Merge pull request #621 from lanvent/dev2

refactor and support plugins for OpenAIBot

bot/chatgpt/chat_gpt_bot.py

@@ -1,6 +1,9 @@
 # encoding:utf-8
 
 from bot.bot import Bot
+from bot.chatgpt.chat_gpt_session import ChatGPTSession
+from bot.openai.open_ai_image import OpenAIImage
+from bot.session_manager import Session, SessionManager
 from bridge.context import ContextType
 from bridge.reply import Reply, ReplyType
 from config import conf, load_config
@@ -10,21 +13,20 @@ from common.expired_dict import ExpiredDict
 import openai
 import time
 
-
 # OpenAI对话模型API (可用)
-class ChatGPTBot(Bot):
+class ChatGPTBot(Bot,OpenAIImage):
     def __init__(self):
+        super().__init__()
         openai.api_key = conf().get('open_ai_api_key')
         if conf().get('open_ai_api_base'):
             openai.api_base = conf().get('open_ai_api_base')
         proxy = conf().get('proxy')
-        self.sessions = SessionManager(model= conf().get("model") or "gpt-3.5-turbo")
         if proxy:
             openai.proxy = proxy
         if conf().get('rate_limit_chatgpt'):
             self.tb4chatgpt = TokenBucket(conf().get('rate_limit_chatgpt', 20))
-        if conf().get('rate_limit_dalle'):
-            self.tb4dalle = TokenBucket(conf().get('rate_limit_dalle', 50))
+        
+        self.sessions = SessionManager(ChatGPTSession, model= conf().get("model") or "gpt-3.5-turbo")
 
     def reply(self, query, context=None):
         # acquire reply content
@@ -45,19 +47,19 @@ class ChatGPTBot(Bot):
                 reply = Reply(ReplyType.INFO, '配置已更新')
             if reply:
                 return reply
-            session = self.sessions.build_session_query(query, session_id)
-            logger.debug("[OPEN_AI] session query={}".format(session))
+            session = self.sessions.session_query(query, session_id)
+            logger.debug("[OPEN_AI] session query={}".format(session.messages))
 
             # if context.get('stream'):
             #     # reply in stream
             #     return self.reply_text_stream(query, new_query, session_id)
 
             reply_content = self.reply_text(session, session_id, 0)
-            logger.debug("[OPEN_AI] new_query={}, session_id={}, reply_cont={}, completion_tokens={}".format(session, session_id, reply_content["content"], reply_content["completion_tokens"]))
+            logger.debug("[OPEN_AI] new_query={}, session_id={}, reply_cont={}, completion_tokens={}".format(session.messages, session_id, reply_content["content"], reply_content["completion_tokens"]))
             if reply_content['completion_tokens'] == 0 and len(reply_content['content']) > 0:
                 reply = Reply(ReplyType.ERROR, reply_content['content'])
             elif reply_content["completion_tokens"] > 0:
-                self.sessions.save_session(reply_content["content"], session_id, reply_content["total_tokens"])
+                self.sessions.session_reply(reply_content["content"], session_id, reply_content["total_tokens"])
                 reply = Reply(ReplyType.TEXT, reply_content["content"])
             else:
                 reply = Reply(ReplyType.ERROR, reply_content['content'])
@@ -86,7 +88,7 @@ class ChatGPTBot(Bot):
             "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
         }
 
-    def reply_text(self, session, session_id, retry_count=0) -> dict:
+    def reply_text(self, session:ChatGPTSession, session_id, retry_count=0) -> dict:
         '''
         call openai's ChatCompletion to get the answer
         :param session: a conversation session
@@ -98,7 +100,7 @@ class ChatGPTBot(Bot):
             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(
-                messages=session, **self.compose_args()
+                messages=session.messages, **self.compose_args()
             )
             # logger.info("[ChatGPT] reply={}, total_tokens={}".format(response.choices[0]['message']['content'], response["usage"]["total_tokens"]))
             return {"total_tokens": response["usage"]["total_tokens"],
@@ -128,31 +130,6 @@ class ChatGPTBot(Bot):
             self.sessions.clear_session(session_id)
             return {"completion_tokens": 0, "content": "请再问我一次吧"}
 
