├── .github └── workflows │ └── docker_build_push.yml ├── Dockerfile ├── LICENSE ├── README.md ├── README_en.md ├── cf-openai-azure-proxy.js └── cf-openai-palm-proxy.js /.github/workflows/docker_build_push.yml: -------------------------------------------------------------------------------- 1 | name: Build and Push Docker Image 2 | 3 | on: 4 | push: 5 | tags: 6 | - 'v*' 7 | 8 | jobs: 9 | build_and_push: 10 | runs-on: ubuntu-latest 11 | 12 | steps: 13 | - name: Checkout code 14 | uses: actions/checkout@v2 15 | 16 | - name: Set up Docker Buildx 17 | uses: docker/setup-buildx-action@v1 18 | 19 | - name: Login to DockerHub 20 | uses: docker/login-action@v1 21 | with: 22 | username: ${{ secrets.DOCKERHUB_USERNAME }} 23 | password: ${{ secrets.DOCKERHUB_TOKEN }} 24 | 25 | - name: Build and push Docker image 26 | uses: docker/build-push-action@v2 27 | with: 28 | context: . 29 | push: true 30 | tags: haibbo/cf-openai-azure-proxy:${{ github.ref_name }} 31 | - name: Build and push Docker image 32 | uses: docker/build-push-action@v2 33 | with: 34 | context: . 35 | push: true 36 | tags: | 37 | haibbo/cf-openai-azure-proxy:${{ github.ref_name }} 38 | haibbo/cf-openai-azure-proxy:latest 39 | platforms: linux/amd64,linux/arm64/v8 40 | - name: Create GitHub Release 41 | id: create_release 42 | uses: actions/create-release@v1 43 | env: 44 | GITHUB_TOKEN: ${{ secrets.ACTIONS_RELEASE_TOKEN }} 45 | with: 46 | tag_name: ${{ github.ref }} 47 | release_name: Release ${{ github.ref }} 48 | body: | 49 | Release notes for ${{ github.ref }} 50 | 51 | ```sh 52 | docker run -d -p 8787:8787 -t cf-azure-openai-proxy \ 53 | --env RESOURCE_NAME=codegpt \ 54 | --env DEPLOY_NAME=gpt3 \ 55 | --env DEPLOY_NAME_GPT4=gpt4 \ 56 | haibbo/cf-openai-azure-proxy:latest 57 | ``` 58 | 59 | 60 | Docker image is available at: 61 | - `haibbo/cf-openai-azure-proxy:${{ github.ref_name }}` 62 | - `haibbo/cf-openai-azure-proxy:latest` 63 | 64 | Enjoy! 65 | draft: false 66 | prerelease: false 67 | 68 | 69 | 70 | -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- 1 | FROM node:18.12-slim 2 | 3 | WORKDIR /app 4 | 5 | # 安装 Cloudflare Workers CLI 工具 6 | RUN npm install -g wrangler@2.15.0 7 | 8 | ENV WRANGLER_SEND_METRICS=false 9 | 10 | ENV DEPLOY_NAME_GPT35="" 11 | ENV DEPLOY_NAME_GPT4="" 12 | 13 | # 复制 Workers 脚本到镜像 14 | COPY cf-openai-azure-proxy.js . 15 | 16 | # 启动本地开发服务器 17 | CMD wrangler dev cf-openai-azure-proxy.js --local --var RESOURCE_NAME:$RESOURCE_NAME DEPLOY_NAME_GPT35:$DEPLOY_NAME_GPT35 DEPLOY_NAME_GPT4:$DEPLOY_NAME_GPT4 18 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2023 haibbo 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # cf-openai-azure-proxy 2 | 3 | English | 4 | 中文 5 | 6 | > 大多数 OpenAI 客户端不支持 Azure OpenAI Service，但Azure OpenAI Service的申请和绑卡都非常简单，并且还提供了免费的额度。此脚本使用免费的 Cloudflare Worker 作为代理，使得支持 OpenAI 的客户端可以直接使用 Azure OpenAI Service。 7 | 8 | ### 支持模型: 9 | - GPT-3 10 | - GPT-4 11 | - DALL-E-3 12 | 13 | 模型子类添加非常容易, 参考下面的使用说明 14 | 15 | ### 项目说明: 16 | - 我没有服务器可以使用吗? 17 | - 这段脚本跑在Cloudflare Worker, 不需要服务器, 不需要绑卡, 每天10W次请求 免费 18 | - 我没有自己的域名可以使用吗? 19 | - 也可以, 参考: https://github.com/haibbo/cf-openai-azure-proxy/issues/3 20 | - 实现打印机模式： 21 | - Azure OpenAI Service's 回复是一段一段回复的 22 | - 返回给客户端的时候， 本项目拆出一条条的消息, 依次给， 达到打印机模式 23 | - 项目也支持 Docker 部署（基于 wrangler） 24 | 25 | ### 部署 26 | 代理 OpenAI 的请求到 Azure OpenAI Serivce，代码部署步骤： 27 | 28 | 1. 注册并登录到 Cloudflare 账户 29 | 2. 创建一个新的 Cloudflare Worker 30 | 3. 将 [cf-openai-azure-proxy.js](./cf-openai-azure-proxy.js) 复制并粘贴到 Cloudflare Worker 编辑器中 31 | 4. 通过修改或环境变量调整 resourceName 和 deployment mapper 的值 32 | 5. 