大语言
结构化输出（Structured Outputs）
开始使用
大语言
结构化输出（Structured Outputs）
使用场景
Structured Outputs 功能使得模型能够生成符合您提供的 JSON Schema 的响应，让生成结果更加可控，易于解析。这一功能既可以方便后续逻辑的解析和处理，也有利于将结果集成到业务系统中，适用于各种自动化和数据处理场景。
支持的模型
以下模型支持结构化输出：
使用方法
在请求中添加以下信息：
-
设置参数：通过
response_format参数指定您定义的 JSON Schema。 -
提示词指引：在提示词中指引模型进行结构化输出。
使用示例
下文提供了完整的 Python 代码示例，演示如何使用 Structured Outputs 功能生成符合您提供的 JSON Schema 的 JSON 响应。
1. 初始化客户端
您需要使用您的派欧算力云 API 密钥初始化客户端。
from openai import OpenAI
client = OpenAI(
base_url="https://api.ppinfra.com/v3/openai",
api_key="<Your API Key>",
)
model = "qwen/qwen-2.5-72b-instruct"
2. 定义 JSON Schema
您需要定义 JSON Schema。以下示例创建了一个从用户输入中提取费用信息的 JSON Schema。
# 定义用于费用跟踪的系统提示。
system_prompt = """You are an expense tracking assistant.
Extract expense information from the user's input and format it according to the provided schema."""
# 定义用于结构化响应的 JSON Schema。
response_format = {
"type": "json_schema",
"json_schema": {
"name": "expense_tracking_schema",
"schema": {
"type": "object",
"properties": {
"expenses": {
"type": "array",
"items": {
"type": "object",
"properties": {
"description": {
"type": "string",
"description": "Description of the expense"
},
"amount": {
"type": "number",
"description": "Amount spent in dollars"
},
"date": {
"type": "string",
"description": "When the expense occurred"
},
"category": {
"type": "string",
"description": "Category of expense (e.g., food, office, travel)"
}
},
"required": [
"description",
"amount"
]
}
},
"total": {
"type": "number",
"description": "Total amount of all expenses"
}
},
"required": [
"expenses",
"total"
],
},
},
}
3. 发起 API 请求
创建 API 请求。此请求包括 response_format 参数，指定了上一步中定义的 JSON schema。
chat_completion = client.chat.completions.create(
model=model,
messages=[
{
"role": "system",
"content": system_prompt,
},
{
"role": "user",
"content": """I spent $120 on dinner at an Italian restaurant last Friday with my colleagues.
Also bought office supplies for $45 on Monday.""",
},
],
max_tokens=1024,
temperature=0.8,
stream=False,
response_format=response_format,
)
response_content = chat_completion.choices[0].message.content
# 解析并美化 JSON
try:
json_response = json.loads(response_content)
prettified_json = json.dumps(json_response, indent=2)
print(prettified_json)
except json.JSONDecodeError:
print("Could not parse response as JSON. Raw response:")
print(response_content)
输出：
{
"expenses": [
{
"date": "2023-03-17",
"description": "Dinner at Italian restaurant",
"amount": 120,
"category": "Food & Dining"
},
{
"date": "2023-03-13",
"description": "Office supplies",
"amount": 45,
"category": "Office Supplies"
}
],
"total": 165
}
完整代码
from openai import OpenAI
import json
client = OpenAI(
base_url="https://api.ppinfra.com/v3/openai",
api_key="<Your API Key>",
)
model = "qwen/qwen-2.5-72b-instruct"
# 使用 JSON Schema 进行结构化输出的示例
# 此示例创建了一个用于提取费用信息的 schema
# 定义用于费用跟踪的系统提示
system_prompt = """You are an expense tracking assistant.
Extract expense information from the user's input and format it according to the provided schema."""
# 定义用于结构化响应的 JSON schema
response_format = {
"type": "json_schema",
"json_schema": {
"name": "expense_tracking_schema",
"schema": {
"type": "object",
"properties": {
"expenses": {
"type": "array",
"items": {
"type": "object",
"properties": {
"description": {
"type": "string",
"description": "Description of the expense"
},
"amount": {
"type": "number",
"description": "Amount spent in dollars"
},
"date": {
"type": "string",
"description": "When the expense occurred"
},
"category": {
"type": "string",
"description": "Category of expense (e.g., food, office, travel)"
}
},
"required": [
"description",
"amount"
]
}
},
"total": {
"type": "number",
"description": "Total amount of all expenses"
}
},
"required": [
"expenses",
"total"
],
},
},
}
chat_completion = client.chat.completions.create(
model=model,
messages=[
{
"role": "system",
"content": system_prompt,
},
{
"role": "user",
"content": """I spent $120 on dinner at an Italian restaurant last Friday with my colleagues.
Also bought office supplies for $45 on Monday.""",
},
],
max_tokens=1024,
temperature=0.8,
stream=False,
response_format=response_format,
)
response_content = chat_completion.choices[0].message.content
# 解析并美化 JSON
try:
json_response = json.loads(response_content)
prettified_json = json.dumps(json_response, indent=2)
print(prettified_json)
except json.JSONDecodeError:
print("Could not parse response as JSON. Raw response:")
print(response_content)