-    def create_img(self, query, retry_count=0):
-        try:
-            if conf().get('rate_limit_dalle') and not self.tb4dalle.get_token():
-                return False, "请求太快了，请休息一下再问我吧"
-            logger.info("[OPEN_AI] image_query={}".format(query))
-            response = openai.Image.create(
-                prompt=query,    #图片描述
-                n=1,             #每次生成图片的数量
-                size="256x256"   #图片大小,可选有 256x256, 512x512, 1024x1024
-            )
-            image_url = response['data'][0]['url']
-            logger.info("[OPEN_AI] image_url={}".format(image_url))
-            return True, image_url
-        except openai.error.RateLimitError as e:
-            logger.warn(e)
-            if retry_count < 1:
-                time.sleep(5)
-                logger.warn("[OPEN_AI] ImgCreate RateLimit exceed, 第{}次重试".format(retry_count+1))
-                return self.create_img(query, retry_count+1)
-            else:
-                return False, "提问太快啦，请休息一下再问我吧"
-        except Exception as e:
-            logger.exception(e)
-            return False, str(e)
-
 
 class AzureChatGPTBot(ChatGPTBot):
     def __init__(self):
@@ -164,123 +141,4 @@ class AzureChatGPTBot(ChatGPTBot):
         args = super().compose_args()
         args["engine"] = args["model"]
         del(args["model"])
-        return args
-
-class SessionManager(object):
-    def __init__(self, model = "gpt-3.5-turbo-0301"):
-        if conf().get('expires_in_seconds'):
-            sessions = ExpiredDict(conf().get('expires_in_seconds'))
-        else:
-            sessions = dict()
-        self.sessions = sessions
-        self.model = model
-
-    def build_session(self, session_id, system_prompt=None):
-        session = self.sessions.get(session_id, [])
-        if len(session) == 0:
-            if system_prompt is None:
-                system_prompt = conf().get("character_desc", "")
-            system_item = {'role': 'system', 'content': system_prompt}
-            session.append(system_item)
-            self.sessions[session_id] = session
-        return session
-
-    def build_session_query(self, query, session_id):
-        '''
-        build query with conversation history
-        e.g.  [
-            {"role": "system", "content": "You are a helpful assistant."},
-            {"role": "user", "content": "Who won the world series in 2020?"},
-            {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
-            {"role": "user", "content": "Where was it played?"}
-        ]
-        :param query: query content
-        :param session_id: session id
-        :return: query content with conversaction
-        '''
-        session = self.build_session(session_id)
-        user_item = {'role': 'user', 'content': query}
-        session.append(user_item)
-        try:
-            total_tokens = num_tokens_from_messages(session, self.model)
-            max_tokens = conf().get("conversation_max_tokens", 1000)
-            total_tokens = self.discard_exceed_conversation(session, max_tokens, total_tokens)
-            logger.debug("prompt tokens used={}".format(total_tokens))
-        except Exception as e:
-            logger.debug("Exception when counting tokens precisely for prompt: {}".format(str(e)))
-
-        return session
-
-    def save_session(self, answer, session_id, total_tokens):
-        max_tokens = conf().get("conversation_max_tokens", 1000)
-        session = self.sessions.get(session_id)
-        if session:
-            # append conversation
-            gpt_item = {'role': 'assistant', 'content': answer}
-            session.append(gpt_item)
-
-        # discard exceed limit conversation
-        tokens_cnt = self.discard_exceed_conversation(session, max_tokens, total_tokens)
-        logger.debug("raw total_tokens={}, savesession tokens={}".format(total_tokens, tokens_cnt))
-
-    def discard_exceed_conversation(self, session, max_tokens, total_tokens):
-        dec_tokens = int(total_tokens)
-        # logger.info("prompt tokens used={},max_tokens={}".format(used_tokens,max_tokens))
-        while dec_tokens > max_tokens:
-            # pop first conversation
-            if len(session) > 2:
-                session.pop(1)
-            elif len(session) == 2 and session[1]["role"] == "assistant":
-                session.pop(1)
-                break
-            elif len(session) == 2 and session[1]["role"] == "user":
-                logger.warn("user message exceed max_tokens. total_tokens={}".format(dec_tokens))
-                break
-            else:
-                logger.debug("max_tokens={}, total_tokens={}, len(sessions)={}".format(max_tokens, dec_tokens, len(session)))
-                break
-            try:
-                cur_tokens = num_tokens_from_messages(session, self.model)
-                dec_tokens = cur_tokens
-            except Exception as e:
-                logger.debug("Exception when counting tokens precisely for query: {}".format(e))
-                dec_tokens = dec_tokens - max_tokens
-        return dec_tokens
-
-    def clear_session(self, session_id):
-        self.sessions[session_id] = []
-
-    def clear_all_session(self):
-        self.sessions.clear()
-
-# refer to https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
-def num_tokens_from_messages(messages, model):
-    """Returns the number of tokens used by a list of messages."""
-    import tiktoken
-    try:
-        encoding = tiktoken.encoding_for_model(model)
-    except KeyError:
-        logger.debug("Warning: model not found. Using cl100k_base encoding.")
-        encoding = tiktoken.get_encoding("cl100k_base")
-    if model == "gpt-3.5-turbo":
-        return num_tokens_from_messages(messages, model="gpt-3.5-turbo-0301")
-    elif model == "gpt-4":
-        return num_tokens_from_messages(messages, model="gpt-4-0314")
-    elif model == "gpt-3.5-turbo-0301":
-        tokens_per_message = 4  # every message follows <|start|>{role/name}\n{content}<|end|>\n
-        tokens_per_name = -1  # if there's a name, the role is omitted
-    elif model == "gpt-4-0314":
-        tokens_per_message = 3
-        tokens_per_name = 1
-    else:
-        logger.warn(f"num_tokens_from_messages() is not implemented for model {model}. Returning num tokens assuming gpt-3.5-turbo-0301.")
-        return num_tokens_from_messages(messages, model="gpt-3.5-turbo-0301")
-    num_tokens = 0
-    for message in messages:
-        num_tokens += tokens_per_message
-        for key, value in message.items():
-            num_tokens += len(encoding.encode(value))
-            if key == "name":
-                num_tokens += tokens_per_name
-    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>
-    return num_tokens
+        return args

