前言

昨天根据 Azure OpenAI Service 注册申请与配置心得 这篇 Blog 成功注册了 Azure OpenAI Service 。

注册过程大家看上面👆的 Blog 文章就好，这里我仅仅就我踩过的坑来做个说明。

⚠️注意

1. Azure OpenAI Service 目前需要以企业的身份来申请试用。

   Azure OpenAI Service 接受来自中国的企业申请，请依照表单如实填写信息，其中最重要的是：

   Your Company Email Address：请填写你在你公司的企业邮箱地址。

   经过我的试验使用自定义域名结尾的邮箱可以作为申请，使用个人邮箱如 gmail/hotmail/QQ/outlook 等。基本都是不给过的。

   Company Website：请填写你所在公司的网站的网址。网址里最好能展示一个与你企业邮箱地址同一个域名的邮箱地址。**你的企业邮箱的域名、公司网站的域名需要保持一致，域名的主体需要与公司的名称保持一致，否则会在审核时被拒绝。**

2. 在使用 Cloudflare 转换 Azure OpenAI Service API 时需要使用 cf-openai-azure-proxy 这个项目。

   其中

   

   • 直接修改他们的值, 如:

   

   注意以上修改代码值的时候请务必加上 ' ' 我一开始没加，一直报错。

   或者通过 cloudflare worker 控制台, 进入 Workers script > Settings > Add variable under Environment Variables.

   

   附上完整代码：

   // The name of your Azure OpenAI Resource.
   const resourceName=RESOURCE_NAME
   
   // The deployment name you chose when you deployed the model.
   const mapper = {
    'gpt-3.5-turbo': DEPLOY_NAME_GPT35,
    'gpt-4': DEPLOY_NAME_GPT4
   };
   
   const apiVersion=2023-05-15
   
   addEventListener(fetch, (event) => {
   event.respondWith(handleRequest(event.request));
   });
   
   async function handleRequest(request) {
   if (request.method === 'OPTIONS') {
    return handleOPTIONS(request)
   }
   
   const url = new URL(request.url);
   if (url.pathname === '/v1/chat/completions') {
    var path=chat/completions
   } else if (url.pathname === '/v1/completions') {
    var path=completions
   } else if (url.pathname === '/v1/models') {
    return handleModels(request)
   } else {
    return new Response('404 Not Found', { status: 404 })
   }
   
   let body;
   if (request.method === 'POST') {
    body = await request.json();
   }
   
   const modelName = body?.model;  
   const deployName = mapper[modelName] || '' 
   
   if (deployName === '') {
    return new Response('Missing model mapper', {
        status: 403
    });
   }
   const fetchAPI = <code>https://${resourceName}.openai.azure.com/openai/deployments/${deployName}/${path}?api-version=${apiVersion}</code>
   
   const authKey = request.headers.get('Authorization');
   if (!authKey) {
    return new Response(Not allowed, {
      status: 403
    });
   }
   
   const payload = {
    method: request.method,
    headers: {
      Content-Type: application/json,
      api-key: authKey.replace('Bearer ', ''),
    },
    body: typeof body === 'object' ? JSON.stringify(body) : '{}',
   };
   
   let response = await fetch(fetchAPI, payload);
   response = new Response(response.body, response);
   response.headers.set(Access-Control-Allow-Origin, *);
   
   if (body?.stream != true){
    return response
   } 
   
   let { readable, writable } = new TransformStream()
   stream(response.body, writable);
   return new Response(readable, response);
   
   }
   
   function sleep(ms) {
   return new Promise(resolve => setTimeout(resolve, ms));
   }
   
   // support printer mode and add newline
   async function stream(readable, writable) {
   const reader = readable.getReader();
   const writer = writable.getWriter();
   
   // const decoder = new TextDecoder();
   const encoder = new TextEncoder();
   const decoder = new TextDecoder();
   // let decodedValue = decoder.decode(value);
   const newline = n;
   const delimiter = nn
   const encodedNewline = encoder.encode(newline);
   
   let buffer = ;
   while (true) {
    let { value, done } = await reader.read();
    if (done) {
      break;
    }
    buffer += decoder.decode(value, { stream: true }); // stream: true is important here,fix the bug of incomplete line
    let lines = buffer.split(delimiter);
   
    // Loop through all but the last line, which may be incomplete.
    for (let i = 0; i < lines.length - 1; i++) {
      await writer.write(encoder.encode(lines[i] + delimiter));
      await sleep(20);
    }
   
    buffer = lines[lines.length - 1];
   }
   
   if (buffer) {
    await writer.write(encoder.encode(buffer));
   }
   await writer.write(encodedNewline)
   await writer.close();
   }
   
   async function handleModels(request) {
   const data = {
    object: list,
    data: []  
   };
   
   for (let key in mapper) {
    data.data.push({
      id: key,
      object: model,
      created: 1677610602,
      owned_by: openai,
      permission: [{
        id: modelperm-M56FXnG1AsIr3SXq8BYPvXJA,
        object: model_permission,
        created: 1679602088,
        allow_create_engine: false,
        allow_sampling: true,
        allow_logprobs: true,
        allow_search_indices: false,
        allow_view: true,
        allow_fine_tuning: false,
        organization: *,
        group: null,
        is_blocking: false
      }],
      root: key,
      parent: null
    });  
   }
   
   const json = JSON.stringify(data, null, 2);
   return new Response(json, {
    headers: { 'Content-Type': 'application/json' },
   });
   }
   
   async function handleOPTIONS(request) {
    return new Response(null, {
      headers: {
        'Access-Control-Allow-Origin': '*',
        'Access-Control-Allow-Methods': '*',
        'Access-Control-Allow-Headers': '*'
      }
    })
   }
   

最后

如果你成功做完了所有的步骤。那么你应该有一个可以取代 OpenAI API 功能的密钥以及 Cloudflare Worker 的 URL 链接了。

使用方法如下：

• 请在设置中选择OpenAI作为Default Service Provider。
• API KEY：请填写Azure OpenAI Service的密钥。
• API URL：请填写Cloudflare Worker的 URL。
• API Model 选择你已部署的 Model，例如gpt-35-turbo (version 0301)，而这个Model需要你在Cloudflare Worker代码最前面的 mapper 中做好映射。