Azure OpenAI开发报错：AzureKeyCredential不是构造函数求助

开发可读取作品集的聊天机器人项目，采用Azure OpenAI GPT-3.5模型，已完成API密钥与端点配置。通过route.js文件中的GET请求测试API调用时，终端报错：TypeError: _azure_openai__WEBPACK_IMPORTED_MODULE_0__.AzureKeyCredential is not a constructor，尝试查阅官方文档、修改导入语法等方法后问题仍未解决。

原代码如下：

import { AzureCliCredential } from "@azure/identity";
import { OpenAIClient, AzureKeyCredential } from "@azure/openai";
import { NextResponse } from "next/server";

const endpoint = process.env.AZURE_OPENAI_ENDPOINT;
const apiKey = process.env.AZURE_OPENAI_API_KEY;
const model = process.env.AZURE_OPENAI_MODEL;

export async function GET(req) {
  const client = new OpenAIClient(endpoint, new AzureKeyCredential(apiKey));

  const messages = [
    {
      role: "system",
      content: "You are a helpful assistant.",
    },
    {
      role: "user",
      content: "Why is python such a popular coding language?",
    },
  ];

  const response = await client.getChatCompletions(model, messages, {
    maxTokens: 128,
  });

  return NextResponse.json({
    message: response.choices[0].message.content,
  });
}

解决方案 #

• 版本适配问题：@azure/openai v4及以上版本中，AzureKeyCredential已从该包迁移至@azure/core-auth包，原导入路径不再有效。
• 调整导入语句：将AzureKeyCredential的导入源改为@azure/core-auth，保留OpenAIClient从@azure/openai导入。
• 安装依赖：若项目未安装@azure/core-auth，执行以下命令安装：

  npm install @azure/core-auth
  # 或使用yarn
  yarn add @azure/core-auth
  

修改后的完整代码：

import { AzureCliCredential } from "@azure/identity";
import { OpenAIClient } from "@azure/openai";
import { AzureKeyCredential } from "@azure/core-auth";
import { NextResponse } from "next/server";

const endpoint = process.env.AZURE_OPENAI_ENDPOINT;
const apiKey = process.env.AZURE_OPENAI_API_KEY;
const model = process.env.AZURE_OPENAI_MODEL;

export async function GET(req) {
  const client = new OpenAIClient(endpoint, new AzureKeyCredential(apiKey));

  const messages = [
    {
      role: "system",
      content: "You are a helpful assistant.",
    },
    {
      role: "user",
      content: "Why is python such a popular coding language?",
    },
  ];

  const response = await client.getChatCompletions(model, messages, {
    maxTokens: 128,
  });

  return NextResponse.json({
    message: response.choices[0].message.content,
  });
}

内容的提问来源于stack exchange，提问作者Ryan goh