bot/chatgpt/chat_gpt_session.py

@@ -0,0 +1,92 @@
+from bot.session_manager import Session
+from common.log import logger
+'''
+    e.g.  [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Who won the world series in 2020?"},
+        {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
+        {"role": "user", "content": "Where was it played?"}
+    ]
+'''
+class ChatGPTSession(Session):
+    def __init__(self, session_id, system_prompt=None, model= "gpt-3.5-turbo"):
+        super().__init__(session_id, system_prompt)
+        self.messages = []
+        self.model = model
+        self.reset()
+    
+    def reset(self):
+        system_item = {'role': 'system', 'content': self.system_prompt}
+        self.messages = [system_item]
+
+    def add_query(self, query):
+        user_item = {'role': 'user', 'content': query}
+        self.messages.append(user_item)
+
+    def add_reply(self, reply):
+        assistant_item = {'role': 'assistant', 'content': reply}
+        self.messages.append(assistant_item)
+    
+    def discard_exceeding(self, max_tokens, cur_tokens= None):
+        precise = True
+        try:
+            cur_tokens = num_tokens_from_messages(self.messages, self.model)
+        except Exception as e:
+            precise = False
+            if cur_tokens is None:
+                raise e
+            logger.debug("Exception when counting tokens precisely for query: {}".format(e))
+        while cur_tokens > max_tokens:
+            if len(self.messages) > 2:
+                self.messages.pop(1)
+            elif len(self.messages) == 2 and self.messages[1]["role"] == "assistant":
+                self.messages.pop(1)
+                if precise:
+                    cur_tokens = num_tokens_from_messages(self.messages, self.model)
+                else:
+                    cur_tokens = cur_tokens - max_tokens
+                break
+            elif len(self.messages) == 2 and self.messages[1]["role"] == "user":
+                logger.warn("user message exceed max_tokens. total_tokens={}".format(cur_tokens))
+                break
+            else:
+                logger.debug("max_tokens={}, total_tokens={}, len(messages)={}".format(max_tokens, cur_tokens, len(self.messages)))
+                break
+            if precise:
+                cur_tokens = num_tokens_from_messages(self.messages, self.model)
+            else:
+                cur_tokens = cur_tokens - max_tokens
+        return cur_tokens
+    
+
+# refer to https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
+def num_tokens_from_messages(messages, model):
+    """Returns the number of tokens used by a list of messages."""
+    import tiktoken
+    try:
+        encoding = tiktoken.encoding_for_model(model)
+    except KeyError:
+        logger.debug("Warning: model not found. Using cl100k_base encoding.")
+        encoding = tiktoken.get_encoding("cl100k_base")
+    if model == "gpt-3.5-turbo":
+        return num_tokens_from_messages(messages, model="gpt-3.5-turbo-0301")
+    elif model == "gpt-4":
+        return num_tokens_from_messages(messages, model="gpt-4-0314")
+    elif model == "gpt-3.5-turbo-0301":
+        tokens_per_message = 4  # every message follows <|start|>{role/name}\n{content}<|end|>\n
+        tokens_per_name = -1  # if there's a name, the role is omitted
+    elif model == "gpt-4-0314":
+        tokens_per_message = 3
+        tokens_per_name = 1
+    else:
+        logger.warn(f"num_tokens_from_messages() is not implemented for model {model}. Returning num tokens assuming gpt-3.5-turbo-0301.")
+        return num_tokens_from_messages(messages, model="gpt-3.5-turbo-0301")
+    num_tokens = 0
+    for message in messages:
+        num_tokens += tokens_per_message
+        for key, value in message.items():
+            num_tokens += len(encoding.encode(value))
+            if key == "name":
+                num_tokens += tokens_per_name
+    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>
+    return num_tokens

