feat: support claude-2 and anthropic-version

README.md

@@ -24,7 +24,7 @@ This project converts the API of Anthropic's Claude model to the OpenAI Chat API
 
 * ✨ Call Claude API like OpenAI ChatGPT API
 * 💦 Support streaming response
-* 🐻 Support `claude-v1.3`, `claude-v1.3-100k` models
+* 🐻 Support `claude-instant-1`, `claude-2` models
 * 🌩️ Deploy by Cloudflare Workers or Docker
 
 ## Getting Started
@@ -61,7 +61,7 @@ The API will then be available at http://localhost:8000. API endpoint: `/v1/chat
 
 ### Usage
 
-When you input the model parameter as `gpt-3.5-turbo` or `gpt-3.5-turbo-0301`, it will be substituted with `claude-v1.3`. otherwise, `claude-v1.3-100k` will be utilized.
+When you input the model parameter as `gpt-3.5-turbo` or `gpt-3.5-turbo-0613`, it will be substituted with `claude-instant-1`. otherwise, `claude-2` will be utilized.
 
 
 #### GUI
@@ -90,7 +90,7 @@ The Claude Completion API has an endpoint `/v1/complete` which takes the followi
 ```json
 {
   "prompt": "\n\nHuman: Hello, AI.\n\nAssistant: ",
-  "model": "claude-v1.3",
+  "model": "claude-instant-1",
   "max_tokens_to_sample": 100,
   "temperature": 1,
   "stream": true

README_CN.md

@@ -24,7 +24,7 @@
 
 * ✨ 以 OpenAI ChatGPT API 的方式调用 Claude API
 * 💦 支持流式响应，实现打印机效果
-* 🐻 支持 `claude-v1.3`，`claude-v1.3-100k` 模型
+* 🐻 支持 `claude-instant-1`，`claude-2` 模型
 * 🌩️ 通过 Cloudflare Workers 或 Docker 部署
 
 ## 开始使用
@@ -60,7 +60,7 @@ docker-compose up
 
 ### 使用方法
 
-当您将模型参数 `model` 为 `gpt-3.5-turbo` 或 `gpt-3.5-turbo-0301` 时，它将替换为 `claude-v1.3`。否则，将使用 `claude-v1.3-100k`。
+当您将模型参数 `model` 为 `gpt-3.5-turbo` 或 `gpt-3.5-turbo-0613` 时，它将替换为 `claude-instant-1`。否则，将使用 `claude-2`。
 
 
 #### 图形界面软件
@@ -89,7 +89,7 @@ Claude Completion API endpoint 为 `/v1/complete`，它接受以下请求格式
 ```json
 {
   "prompt": "\n\nHuman: Hello, AI.\n\nAssistant: ",
-  "model": "claude-v1.3",
+  "model": "claude-instant-1",
   "max_tokens_to_sample": 100,
   "temperature": 1,
   "stream": true

README_JA.md

@@ -24,7 +24,7 @@
 
 - ✨ OpenAI ChatGPT API のように Claude API を呼び出す
 - 💦 ストリーミングレス対応
-- 🐻 `claude-v1.3`, `claude-v1.3-100k` のモデルをサポートする
+- 🐻 `claude-instant-1`, `claude-2` のモデルをサポートする
 - 🌩️ Cloudflare Workers や Docker でデプロイする
 
 ## はじめに
@@ -60,7 +60,7 @@ docker-compose up
 
 ### 使用方法
 
-モデルパラメータを `gpt-3.5-turbo` または `gpt-3.5-turbo-0301` と入力すると `claude-v1.3` に置換されます。
+モデルパラメータを `gpt-3.5-turbo` または `gpt-3.5-turbo-0613` と入力すると `claude-instant-1` に置換されます。
 