保存并部署 Cloudflare Worker 33 | 6. https://github.com/haibbo/cf-openai-azure-proxy/issues/3 **可选**绑定自定义域名: 在 Worker 详情页 -> Trigger -> Custom Domains 中为这个 Worker 添加一个自定义域名 34 | 35 | 36 | ### 使用说明 37 | 38 | 先得到 resourceName 和 deployment mapper, 登录到Azure的后台: 39 | 40 | env 41 | 42 | #### 这里有两种做法: 43 | - 直接修改他们的值, 如: 44 | ```js 45 | // The name of your Azure OpenAI Resource. 46 | const resourceName="codegpt" 47 | 48 | // deployment model mapper 49 | const mapper = { 50 | 'gpt-3.5-turbo': 'gpt3', 51 | 'gpt-4': 'gpt4', 52 | 'dall-e-3': 'dalle3' 53 | }; 54 | 其他的map规则直接按这样的格式续写即可 55 | ``` 56 | - 或者通过 cloudflare worker 控制台, 进入 Workers script > Settings > Add variable under Environment Variables. 57 | 58 | env 59 | 60 | ### 客户端 61 | 以 OpenCat 为例: 自定义 API 域名填写 第六步绑定的域名: 62 | 63 | opencat 64 | 65 | 我已经尝试了多种客户端, 如果遇到其他客户端有问题, 欢迎创建issue. 66 | -------------------------------------------------------------------------------- /README_en.md: -------------------------------------------------------------------------------- 1 | # cf-openai-azure-proxy 2 | 3 | English | 4 | 中文 5 | 6 | > Most OpenAI clients do not support Azure OpenAI Service, but the application for Azure OpenAI Service is very simple, and it also provides free quotas. This script uses a free Cloudflare Worker as a proxy, allowing OpenAI-supported clients to directly use Azure OpenAI Service. 7 | 8 | This script proxies requests to Azure OpenAI Service for OpenAI clients. The code deployment steps are as follows: 9 | 10 | Register and log in to your Cloudflare account. 11 | - Create a new Cloudflare Worker. 12 | - Copy and paste cf-openai-azure-proxy.js into the Cloudflare Worker editor. 13 | - Adjust the values of **resourceName** and deployment **mapper** by either direct modification or using environment variables.. 14 | - Save and deploy the Cloudflare Worker. 15 | - https://github.com/haibbo/cf-openai-azure-proxy/issues/3 Optional: Bind a custom domain name: Add a custom domain name for this worker in the Worker details page -> Trigger -> Custom Domains. 16 | 17 | ## Instructions 18 | First obtain the resourceName and deployment mapper, and log in to the Azure portal: 19 | 20 | env 21 | 22 | #### There are two ways to do this: 23 | - Directly modify their values, such as: 24 | ```js 25 | // The name of your Azure OpenAI Resource. 26 | const resourceName="codegpt" 27 | 28 | const mapper:any = { 29 | 'gpt-3.5-turbo': 'gpt3', 30 | 'gpt-4': 'gpt4' 31 | }; 32 | ``` 33 | Other map rules can be continued directly in this format. 34 | - **OR** go to the Cloudflare Worker console, navigate to Workers script > Settings > Add variable under Environment Variables. 35 | 36 | env 37 | 38 | ## Client 39 | Take OpenCat as an example: fill in the custom API domain name with the domain name bound in step 6: 40 | 41 | opencat 42 | I have tried multiple clients. If you encounter problems with other clients, please feel free to create an issue. 43 | 44 | QA: 45 | 46 | - Do I need a server to use this? 47 | - This script runs on Cloudflare Worker and does not require a server or a bound card. It is free for up to 100,000 requests per day. 48 | - Do I need my own domain name to use this? 49 | - No, it is not necessary. Refer to: https://github.com/haibbo/cf-openai-azure-proxy/issues/3 50 | -------------------------------------------------------------------------------- /cf-openai-azure-proxy.js: -------------------------------------------------------------------------------- 1 | // The name of your Azure OpenAI Resource. 