bot/openai/open_ai_bot.py

@@ -1,6 +1,9 @@
 # encoding:utf-8
 
 from bot.bot import Bot
+from bot.openai.open_ai_image import OpenAIImage
+from bot.openai.open_ai_session import OpenAISession
+from bot.session_manager import SessionManager
 from bridge.context import ContextType
 from bridge.reply import Reply, ReplyType
 from config import conf
@@ -11,8 +14,9 @@ import time
 user_session = dict()
 
 # OpenAI对话模型API (可用)
-class OpenAIBot(Bot):
+class OpenAIBot(Bot, OpenAIImage):
     def __init__(self):
+        super().__init__()
         openai.api_key = conf().get('open_ai_api_key')
         if conf().get('open_ai_api_base'):
             openai.api_base = conf().get('open_ai_api_base')
@@ -20,32 +24,43 @@ class OpenAIBot(Bot):
         if proxy:
             openai.proxy = proxy
 
+        self.sessions = SessionManager(OpenAISession, model= conf().get("model") or "text-davinci-003")
 
     def reply(self, query, context=None):
         # acquire reply content
         if context and context.type:
             if context.type == ContextType.TEXT:
                 logger.info("[OPEN_AI] query={}".format(query))
-                from_user_id = context['session_id']
+                session_id = context['session_id']
                 reply = None
                 if query == '#清除记忆':
-                    Session.clear_session(from_user_id)
+                    self.sessions.clear_session(session_id)
                     reply = Reply(ReplyType.INFO, '记忆已清除')
                 elif query == '#清除所有':
-                    Session.clear_all_session()
+                    self.sessions.clear_all_session()
                     reply = Reply(ReplyType.INFO, '所有人记忆已清除')
                 else:
-                    new_query = Session.build_session_query(query, from_user_id)
+                    session = self.sessions.session_query(query, session_id)
+                    new_query = str(session)
                     logger.debug("[OPEN_AI] session query={}".format(new_query))
 
-                    reply_content = self.reply_text(new_query, from_user_id, 0)
-                    logger.debug("[OPEN_AI] new_query={}, user={}, reply_cont={}".format(new_query, from_user_id, reply_content))
-                    if reply_content and query:
-                        Session.save_session(query, reply_content, from_user_id)
-                    reply = Reply(ReplyType.TEXT, reply_content)
+                    total_tokens, completion_tokens, reply_content = self.reply_text(new_query, session_id, 0)
+                    logger.debug("[OPEN_AI] new_query={}, session_id={}, reply_cont={}, completion_tokens={}".format(new_query, session_id, reply_content, completion_tokens))
+
+                    if total_tokens == 0 :
+                        reply = Reply(ReplyType.ERROR, reply_content)
+                    else:
+                        self.sessions.session_reply(reply_content, session_id, total_tokens)
+                        reply = Reply(ReplyType.TEXT, reply_content)
                 return reply
             elif context.type == ContextType.IMAGE_CREATE:
-                return self.create_img(query, 0)
+                ok, retstring = self.create_img(query, 0)
+                reply = None
+                if ok:
+                    reply = Reply(ReplyType.IMAGE_URL, retstring)
+                else:
+                    reply = Reply(ReplyType.ERROR, retstring)
+                return reply
 
