如何让通义千问大模型支持结构化输出？

之前的文章提到通义千问API无法通过with_structured_output/json schema的方式支持结构化输出，如果就是想使用通义千问大模型做结构化输出，应该怎么办呢？有两种办法

使用Ollama来运行通义千问大模型

从Ollama博客文章 Structured output 中了解到，Ollama已经支持结构化输出了，这个功能是在Ollama 0.5.0版本 引入的。通过Ollama把qwen3大模型在本地运行起来，使用下面的代码就能看到效果。

from langchain.chat_models import init_chat_model  
from langchain_core.prompts import ChatPromptTemplate  
from pydantic import BaseModel, Field  
  
llm = init_chat_model(  
    model_provider="openai",  
    model="qwen3:8b",  
    base_url="http://localhost:11434/v1",  
    api_key="123456"  
)  
  
tagging_prompt = ChatPromptTemplate.from_template(  
    """    
        Extract the desired information from the following passage.  
        Only extract the properties mentioned in the 'Classification' function.        Passage:        {input}    """)  
  
class Classification(BaseModel):  
    sentiment: str = Field(description="The sentiment of the text")  
    aggressiveness: int = Field(description="How aggressive the text is on a scale from 1 to 10")  
    language: str = Field(description="The language the text is written in")  
  
structured_llm = llm.with_structured_output(Classification)  
  
inp = "Estoy increiblemente contento de haberte conocido! Creo que seremos muy buenos amigos!"  
prompt = tagging_prompt.invoke({"input": inp})  
response = structured_llm.invoke(prompt)  
print(response)

程序能够正常运行并打印出Classification这个结构化的类对象内容（如下）

sentiment='positive' aggressiveness=0 language='Spanish'

如果把代码中llm对象构造时的参数换回通义千问API的值，也即是改为

llm = init_chat_model(  
    model_provider="openai",  
    model="qwen-plus-latest",  
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",  
    api_key=os.environ["QWEN_API_KEY"]
)

会发现之前提到的错误再次出现了，这证明不是通义千问大模型不支持结构化输出，而是通义千问API服务不支持。

如果不想使用Ollama本地运行大模型，而是想使用通义千问API服务支持结构化输出，该怎么办呢？

使用 PydanticOutputParser 自动生成提示词

要使用结构化输出，通义千问文档里提到

您可以在提示词中明确描述所需 JSON 的键值结构和数据类型，并提供标准数据样例，这会帮助大模型达到类似效果。

借助 LangChain 的 PydanticOutputParser 可以帮我们自动生成数据结构的提示词。参考下面的代码

import os  
from langchain_core.output_parsers import PydanticOutputParser  
from langchain_core.prompts import ChatPromptTemplate  
from pydantic import BaseModel, Field  
from langchain.chat_models import init_chat_model  
  
llm = init_chat_model(  
    model_provider="openai",  
    model="qwen-plus-latest",  
    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",  
    api_key=os.environ["QWEN_API_KEY"]  
)  
  
class Classification(BaseModel):  
    sentiment: str = Field(description="The sentiment of the text")  
    aggressiveness: int = Field(description="How aggressive the text is on a scale from 1 to 10")  
    language: str = Field(description="The language the text is written in")  
  
parser = PydanticOutputParser(pydantic_object=Classification)  
  
tagging_prompt = ChatPromptTemplate.from_messages([  
    ("system",  
     "Extract the desired information into JSON format from the following passage."  
    "{format_instructions}"     ),  
    ("human", "{input}")  
]).partial(format_instructions=parser.get_format_instructions())  
  
inp = "Estoy increiblemente contento de haberte conocido! Creo que seremos muy buenos amigos!"  
print(tagging_prompt.invoke({"input": inp}))  
  
chain = tagging_prompt | llm | parser  
response = chain.invoke({"input": inp})  
print(response)

第30行代码会打印出格式化后的提示词，第一行后面的内容都是API根据提供的类结构生成的提示词。

    Extract the desired information into JSON format from the following passage.
    
    The output should be formatted as a JSON instance that conforms to the JSON schema below.
    As an example, for the schema {"properties": {"foo": {"title": "Foo", "description": "a list of strings", "type": "array", "items": {"type": "string"}}}, "required": ["foo"]}
    the object {"foo": ["bar", "baz"]} is a well-formatted instance of the schema. The object {"properties": {"foo": ["bar", "baz"]}} is not well-formatted.
    Here is the output schema:
    {
      "properties": {
        "sentiment": {
          "description": "The sentiment of the text",
          "title": "Sentiment",
          "type": "string"
        },
        "aggressiveness": {
          "description": "How aggressive the text is on a scale from 1 to 10",
          "title": "Aggressiveness",
          "type": "integer"
        },
        "language": {
          "description": "The language the text is written in",
          "title": "Language",
          "type": "string"
        }
      },
      "required": [
        "sentiment",
        "aggressiveness",
        "language"
      ]
    }

真是太方便了，这下可以高效的从文档里提取结构化数据啦。

思考

从 Ollama 0.5.0的代码修改记录 来看，Ollama只用了很少的代码修改就支持了结构化输出。那么通义千问API服务什么时候能支持起来呢？

本文由博客一文多发平台 OpenWrite 发布！