Merge pull request #586 from fangpin/railway

Support Railway deployment

README.md

@@ -54,6 +54,9 @@
 
 > 项目中使用的对话模型是 davinci，计费方式是约每 750 字 (包含请求和回复) 消耗 $0.02，图片生成是每张消耗 $0.016，账号创建有免费的 $18 额度 (更新3.25: 最新注册的已经无免费额度了)，使用完可以更换邮箱重新注册。
 
+#### 1.1 ChapGPT service On Azure
+一种替换以上的方法是使用Azure推出的[ChatGPT service](https://azure.microsoft.com/en-in/products/cognitive-services/openai-service/)。它host在公有云Azure上，因此不需要VPN就可以直接访问。不过目前仍然处于preview阶段。新用户可以通过Try Azure for free来薅一段时间的羊毛
+
 
 ### 2.运行环境
 
@@ -94,7 +97,7 @@ pip3 install --upgrade openai
 # config.json文件内容示例
 { 
   "open_ai_api_key": "YOUR API KEY",                          # 填入上面创建的 OpenAI API KEY
-  "model": "gpt-3.5-turbo",                                   # 模型名称
+  "model": "gpt-3.5-turbo",                                   # 模型名称。当use_azure_chatgpt为true时，其名称为Azure上model deployment名称
   "proxy": "127.0.0.1:7890",                                  # 代理客户端的ip和端口
   "single_chat_prefix": ["bot", "@bot"],                      # 私聊时文本需要包含该前缀才能触发机器人回复
   "single_chat_reply_prefix": "[bot] ",                       # 私聊时自动回复的前缀，用于区分真人
@@ -104,7 +107,8 @@ pip3 install --upgrade openai
   "image_create_prefix": ["画", "看", "找"],                   # 开启图片回复的前缀
   "conversation_max_tokens": 1000,                            # 支持上下文记忆的最多字符数
   "speech_recognition": false,                                # 是否开启语音识别
-  "character_desc": "你是ChatGPT, 一个由OpenAI训练的大型语言模型, 你旨在回答并解决人们的任何问题，并且可以使用多种语言与人交流。",  # 人格描述
+  "use_azure_chatgpt": false,                                 # 是否使用Azure ChatGPT service代替openai ChatGPT service. 当设置为true时需要设置 open_ai_api_base，如 https://xxx.openai.azure.com/
+  "character_desc": "你是ChatGPT, 一个由OpenAI训练的大型语言模型, 你旨在回答并解决人们的任何问题，并且可以使用多种语言与人交流。",  # 人格描述,
 }
 ```
 **配置说明：**
@@ -173,6 +177,14 @@ nohup python3 app.py & tail -f nohup.out          # 在后台运行程序并通
 
 参考文档 [Docker部署](https://github.com/limccn/chatgpt-on-wechat/wiki/Docker%E9%83%A8%E7%BD%B2)   (Contributed by [limccn](https://github.com/limccn))。
 
+### 4. Railway部署
+[Use with Railway](#use-with-railway)(PaaS, Free, Stable, ✅Recommended)
+> Railway offers $5 (500 hours) of runtime per month
+1. Click the [Railway](https://railway.app/) button to go to the Railway homepage
+2. Click the `Start New Project` button.
+3. Click the `Deploy from Github repo` button.
+4. Choose your repo (you can fork this repo firstly)
+5. Set environment variable to override settings in config-template.json, such as: model, open_ai_api_base, open_ai_api_key, use_azure_chatgpt etc.
 
 ## 常见问题
 

app.py

@@ -5,7 +5,8 @@ from channel import channel_factory
 from common.log import logger
 
 from plugins import *
-if __name__ == '__main__':
+
+def run():
     try:
         # load config
         config.load_config()
@@ -21,3 +22,6 @@ if __name__ == '__main__':
     except Exception as e:
         logger.error("App startup failed!")
         logger.exception(e)
+
+if __name__ == '__main__':
+    run()

bot/bot_factory.py

@@ -24,4 +24,9 @@ def create_bot(bot_type):
         # OpenAI 官方对话模型API
         from bot.openai.open_ai_bot import OpenAIBot
         return OpenAIBot()
+
+    elif bot_type == const.CHATGPTONAZURE:
+        # Azure chatgpt service https://azure.microsoft.com/en-in/products/cognitive-services/openai-service/
+        from bot.chatgpt.chat_gpt_bot import AzureChatGPTBot
+        return AzureChatGPTBot()
     raise RuntimeError

bot/chatgpt/chat_gpt_bot.py

@@ -150,6 +150,61 @@ class ChatGPTBot(Bot):
             return False, str(e)
 
 
+class AzureChatGPTBot(ChatGPTBot):
+    def __init__(self):
+        super().__init__()
+        openai.api_type = "azure"
+        openai.api_version = "2023-03-15-preview"
+
+    def reply_text(self, session, session_id, retry_count=0) ->dict:
+        '''
+        call openai's ChatCompletion to get the answer
+        :param session: a conversation session
+        :param session_id: session id
+        :param retry_count: retry count
+        :return: {}
+        '''
+        try:
+            if conf().get('rate_limit_chatgpt') and not self.tb4chatgpt.get_token():
+                return {"completion_tokens": 0, "content": "提问太快啦，请休息一下再问我吧"}
+            response = openai.ChatCompletion.create(
+                engine=conf().get("model") or "gpt-3.5-turbo",  # the model deployment name on Azure
+                messages=session,
+                temperature=conf().get('temperature', 0.9),  # 值在[0,1]之间，越大表示回复越具有不确定性
+                #max_tokens=4096,  # 回复最大的字符数
+                top_p=1,
+                frequency_penalty=conf().get('frequency_penalty', 0.0),  # [-2,2]之间，该值越大则更倾向于产生不同的内容
+                presence_penalty=conf().get('presence_penalty', 0.0),  # [-2,2]之间，该值越大则更倾向于产生不同的内容
+            )
+            # logger.info("[ChatGPT] reply={}, total_tokens={}".format(response.choices[0]['message']['content'], response["usage"]["total_tokens"]))
+            return {"total_tokens": response["usage"]["total_tokens"], 
+                    "completion_tokens": response["usage"]["completion_tokens"], 
+                    "content": response.choices[0]['message']['content']}
+        except openai.error.RateLimitError as e:
+            # rate limit exception
+            logger.warn(e)
+            if retry_count < 1:
+                time.sleep(5)
+                logger.warn("[OPEN_AI] RateLimit exceed, 第{}次重试".format(retry_count+1))
+                return self.reply_text(session, session_id, retry_count+1)
+            else:
+                return {"completion_tokens": 0, "content": "提问太快啦，请休息一下再问我吧"}
+        except openai.error.APIConnectionError as e:
+            # api connection exception
+            logger.warn(e)
+            logger.warn("[OPEN_AI] APIConnection failed")
+            return {"completion_tokens": 0, "content":"我连接不到你的网络"}
+        except openai.error.Timeout as e:
+            logger.warn(e)
+            logger.warn("[OPEN_AI] Timeout")
+            return {"completion_tokens": 0, "content":"我没有收到你的消息"}
+        except Exception as e:
+            # unknown exception
+            logger.exception(e)
+            Session.clear_session(session_id)
+            return {"completion_tokens": 0, "content": "请再问我一次吧"}
+
+
 class SessionManager(object):
     def __init__(self):
         if conf().get('expires_in_seconds'):

bridge/bridge.py

@@ -19,6 +19,8 @@ class Bridge(object):
         model_type = conf().get("model")
         if model_type in ["text-davinci-003"]:
             self.btype['chat'] = const.OPEN_AI
+        if conf().get("use_azure_chatgpt"):
+            self.btype['chat'] = const.CHATGPTONAZURE
         self.bots={}
 
     def get_bot(self,typename):

common/const.py

@@ -1,4 +1,5 @@
 # bot_type
 OPEN_AI = "openAI"
 CHATGPT = "chatGPT"
-BAIDU = "baidu"
+BAIDU = "baidu"
+CHATGPTONAZURE = "chatGPTOnAzure"

config-template.json

@@ -2,6 +2,7 @@
   "open_ai_api_key": "YOUR API KEY",
   "model": "gpt-3.5-turbo",
   "proxy": "",
+  "use_azure_chatgpt": false,
   "single_chat_prefix": ["bot", "@bot"],
   "single_chat_reply_prefix": "[bot] ",
   "group_chat_prefix": ["@bot"],

config.py

@@ -10,11 +10,18 @@ def load_config():
     global config
     config_path = "./config.json"
     if not os.path.exists(config_path):
-        raise Exception('配置文件不存在，请根据config-template.json模板创建config.json文件')
+        logger.info('配置文件不存在，将使用config-template.json模板')
+        config_path = "./config-template.json"
 
     config_str = read_file(config_path)
     # 将json字符串反序列化为dict类型
     config = json.loads(config_str)
+
+    # override config with environment variables.
+    # Some online deployment platforms (e.g. Railway) deploy project from github directly. So you shouldn't put your secrets like api key in a config file, instead use environment variables to override the default config.
+    for name, value in os.environ.items():
+        config[name] = value
+
     logger.info("[INIT] load config: {}".format(config))
 
 

main.py

@@ -0,0 +1,5 @@
+# entry point for online railway deployment
+from app import run
+
+if __name__ == '__main__':
+	run()