     def reply_text(self, query, user_id, retry_count=0):
         try:
@@ -60,8 +75,10 @@ class OpenAIBot(Bot):
                 stop=["\n\n\n"]
             )
             res_content = response.choices[0]['text'].strip().replace('<|endoftext|>', '')
+            total_tokens = response["usage"]["total_tokens"]
+            completion_tokens = response["usage"]["completion_tokens"]
             logger.info("[OPEN_AI] reply={}".format(res_content))
-            return res_content
+            return total_tokens, completion_tokens, res_content
         except openai.error.RateLimitError as e:
             # rate limit exception
             logger.warn(e)
@@ -70,106 +87,9 @@ class OpenAIBot(Bot):
                 logger.warn("[OPEN_AI] RateLimit exceed, 第{}次重试".format(retry_count+1))
                 return self.reply_text(query, user_id, retry_count+1)
             else:
-                return "提问太快啦，请休息一下再问我吧"
+                return 0,0, "提问太快啦，请休息一下再问我吧"
         except Exception as e:
             # unknown exception
             logger.exception(e)
-            Session.clear_session(user_id)
-            return "请再问我一次吧"
-
-
-    def create_img(self, query, retry_count=0):
-        try:
-            logger.info("[OPEN_AI] image_query={}".format(query))
-            response = openai.Image.create(
-                prompt=query,    #图片描述
-                n=1,             #每次生成图片的数量
-                size="256x256"   #图片大小,可选有 256x256, 512x512, 1024x1024
-            )
-            image_url = response['data'][0]['url']
-            logger.info("[OPEN_AI] image_url={}".format(image_url))
-            return image_url
-        except openai.error.RateLimitError as e:
-            logger.warn(e)
-            if retry_count < 1:
-                time.sleep(5)
-                logger.warn("[OPEN_AI] ImgCreate RateLimit exceed, 第{}次重试".format(retry_count+1))
-                return self.reply_text(query, retry_count+1)
-            else:
-                return "提问太快啦，请休息一下再问我吧"
-        except Exception as e:
-            logger.exception(e)
-            return None
-
-
-class Session(object):
-    @staticmethod
-    def build_session_query(query, user_id):
-        '''
-        build query with conversation history
-        e.g.  Q: xxx
-              A: xxx
-              Q: xxx
-        :param query: query content
-        :param user_id: from user id
-        :return: query content with conversaction
-        '''
-        prompt = conf().get("character_desc", "")
-        if prompt:
-            prompt += "<|endoftext|>\n\n\n"
-        session = user_session.get(user_id, None)
-        if session:
-            for conversation in session:
-                prompt += "Q: " + conversation["question"] + "\n\n\nA: " + conversation["answer"] + "<|endoftext|>\n"
-            prompt += "Q: " + query + "\nA: "
-            return prompt
-        else:
-            return prompt + "Q: " + query + "\nA: "
-
-    @staticmethod
-    def save_session(query, answer, user_id):
-        max_tokens = conf().get("conversation_max_tokens")
-        if not max_tokens:
-            # default 3000
-            max_tokens = 1000
-        conversation = dict()
-        conversation["question"] = query
-        conversation["answer"] = answer
-        session = user_session.get(user_id)
-        logger.debug(conversation)
-        logger.debug(session)
-        if session:
-            # append conversation
-            session.append(conversation)
-        else:
-            # create session
-            queue = list()
-            queue.append(conversation)
-            user_session[user_id] = queue
-
-        # discard exceed limit conversation
-        Session.discard_exceed_conversation(user_session[user_id], max_tokens)
-
-
-    @staticmethod
-    def discard_exceed_conversation(session, max_tokens):
-        count = 0
-        count_list = list()
-        for i in range(len(session)-1, -1, -1):
-            # count tokens of conversation list
-            history_conv = session[i]
-            count += len(history_conv["question"]) + len(history_conv["answer"])
-            count_list.append(count)
-
-        for c in count_list:
-            if c > max_tokens:
-                # pop first conversation
-                session.pop(0)
-
-    @staticmethod
-    def clear_session(user_id):
-        user_session[user_id] = []
-
-    @staticmethod
-    def clear_all_session():
-        user_session.clear()
+            self.sessions.clear_session(user_id)
+            return 0,0, "请再问我一次吧"