2 | const resourceName=RESOURCE_NAME 3 | 4 | // The deployment name you chose when you deployed the model. 5 | const mapper = { 6 | 'gpt-3.5-turbo': DEPLOY_NAME_GPT35, 7 | 'gpt-3.5-turbo-0613': DEPLOY_NAME_GPT35, 8 | 'gpt-3.5-turbo-1106': DEPLOY_NAME_GPT35, 9 | 'gpt-3.5-turbo-16k': DEPLOY_NAME_GPT35, 10 | 'gpt-4': DEPLOY_NAME_GPT4, 11 | 'gpt-4-0613': DEPLOY_NAME_GPT4, 12 | 'gpt-4-1106-preview': DEPLOY_NAME_GPT4, 13 | 'gpt-4-32k': DEPLOY_NAME_GPT4, 14 | 'dall-e-3': typeof DEPLOY_NAME_DALLE3 !== 'undefined' ? DEPLOY_NAME_DALLE3 : "dalle3", 15 | }; 16 | 17 | const apiVersion="2023-12-01-preview" 18 | 19 | addEventListener("fetch", (event) => { 20 | event.respondWith(handleRequest(event.request)); 21 | }); 22 | 23 | async function handleRequest(request) { 24 | if (request.method === 'OPTIONS') { 25 | return handleOPTIONS(request) 26 | } 27 | 28 | const url = new URL(request.url); 29 | if (url.pathname.startsWith("//")) { 30 | url.pathname = url.pathname.replace('/',"") 31 | } 32 | if (url.pathname === '/v1/chat/completions') { 33 | var path="chat/completions" 34 | } else if (url.pathname === '/v1/images/generations') { 35 | var path="images/generations" 36 | } else if (url.pathname === '/v1/completions') { 37 | var path="completions" 38 | } else if (url.pathname === '/v1/models') { 39 | return handleModels(request) 40 | } else { 41 | return new Response('404 Not Found', { status: 404 }) 42 | } 43 | 44 | let body; 45 | if (request.method === 'POST') { 46 | body = await request.json(); 47 | } 48 | 49 | const modelName = body?.model; 50 | const deployName = mapper[modelName] || '' 51 | 52 | if (deployName === '') { 53 | return new Response('Missing model mapper', { 54 | status: 403 55 | }); 56 | } 57 | const fetchAPI = `https://${resourceName}.openai.azure.com/openai/deployments/${deployName}/${path}?api-version=${apiVersion}` 58 | 59 | const authKey = request.headers.get('Authorization'); 60 | if (!authKey) { 61 | return new Response("Not allowed", { 62 | status: 403 63 | }); 64 | } 65 | 66 | const payload = { 67 | method: request.method, 68 | headers: { 69 | "Content-Type": "application/json", 70 | "api-key": authKey.replace('Bearer ', ''), 71 | }, 72 | body: typeof body === 'object' ? JSON.stringify(body) : '{}', 73 | }; 74 | 75 | let response = await fetch(fetchAPI, payload); 76 | response = new Response(response.body, response); 77 | response.headers.set("Access-Control-Allow-Origin", "*"); 78 | 79 | if (body?.stream != true){ 80 | return response 81 | } 82 | 83 | let { readable, writable } = new TransformStream() 84 | stream(response.body, writable); 85 | return new Response(readable, response); 86 | 87 | } 88 | 89 | function sleep(ms) { 90 | return new Promise(resolve => setTimeout(resolve, ms)); 91 | } 92 | 93 | // support printer mode and add newline 94 | async function stream(readable, writable) { 95 | const reader = readable.getReader(); 96 | const writer = writable.getWriter(); 97 | 98 | // const decoder = new TextDecoder(); 99 | const encoder = new TextEncoder(); 100 | const decoder = new TextDecoder(); 101 | // let decodedValue = decoder.decode(value); 102 | const newline = "\n"; 103 | const delimiter = "\n\n" 104 | const encodedNewline = encoder.encode(newline); 105 | 106 | let buffer = ""; 107 | while (true) { 108 | let { value, done } = await reader.read(); 109 | if (done) { 110 | break; 111 | } 112 | buffer += decoder.decode(value, { stream: true }); // stream: true is important here,fix the bug of incomplete line 113 | let lines = buffer.split(delimiter); 114 | 115 | // Loop through all but the last line, which may be incomplete. 