 #### GUI
 
@@ -88,7 +88,7 @@ Claude Completion API にはエンドポイント `/v1/complete` があり、以
 ```json
 {
   "prompt": "\n\nHuman: Hello, AI.\n\nAssistant: ",
-  "model": "claude-v1.3",
+  "model": "claude-instant-1",
   "max_tokens_to_sample": 100,
   "temperature": 1,
   "stream": true

claude_to_chatgpt/adapter.py

@@ -42,15 +42,15 @@ def convert_messages_to_prompt(self, messages):
         return prompt
 
     def openai_to_claude_params(self, openai_params):
-        model = model_map.get(openai_params["model"], "claude-v1.3-100k")
+        model = model_map.get(openai_params["model"], "claude-2")
         messages = openai_params["messages"]
 
         prompt = self.convert_messages_to_prompt(messages)
 
         claude_params = {
             "model": model,
             "prompt": prompt,
-            "max_tokens_to_sample": 100000 if model == "claude-v1.3-100k" else 9016,
+            "max_tokens_to_sample": 100000,
         }
 
         if openai_params.get("max_tokens"):
@@ -67,13 +67,14 @@ def openai_to_claude_params(self, openai_params):
 
         return claude_params
 
-    def claude_to_chatgpt_response_stream(self, claude_response, prev_decoded_response):
-        completion_tokens = num_tokens_from_string(claude_response.get("completion", ""))
+    def claude_to_chatgpt_response_stream(self, claude_response):
+        completion = claude_response.get("completion", "")
+        completion_tokens = num_tokens_from_string(completion)
         openai_response = {
             "id": f"chatcmpl-{str(time.time())}",
             "object": "chat.completion.chunk",
             "created": int(time.time()),
-            "model": "gpt-3.5-turbo-0301",
+            "model": "gpt-3.5-turbo-0613",
             "usage": {
                 "prompt_tokens": 0,
                 "completion_tokens": completion_tokens,
@@ -83,9 +84,7 @@ def claude_to_chatgpt_response_stream(self, claude_response, prev_decoded_respon
                 {
                     "delta": {
                         "role": "assistant",
-                        "content": claude_response.get("completion", "").removeprefix(
-                            prev_decoded_response.get("completion", "")
-                        ),
+                        "content": completion,
                     },
                     "index": 0,
                     "finish_reason": stop_reason_map[claude_response.get("stop_reason")]
@@ -98,12 +97,14 @@ def claude_to_chatgpt_response_stream(self, claude_response, prev_decoded_respon
         return openai_response
 
     def claude_to_chatgpt_response(self, claude_response):
-        completion_tokens = num_tokens_from_string(claude_response.get("completion", ""))
+        completion_tokens = num_tokens_from_string(
+            claude_response.get("completion", "")
+        )
         openai_response = {
             "id": f"chatcmpl-{str(time.time())}",
             "object": "chat.completion",
             "created": int(time.time()),
-            "model": "gpt-3.5-turbo-0301",
+            "model": "gpt-3.5-turbo-0613",
             "usage": {
                 "prompt_tokens": 0,
                 "completion_tokens": completion_tokens,
@@ -137,7 +138,9 @@ async def chat(self, request: Request):
                     f"{self.claude_base_url}/v1/complete",
                     headers={
                         "x-api-key": api_key,
+                        "accept": "application/json",
                         "content-type": "application/json",
+                        "anthropic-version": "2023-06-01",
                     },
                     json=claude_params,
                 )
@@ -152,30 +155,38 @@ async def chat(self, request: Request):
                     f"{self.claude_base_url}/v1/complete",
                     headers={
                         "x-api-key": api_key,
+                        "accept": "application/json",
                         "content-type": "application/json",
+                        "anthropic-version": "2023-06-01",
                     },
                     json=claude_params,
                 ) as response:
                     if response.is_error:
                         raise Exception(f"Error: {response.status_code}")
-                    prev_decoded_line = {}
                     async for line in response.aiter_lines():
                         if line:
-                            if line == "data: [DONE]":
-                                yield "[DONE]"
-                                break
                             stripped_line = line.lstrip("data:")
                             if stripped_line:
                                 try:
                                     decoded_line = json.loads(stripped_line)