bot/openai/open_ai_image.py

@@ -0,0 +1,37 @@
+import time
+import openai
+from common.token_bucket import TokenBucket
+from common.log import logger
+from config import conf
+
+# OPENAI提供的画图接口
+class OpenAIImage(object):
+    def __init__(self):
+        openai.api_key = conf().get('open_ai_api_key')
+        if conf().get('rate_limit_dalle'):
+            self.tb4dalle = TokenBucket(conf().get('rate_limit_dalle', 50))
+            
+    def create_img(self, query, retry_count=0):
+        try:
+            if conf().get('rate_limit_dalle') and not self.tb4dalle.get_token():
+                return False, "请求太快了，请休息一下再问我吧"
+            logger.info("[OPEN_AI] image_query={}".format(query))
+            response = openai.Image.create(
+                prompt=query,    #图片描述
+                n=1,             #每次生成图片的数量
+                size="256x256"   #图片大小,可选有 256x256, 512x512, 1024x1024
+            )
+            image_url = response['data'][0]['url']
+            logger.info("[OPEN_AI] image_url={}".format(image_url))
+            return True, image_url
+        except openai.error.RateLimitError as e:
+            logger.warn(e)
+            if retry_count < 1:
+                time.sleep(5)
+                logger.warn("[OPEN_AI] ImgCreate RateLimit exceed, 第{}次重试".format(retry_count+1))
+                return self.create_img(query, retry_count+1)
+            else:
+                return False, "提问太快啦，请休息一下再问我吧"
+        except Exception as e:
+            logger.exception(e)
+            return False, str(e)

bot/openai/open_ai_session.py

@@ -0,0 +1,77 @@
+from bot.session_manager import Session
+from common.log import logger
+class OpenAISession(Session):
+    def __init__(self, session_id, system_prompt=None, model= "text-davinci-003"):
+        super().__init__(session_id, system_prompt)
+        self.conversation = []
+        self.model = model
+        self.reset()
+    
+    def reset(self):
+        pass
+
+    def add_query(self, query):
+        question = {'type': 'question', 'content': query}
+        self.conversation.append(question)
+
+    def add_reply(self, reply):
+        answer = {'type': 'answer', 'content': reply}
+        self.conversation.append(answer)
+    def __str__(self):
+        '''
+        e.g.  Q: xxx
+              A: xxx
+              Q: xxx
+        '''
+        prompt = self.system_prompt
+        if prompt:
+            prompt += "<|endoftext|>\n\n\n"
+        for item in self.conversation:
+            if item['type'] == 'question':
+                prompt += "Q: " + item['content'] + "\n"
+            elif item['type'] == 'answer':
+                prompt += "\n\nA: " + item['content'] + "<|endoftext|>\n"
+
+        if len(self.conversation) > 0 and self.conversation[-1]['type'] == 'question':
+            prompt += "A: "
+        return prompt
+
+    def discard_exceeding(self, max_tokens, cur_tokens= None):
+        precise = True
+        try:
+            cur_tokens = num_tokens_from_string(str(self), self.model)
+        except Exception as e:
+            precise = False
+            if cur_tokens is None:
+                raise e
+            logger.debug("Exception when counting tokens precisely for query: {}".format(e))
+        while cur_tokens > max_tokens:
+            if len(self.conversation) > 1:
+                self.conversation.pop(0)
+            elif len(self.conversation) == 1 and self.conversation[0]["type"] == "answer":
+                self.conversation.pop(0)
+                if precise:
+                    cur_tokens = num_tokens_from_string(str(self), self.model)
+                else:
+                    cur_tokens = len(str(self))
+                break
+            elif len(self.conversation) == 1 and self.conversation[0]["type"] == "question":
+                logger.warn("user question exceed max_tokens. total_tokens={}".format(cur_tokens))
+                break
+            else:
+                logger.debug("max_tokens={}, total_tokens={}, len(conversation)={}".format(max_tokens, cur_tokens, len(self.conversation)))
+                break
+            if precise:
+                cur_tokens = num_tokens_from_string(str(self), self.model)
+            else:
+                cur_tokens = len(str(self))
+        return cur_tokens
+    
+
+# refer to https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
+def num_tokens_from_string(string: str, model: str) -> int:
+    """Returns the number of tokens in a text string."""
+    import tiktoken
+    encoding = tiktoken.encoding_for_model(model)
+    num_tokens = len(encoding.encode(string,disallowed_special=()))
+    return num_tokens