116 | for (let i = 0; i < lines.length - 1; i++) { 117 | await writer.write(encoder.encode(lines[i] + delimiter)); 118 | await sleep(20); 119 | } 120 | 121 | buffer = lines[lines.length - 1]; 122 | } 123 | 124 | if (buffer) { 125 | await writer.write(encoder.encode(buffer)); 126 | } 127 | await writer.write(encodedNewline) 128 | await writer.close(); 129 | } 130 | 131 | async function handleModels(request) { 132 | const data = { 133 | "object": "list", 134 | "data": [] 135 | }; 136 | 137 | for (let key in mapper) { 138 | data.data.push({ 139 | "id": key, 140 | "object": "model", 141 | "created": 1677610602, 142 | "owned_by": "openai", 143 | "permission": [{ 144 | "id": "modelperm-M56FXnG1AsIr3SXq8BYPvXJA", 145 | "object": "model_permission", 146 | "created": 1679602088, 147 | "allow_create_engine": false, 148 | "allow_sampling": true, 149 | "allow_logprobs": true, 150 | "allow_search_indices": false, 151 | "allow_view": true, 152 | "allow_fine_tuning": false, 153 | "organization": "*", 154 | "group": null, 155 | "is_blocking": false 156 | }], 157 | "root": key, 158 | "parent": null 159 | }); 160 | } 161 | 162 | const json = JSON.stringify(data, null, 2); 163 | return new Response(json, { 164 | headers: { 'Content-Type': 'application/json' }, 165 | }); 166 | } 167 | 168 | async function handleOPTIONS(request) { 169 | return new Response(null, { 170 | headers: { 171 | 'Access-Control-Allow-Origin': '*', 172 | 'Access-Control-Allow-Methods': '*', 173 | 'Access-Control-Allow-Headers': '*' 174 | } 175 | }) 176 | } 177 | 178 | -------------------------------------------------------------------------------- /cf-openai-palm-proxy.js: -------------------------------------------------------------------------------- 1 | // The deployment name you chose when you deployed the model. 2 | const chatmodel = 'chat-bison-001'; 3 | const textmodel = 'text-bison-001'; 4 | 5 | addEventListener("fetch", (event) => { 6 | event.respondWith(handleRequest(event.request)); 7 | }); 8 | 9 | async function handleRequest(request) { 10 | if (request.method === 'OPTIONS') { 11 | return handleOPTIONS(request) 12 | } 13 | 14 | const url = new URL(request.url); 15 | if (url.pathname === '/v1/chat/completions') { 16 | var path = "generateMessage" 17 | var deployName = chatmodel; 18 | } else if (url.pathname === '/v1/completions') { 19 | var path = "generateText" 20 | var deployName = textmodel; 21 | } else { 22 | return new Response('404 Not Found', { status: 404 }) 23 | } 24 | 25 | let body; 26 | if (request.method === 'POST') { 27 | body = await request.json(); 28 | } 29 | 30 | const authKey = request.headers.get('Authorization'); 31 | if (!authKey) { 32 | return new Response("Not allowed", { status: 403 }); 33 | } 34 | 35 | // Remove 'Bearer ' from the start of authKey 36 | const apiKey = authKey.replace('Bearer ', ''); 37 | 38 | const fetchAPI = `https://generativelanguage.googleapis.com/v1beta2/models/${deployName}:${path}?key=${apiKey}` 39 | 40 | // Transform request body from OpenAI to PaLM format 41 | const transformedBody = { 42 | temperature: body?.temperature, 43 | candidateCount: body?.n, 44 | topP: body?.top_p, 45 | prompt: { 46 | context: body?.messages?.find(msg => msg.role === 'system')?.content, 47 | messages: body?.messages?.filter(msg => msg.role !