support gpt-3.5-turbo api

Signed-off-by: pengzhile <[email protected]>

Dockerfile

@@ -4,4 +4,4 @@ WORKDIR /opt/app
 ADD . .
 RUN pip install .
 
-ENTRYPOINT ["bin/startup.sh"]
+ENTRYPOINT ["/opt/app/bin/startup.sh"]

README.md

@@ -9,8 +9,6 @@
 [![PyPi workflow](https://github.com/pengzhile/pandora/actions/workflows/python-publish.yml/badge.svg)](https://github.com/pengzhile/pandora/actions/workflows/python-publish.yml)
 [![Docker workflow](https://github.com/pengzhile/pandora/actions/workflows/docker-publish.yml/badge.svg)](https://github.com/pengzhile/pandora/actions/workflows/docker-publish.yml)
 
-[English](https://github.com/pengzhile/pandora/blob/master/doc/README.en.md)
-
 ### `潘多拉`，一个命令行的`ChatGPT`。
 
 ### 实现了网页版`ChatGPT`的主要操作。能过`Cloudflare`，理论上速度还可以。
@@ -76,13 +74,15 @@
 * `-p` 或 `--proxy` 指定代理，格式：`protocol://user:pass@ip:port`。
 * `-t` 或 `--token_file` 指定一个存放`Access Token`的文件，使用`Access Token`登录。
 * `-s` 或 `--server` 以`http`服务方式启动，格式：`ip:port`。
+* `-a` 或 `--api` 使用`gpt-3.5-turbo`API请求，**你可能需要向`OpenAI`支付费用**。
 * `-v` 或 `--verbose` 显示调试信息，且出错时打印异常堆栈信息，供查错使用。
 
 ## Docker环境变量
 
 * `PANDORA_ACCESS_TOKEN` 指定`Access Token`字符串。
 * `PANDORA_PROXY` 指定代理，格式：`protocol://user:pass@ip:port`。
 * `PANDORA_SERVER` 以`http`服务方式启动，格式：`ip:port`。
+* `PANDORA_API` 使用`gpt-3.5-turbo`API请求，**你可能需要向`OpenAI`支付费用**。
 * `PANDORA_VERBOSE` 显示调试信息，且出错时打印异常堆栈信息，供查错使用。
 * 使用Docker方式，设置环境变量即可，无视上述`程序参数`。
 

bin/startup.sh

@@ -17,6 +17,10 @@ if [ -n "${PANDORA_SERVER}" ]; then
   PANDORA_ARGS="${PANDORA_ARGS} -s ${PANDORA_SERVER}"
 fi
 
+if [ -n "${PANDORA_API}" ]; then
+  PANDORA_ARGS="${PANDORA_ARGS} -a"
+fi
+
 if [ -n "${PANDORA_VERBOSE}" ]; then
   PANDORA_ARGS="${PANDORA_ARGS} -v"
 fi

doc/HTTP-API.md

@@ -65,8 +65,8 @@
 * **JSON字段：**
     * `prompt` 提问的内容。
     * `model` 对话使用的模型，通常整个会话中保持不变。
-    * `last_user_message_id` 上一条用户发送消息的ID。
-    * `last_parent_message_id` 上一条用户发送消息的父消息ID。
+    * `message_id` 上一条用户发送消息的ID。
+    * `parent_message_id` 上一条用户发送消息的父消息ID。
     * `conversation_id` 会话ID，在这个接口不可不传。
     * `stream` 是否使用流的方式输出内容，默认为：`True`
 * **接口描述：** 让`ChatGPT`重新生成回复。

requirements.txt

@@ -5,6 +5,7 @@ requests[socks]==2.28.2
 rich==13.3.1
 appdirs==1.4.4
 werkzeug==2.2.3
-flask[async]==2.2.3
+flask==2.2.3
 flask-cors==3.0.10
 waitress==2.1.2
+tiktoken==0.3.0

setup.py

@@ -17,7 +17,7 @@
     long_description=long_description,
     long_description_content_type='text/markdown',
     url="https://github.com/pengzhile/pandora",
-    packages=['pandora', 'pandora.openai', 'pandora.bots'],
+    packages=['pandora', 'pandora.openai', 'pandora.bots', 'pandora.turbo'],
     package_dir={'pandora': 'src/pandora'},
     include_package_data=True,
     install_requires=[
@@ -28,9 +28,10 @@
         'rich == 13.3.1',
         'appdirs == 1.4.4',
         'werkzeug == 2.2.3',
-        'flask[async] == 2.2.3',
+        'flask == 2.2.3',
         'flask-cors == 3.0.10',
         'waitress == 2.1.2',
+        'tiktoken == 0.3.0',
     ],
     entry_points={
         "console_scripts": [

src/pandora/__init__.py

@@ -1,3 +1,3 @@
 # -*- coding: utf-8 -*-
 
-__version__ = '0.4.4'
+__version__ = '0.5.0'

src/pandora/bots/legacy.py

@@ -1,13 +1,11 @@
 # -*- coding: utf-8 -*-
 
-import asyncio
 import re
 import uuid
 
 from rich.prompt import Prompt, Confirm
 
 from .. import __version__
-from ..openai.api import ChatGPT
 from ..openai.utils import Console
 
 
@@ -35,7 +33,7 @@ def __init__(self, title=None, conversation_id=None, model_slug=None, user_promp
 
 
 class ChatBot:
-    def __init__(self, chatgpt: ChatGPT):
+    def __init__(self, chatgpt):
         self.chatgpt = chatgpt
         self.state = None
 