-                                    # yield decoded_line
-                                    openai_response = (
-                                        self.claude_to_chatgpt_response_stream(
-                                            decoded_line, prev_decoded_line
+                                    stop_reason = decoded_line.get("stop_reason")
+                                    if stop_reason:
+                                        yield self.claude_to_chatgpt_response_stream(
+                                            {
+                                                "completion": "",
+                                                "stop_reason": stop_reason,
+                                            }
                                         )
-                                    )
-                                    prev_decoded_line = decoded_line
-                                    yield openai_response
+                                        yield "[DONE]"
+                                    else:
+                                        completion = decoded_line.get("completion")
+                                        if completion:
+                                            openai_response = (
+                                                self.claude_to_chatgpt_response_stream(
+                                                    decoded_line
+                                                )
+                                            )
+                                            yield openai_response
                                 except json.JSONDecodeError as e:
                                     logger.debug(
                                         f"Error decoding JSON: {e}"

claude_to_chatgpt/models.py

@@ -24,7 +24,7 @@
         "parent": None,
     },
     {
-        "id": "gpt-3.5-turbo-0301",
+        "id": "gpt-3.5-turbo-0613",
         "object": "model",
         "created": 1677649963,
         "owned_by": "openai",
@@ -44,7 +44,7 @@
                 "is_blocking": False,
             }
         ],
-        "root": "gpt-3.5-turbo-0301",
+        "root": "gpt-3.5-turbo-0613",
         "parent": None,
     },
     {
@@ -72,7 +72,7 @@
         "parent": None,
     },
     {
-        "id": "gpt-4-0314",
+        "id": "gpt-4-0613",
         "object": "model",
         "created": 1678604601,
         "owned_by": "openai",
@@ -92,14 +92,14 @@
                 "is_blocking": False,
             }
         ],
-        "root": "gpt-4-0314",
+        "root": "gpt-4-0613",
         "parent": None,
     },
 ]
 
 model_map = {
-    "gpt-3.5-turbo": "claude-v1.3",
-    "gpt-3.5-turbo-0301": "claude-v1.3",
-    "gpt-4": "claude-v1.3-100k",
-    "gpt-4-0314": "claude-v1.3-100k",
+    "gpt-3.5-turbo": "claude-instant-1",
+    "gpt-3.5-turbo-0613": "claude-instant-1",
+    "gpt-4": "claude-2",
+    "gpt-4-0613": "claude-2",
 }

cloudflare-worker.js

@@ -45,7 +45,7 @@ function claudeToChatGPTResponse(claudeResponse, stream = false) {
   const result = {
     id: `chatcmpl-${timestamp}`,
     created: timestamp,
-    model: 'gpt-3.5-turbo-0301',
+    model: 'gpt-3.5-turbo-0613',
     usage: {
       prompt_tokens: 0,
       completion_tokens: completionTokens,
@@ -149,12 +149,12 @@ async function handleRequest(request) {
 
     const requestBody = await request.json();
     const { model, messages, temperature, stop, stream } = requestBody;
-    const claudeModel = model_map[model] || 'claude-v1.3-100k';
+    const claudeModel = model_map[model] || 'claude-2';
 
     // OpenAI API 转换为 Claude API
     const prompt = convertMessagesToPrompt(messages);
     let maxTokensToSample = 100000;
-    if (model !== 'claude-v1.3-100k') {
+    if (model !== 'claude-2') {
       maxTokensToSample = 9016;
     }
     const claudeRequestBody = {
@@ -237,7 +237,7 @@ const models_list = [
     parent: null,
   },
   {
-    id: 'gpt-3.5-turbo-0301',
+    id: 'gpt-3.5-turbo-0613',
     object: 'model',
     created: 1677649963,
     owned_by: 'openai',
@@ -257,7 +257,7 @@ const models_list = [
         is_blocking: false,
       },
     ],
-    root: 'gpt-3.5-turbo-0301',
+    root: 'gpt-3.5-turbo-0613',
     parent: null,
   },
   {
@@ -305,14 +305,14 @@ const models_list = [
         is_blocking: false,
       },
     ],
-    root: 'gpt-4-0314',
+    root: 'gpt-4-0613',
     parent: null,
   },
 ];
 
 const model_map = {
-  'gpt-3.5-turbo': 'claude-v1.3',
-  'gpt-3.5-turbo-0301': 'claude-v1.3',
-  'gpt-4': 'claude-v1.3-100k',
-  'gpt-4-0314': 'claude-v1.3-100k',
+  'gpt-3.5-turbo': 'claude-instant-1',
+  'gpt-3.5-turbo-0613': 'claude-instant-1',
+  'gpt-4': 'claude-2',
+  'gpt-4-0613': 'claude-2',
 };