diff --git a/README.md b/README.md
index 03f6933..1a72325 100644
--- a/README.md
+++ b/README.md
@@ -5,14 +5,23 @@
> 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。
+### 项目说明:
+- 我没有服务器可以使用吗?
+ - 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
+- 我没有自己的域名可以使用吗?
+ - 也可以, 参考: https://github.com/haibbo/cf-openai-azure-proxy/issues/3
+- 实现打印机模式：
+ - Azure OpenAI Service's 回复是一段一段回复的（当前仅GPT-4有此问题， GPT-3的问题Azure已经解决）
+ - 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
+- 项目也支持 Docker 部署（基于 wrangler）
-
+### 部署
代理 OpenAI 的请求到 Azure OpenAI Serivce，代码部署步骤：
1. 注册并登录到 Cloudflare 账户
2. 创建一个新的 Cloudflare Worker
3. 将 [cf-openai-azure-proxy.js](./cf-openai-azure-proxy.js) 复制并粘贴到 Cloudflare Worker 编辑器中
-4. 通过修改或环境变量调整 resourceName 和 deployName 的值
+4. 通过修改或环境变量调整 resourceName 和 deployment mapper 的值
5. 保存并部署 Cloudflare Worker
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
@@ -29,14 +38,15 @@
// The name of your Azure OpenAI Resource.
const resourceName="codegpt"
-const mapper:any = {
+// deployment model mapper
+const mapper = {
'gpt-3.5-turbo': 'gpt35',
'gpt-4': 'gpt4'
};
其他的map规则直接按这样的格式续写即可
```
- 或者通过 cloudflare worker 控制台, 进入 Workers script > Settings > Add variable under Environment Variables.
-
+
### 客户端
以 OpenCat 为例: 自定义 API 域名填写 第六步绑定的域名:
@@ -44,10 +54,3 @@ const mapper:any = {
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.
-
-QA:
-- 我没有服务器可以使用吗?
- - 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
-- 我没有自己的域名可以使用吗?
- - 也可以, 参考: https://github.com/haibbo/cf-openai-azure-proxy/issues/3
-
diff --git a/README_en.md b/README_en.md
index cc30750..9d043b7 100644
--- a/README_en.md
+++ b/README_en.md
@@ -10,7 +10,7 @@ This script proxies requests to Azure OpenAI Service for OpenAI clients. The cod
Register and log in to your Cloudflare account.
- Create a new Cloudflare Worker.
- Copy and paste cf-openai-azure-proxy.js into the Cloudflare Worker editor.
-- Adjust the values of **resourceName** and **deployName** by either direct modification or using environment variables..
+- Adjust the values of **resourceName** and deployment **mapper** by either direct modification or using environment variables..
- Save and deploy the Cloudflare Worker.
- 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.
@@ -29,12 +29,10 @@ const resourceName="codegpt"
'gpt-4': 'gpt4'
};
```
-**Mapper configuration example**: If you have deployed the GPT-3.5 Turbo and GPT-4 models on Azure with deployment names 'gpt35' and 'gpt4', respectively, then the mapper should be configured as follows.
Other map rules can be continued directly in this format.
+- **OR** go to the Cloudflare Worker console, navigate to Workers script > Settings > Add variable under Environment Variables.
-
-- go to the Cloudflare Worker console, navigate to Workers script > Settings > Add variable under Environment Variables.
-
+
## Client
Take OpenCat as an example: fill in the custom API domain name with the domain name bound in step 6:
diff --git a/cf-openai-azure-proxy.js b/cf-openai-azure-proxy.js
index 9695c9a..7c36390 100644
--- a/cf-openai-azure-proxy.js
+++ b/cf-openai-azure-proxy.js
@@ -9,8 +9,6 @@ const mapper = {
const apiVersion="2023-03-15-preview"
-
-
addEventListener("fetch", (event) => {
event.respondWith(handleRequest(event.request));
});
@@ -62,8 +60,14 @@ async function handleRequest(request) {
body: typeof body === 'object' ? JSON.stringify(body) : '{}',
};
- let { readable, writable } = new TransformStream()
const response = await fetch(fetchAPI, payload);
+
+ // Since Azure has fixed gpt-3's printer effect, do not have to stream it.
+ if (modelName.startsWith('gpt-3') || body?.stream != true){
+ return response
+ }
+
+ let { readable, writable } = new TransformStream()
stream(response.body, writable);
return new Response(readable, response);
@@ -98,7 +102,7 @@ async function stream(readable, writable) {
// 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(30);
+ await sleep(50);
}
buffer = lines[lines.length - 1];