@@ -285,9 +283,9 @@ def __talk(self, prompt):
         else:
             self.state.user_prompt = ChatPrompt(prompt, parent_id=self.state.chatgpt_prompt.message_id)
 
-        generator = self.chatgpt.talk(prompt, self.state.model_slug, self.state.user_prompt.message_id,
-                                      self.state.user_prompt.parent_id, self.state.conversation_id)
-        asyncio.run(self.__print_reply(generator))
+        status, _, generator = self.chatgpt.talk(prompt, self.state.model_slug, self.state.user_prompt.message_id,
+                                                 self.state.user_prompt.parent_id, self.state.conversation_id)
+        self.__print_reply(status, generator)
 
         self.state.user_prompts.append(self.state.user_prompt)
 
@@ -302,15 +300,19 @@ def __regenerate_reply(self, state):
             Console.error('#### Conversation has not been created.')
             return
 
-        generator = self.chatgpt.regenerate_reply(state.user_prompt.prompt, state.model_slug, state.conversation_id,
-                                                  state.user_prompt.message_id, state.user_prompt.parent_id)
+        status, _, generator = self.chatgpt.regenerate_reply(state.user_prompt.prompt, state.model_slug,
+                                                             state.conversation_id, state.user_prompt.message_id,
+                                                             state.user_prompt.parent_id)
         print()
         Console.success_b('ChatGPT:')
-        asyncio.run(self.__print_reply(generator))
+        self.__print_reply(status, generator)
+
+    def __print_reply(self, status, generator):
+        if 200 != status:
+            raise Exception(status, next(generator))
 
-    async def __print_reply(self, generator):
         p = 0
-        async for result in await generator:
+        for result in generator:
             if result['error']:
                 raise Exception(result['error'])
 

src/pandora/bots/server.py

@@ -1,6 +1,5 @@
 # -*- coding: utf-8 -*-
 
-import asyncio
 import logging
 from os.path import join, abspath, dirname
 
@@ -12,14 +11,14 @@
 from werkzeug.serving import WSGIRequestHandler
 
 from .. import __version__
-from ..openai.api import ChatGPT
+from ..openai.api import API
 
 
 class ChatBot:
     __default_ip = '127.0.0.1'
     __default_port = 8008
 
-    def __init__(self, chatgpt: ChatGPT, debug=False):
+    def __init__(self, chatgpt, debug=False):
         self.chatgpt = chatgpt
         self.debug = debug
         self.log_level = logging.DEBUG if debug else logging.INFO
@@ -103,8 +102,8 @@ def list_models(self):
         return self.__proxy_result(self.chatgpt.list_models(True))
 
     def list_conversations(self):
-        offset = request.args.get('offset', 1)
-        limit = request.args.get('limit', 20)
+        offset = request.args.get('offset', '1')
+        limit = request.args.get('limit', '20')
 
         return self.__proxy_result(self.chatgpt.list_conversations(offset, limit, True))
 
@@ -129,7 +128,7 @@ def gen_conversation_title(self, conversation_id):
 
         return self.__proxy_result(self.chatgpt.gen_conversation_title(conversation_id, model, message_id, True))
 
-    async def talk(self):
+    def talk(self):
         payload = request.json
         prompt = payload['prompt']
         model = payload['model']
@@ -138,53 +137,32 @@ async def talk(self):
         conversation_id = payload.get('conversation_id')
         stream = payload.get('stream', True)
 
-        async def __talk():
-            generator = self.chatgpt.talk(prompt, model, message_id, parent_message_id, conversation_id, stream)
-            async for line in await generator:
-                yield line
+        return self.__process_stream(
+            *self.chatgpt.talk(prompt, model, message_id, parent_message_id, conversation_id, stream), stream)
 
-        if stream:
-            return Response(self.__to_sync(__talk()), mimetype='text/event-stream')
-
-        last_json = None
-        async for json in __talk():
-            last_json = json
-
-        return jsonify(last_json)
-
-    async def regenerate(self):
+    def regenerate(self):
         payload = request.json
         prompt = payload['prompt']
         model = payload['model']
-        last_user_message_id = payload['last_user_message_id']
-        last_parent_message_id = payload['last_parent_message_id']
+        message_id = payload['message_id']
+        parent_message_id = payload['parent_message_id']
         conversation_id = payload['conversation_id']
         stream = payload.get('stream', True)
 
-        async def __generate():
-            generator = self.chatgpt.regenerate_reply(prompt, model, conversation_id, last_user_message_id,
-                                                      last_parent_message_id, stream)
-            async for line in await generator:
-                yield line
+        return self.__process_stream(
+            *self.chatgpt.regenerate_reply(prompt, model, conversation_id, message_id, parent_message_id, stream),
+            stream)
 
+    @staticmethod
+    def __process_stream(status, headers, generator, stream):
         if stream:
-            return Response(self.__to_sync(__generate()), mimetype='text/event-stream')
+            return Response(API.wrap_stream_out(generator, status), mimetype=headers['Content-Type'], status=status)
 
         last_json = None
-        async for json in __generate():
+        for json in generator:
             last_json = json
 
-        return jsonify(last_json)
-
-    @staticmethod
-    def __to_sync(generator):
-        loop = asyncio.new_event_loop()
-
-        while True:
-            try:
-                yield loop.run_until_complete(generator.__anext__())
-            except StopAsyncIteration:
-                break
+        return make_response(last_json, status)
 
     @staticmethod
     def __proxy_result(remote_resp):