== 'system').map(msg => ({ 48 | // author: msg.role === 'user' ? '0' : '1', 49 | content: msg.content, 50 | })), 51 | }, 52 | }; 53 | 54 | const payload = { 55 | method: request.method, 56 | headers: { 57 | "Content-Type": "application/json", 58 | }, 59 | body: JSON.stringify(transformedBody), 60 | }; 61 | 62 | const response = await fetch(fetchAPI, payload); 63 | const palmData = await response.json(); 64 | 65 | // Transform response from PaLM to OpenAI format 66 | const transformedResponse = transformResponse(palmData); 67 | 68 | if (body?.stream != true){ 69 | return new Response(JSON.stringify(transformedResponse), { 70 | headers: {'Content-Type': 'application/json', 71 | 'Access-Control-Allow-Origin': '*', 72 | 'Access-Control-Allow-Methods': '*', 73 | 'Access-Control-Allow-Headers': '*' } 74 | }); 75 | } else { 76 | let { readable, writable } = new TransformStream(); 77 | streamResponse(transformedResponse, writable); 78 | return new Response(readable, { 79 | headers: {'Content-Type': 'text/event-stream', 80 | 'Access-Control-Allow-Origin': '*', 81 | 'Access-Control-Allow-Methods': '*', 82 | 'Access-Control-Allow-Headers': '*' } 83 | }); 84 | } 85 | } 86 | 87 | function streamResponse(response, writable) { 88 | let encoder = new TextEncoder(); 89 | let writer = writable.getWriter(); 90 | 91 | let content = response.choices[0].message.content; 92 | 93 | // Split the content into chunks, and send each chunk as a separate event 94 | let chunks = content.match(/\s+|\S+/g) || []; 95 | chunks.forEach((chunk, i) => { 96 | let chunkResponse = { 97 | ...response, 98 | object: "chat.completion.chunk", 99 | choices: [{ 100 | index: response.choices[0].index, 101 | delta: { ...response.choices[0].message, content: chunk }, 102 | finish_reason: i === chunks.length - 1 ? 'stop' : null // Set 'stop' for the last chunk 103 | }], 104 | usage: null 105 | }; 106 | 107 | writer.write(encoder.encode(`data: ${JSON.stringify(chunkResponse)}\n\n`)); 108 | }); 109 | 110 | // Write the done signal 111 | writer.write(encoder.encode(`data: [DONE]\n`)); 112 | 113 | writer.close(); 114 | } 115 | 116 | 117 | // Function to transform the response 118 | function transformResponse(palmData) { 119 | // Check if the 'candidates' array exists and if it's not empty 120 | if (!palmData.candidates || palmData.candidates.length === 0) { 121 | // If it doesn't exist or is empty, create a default candidate message 122 | palmData.candidates = [ 123 | { 124 | "author": "1", 125 | "content": "Ooops, the model returned nothing" 126 | } 127 | ]; 128 | } 129 | 130 | return { 131 | id: "chatcmpl-QXlha2FBbmROaXhpZUFyZUF3ZXNvbWUK", 132 | object: 'chat.completion', 133 | created: Math.floor(Date.now() / 1000), // Current Unix timestamp 134 | model: 'gpt-3.5-turbo', // Static model name 135 | usage: { 136 | prompt_tokens: palmData.messages.length, // This is a placeholder. Replace with actual token count if available 137 | completion_tokens: palmData.candidates.length, // This is a placeholder. Replace with actual token count if available 138 | total_tokens: palmData.messages.length + palmData.candidates.length, // This is a placeholder. Replace with actual token count if available 139 | }, 140 | choices: palmData.candidates.map((candidate, index) => ({ 141 | message: { 142 | role: 'assistant', 143 | content: candidate.content, 144 | }, 145 | finish_reason: 'stop', // Static finish reason 146 | index: index, 147 | })), 148 | }; 149 | } 150 | 151 | async function handleOPTIONS(request) { 152 | return new Response("pong", { 153 | headers: { 154 | 'Access-Control-Allow-Origin': '*', 155 | 'Access-Control-Allow-Methods': '*', 156 | 'Access-Control-Allow-Headers': '*' 157 | } 158 | }) 159 | } 160 | --------------------------------------------------------------------------------