關(guān)于使用ChatOpenAI調(diào)用阿里云千問(wèn)模型無(wú)法實(shí)現(xiàn)結(jié)構(gòu)化輸出的解決方案

發(fā)布于 2025-11-6 07:30

瀏覽

0收藏

阿里云提供的 “OpenAI 兼容接口”主要是為了支持標(biāo)準(zhǔn) OpenAI API 調(diào)用。

但是并不是完全兼容：尤其是當(dāng)我們使用下面的方式期待模型產(chǎn)生結(jié)構(gòu)化輸出時(shí)：

from pydantic import BaseModel, Field


class Movie(BaseModel):
    """A movie with details."""
    title: str = Field(..., description="The title of the movie")
    year: int = Field(..., description="The year the movie was released")
    director: str = Field(..., description="The director of the movie")
    rating: float = Field(..., description="The movie's rating out of 10")


model_with_structure = model.with_structured_output(Movie)
response = model_with_structure.invoke("Provide details about the movie Inception")
print(response)  # Movie(title="Inception", year=2010, director="Christopher Nolan", rating=8.8)

會(huì)出現(xiàn)如下 400 錯(cuò)誤：

BadRequestError: Error code: 400 - {'error': {'code': 'invalid_parameter_error', 'param': None, 'message': "<400> InternalError.Algo.InvalidParameter: 'messages' must contain the word 'json' in some form, to use 'response_format' of type 'json_object'.", 'type': 'invalid_request_error'}, 'id': 'chatcmpl-9b2223da-21bb-4299-b46a-efaf01e38fb3', 'request_id': '9b2223da-21bb-4299-b46a-efaf01e38fb3'}

這個(gè)問(wèn)題主要是阿里云限制了用戶(hù)期望模型產(chǎn)生結(jié)構(gòu)化輸出時(shí)要在提示詞中包含 JSON 字樣。

當(dāng)然我們可以安裝 0.2.14 版本的 langchain-openai 來(lái)規(guī)避這種問(wèn)題的產(chǎn)生，但是當(dāng) langchain 更新到1.0版本之后，0.2.14 版本的 langchain-openai 依賴(lài)的 langchain-core 就開(kāi)始與 1.0 版本的langchain不兼容了。

所以為了大家能夠盡快全面升級(jí)到 1.0 版本的 langchain，我試驗(yàn)出了如下方法來(lái)暫時(shí)解決結(jié)構(gòu)化輸出的問(wèn)題，本質(zhì)上還是依賴(lài)于提示詞工程。

from langchain_core.output_parsers import JsonOutputParser
from pydantic import BaseModel, Field




class Movie(BaseModel):
    """A movie with details."""
    title: str = Field(..., description="The title of the movie")
    year: int = Field(..., description="The year the movie was released")
    director: str = Field(..., description="The director of the movie")
    rating: float = Field(..., description="The movie's rating out of 10")




structured_output_parser = JsonOutputParser(pydantic_object=Movie)
structured_output_parser_chain = model | structured_output_parser
response = model.invoke(rewrite_question_string_parser.get_format_instructions() + "Provide details about the movie Inception")

本文轉(zhuǎn)載自??PyTorch研習(xí)社??，作者：南七無(wú)名式

標(biāo)簽

ChatOpenAI

阿里云

千問(wèn)模型

已于2025-11-6 07:30:23修改

贊

回復(fù)