",
				"value"
			]
		}
	},
	"additionalProperties": false,
	"required": [
		"linked_list"
	]
}

Compartir vía

Salidas estructuradas

Modelos admitidos

Compatibilidad con API

Introducción

Output

Output

Llamada a funciones con salidas estructuradas

Esquemas y limitaciones admitidas

Tipos admitidos

Todos los campos deben ser obligatorios

Profundidad de anidamiento

additionalProperties: false debe establecerse siempre en objetos

Ordenación de claves

Palabras clave específicas del tipo no admitidas

Los esquemas anidados que usan anyOf deben cumplir con el subconjunto de esquema JSON general

Se admiten definiciones

Se admiten esquemas recursivos

Comentarios

Recursos adicionales