bot/session_manager.py

@@ -0,0 +1,81 @@
+from common.expired_dict import ExpiredDict
+from common.log import logger
+from config import conf
+
+class Session(object):
+    def __init__(self, session_id, system_prompt=None):
+        self.session_id = session_id
+        if system_prompt is None:
+            self.system_prompt = conf().get("character_desc", "")
+        else:
+            self.system_prompt = system_prompt
+
+    # 重置会话
+    def reset(self):
+        raise NotImplementedError
+
+    def set_system_prompt(self, system_prompt):
+        self.system_prompt = system_prompt
+        self.reset()
+
+    def add_query(self, query):
+        raise NotImplementedError
+
+    def add_reply(self, reply):
+        raise NotImplementedError
+    
+    def discard_exceeding(self, max_tokens=None, cur_tokens=None):
+        raise NotImplementedError
+
+
+
+class SessionManager(object):
+    def __init__(self, sessioncls, **session_args):
+        if conf().get('expires_in_seconds'):
+            sessions = ExpiredDict(conf().get('expires_in_seconds'))
+        else:
+            sessions = dict()
+        self.sessions = sessions
+        self.sessioncls = sessioncls
+        self.session_args = session_args
+
+    def build_session(self, session_id, system_prompt=None):
+        '''
+            如果session_id不在sessions中，创建一个新的session并添加到sessions中
+            如果system_prompt不会空，会更新session的system_prompt并重置session
+        '''
+        if session_id not in self.sessions:
+            self.sessions[session_id] = self.sessioncls(session_id, system_prompt, **self.session_args)
+        elif system_prompt is not None: # 如果有新的system_prompt，更新并重置session
+            self.sessions[session_id].set_system_prompt(system_prompt)
+        session = self.sessions[session_id]
+        return session
+    
+    def session_query(self, query, session_id):
+        session = self.build_session(session_id)
+        session.add_query(query)
+        try:
+            max_tokens = conf().get("conversation_max_tokens", 1000)
+            total_tokens = session.discard_exceeding(max_tokens, None)
+            logger.debug("prompt tokens used={}".format(total_tokens))
+        except Exception as e:
+            logger.debug("Exception when counting tokens precisely for prompt: {}".format(str(e)))
+        return session
+
+    def session_reply(self, reply, session_id, total_tokens = None):
+        session = self.build_session(session_id)
+        session.add_reply(reply)
+        try:
+            max_tokens = conf().get("conversation_max_tokens", 1000)
+            tokens_cnt = session.discard_exceeding(max_tokens, total_tokens)
+            logger.debug("raw total_tokens={}, savesession tokens={}".format(total_tokens, tokens_cnt))
+        except Exception as e:
+            logger.debug("Exception when counting tokens precisely for session: {}".format(str(e)))
+        return session
+
+    def clear_session(self, session_id):
+        if session_id in self.sessions:
+            del(self.sessions[session_id])
+
+    def clear_all_session(self):
+        self.sessions.clear()

plugins/dungeon/dungeon.py

@@ -52,7 +52,7 @@ class Dungeon(Plugin):
         if e_context['context'].type != ContextType.TEXT:
             return
         bottype = Bridge().get_bot_type("chat")
-        if bottype != const.CHATGPT:
+        if bottype not in (const.CHATGPT, const.OPEN_AI):
             return
         bot = Bridge().get_bot("chat")
         content = e_context['context'].content[:]

plugins/godcmd/godcmd.py

@@ -179,7 +179,7 @@ class Godcmd(Plugin):
                 elif cmd == "id":
                     ok, result = True, f"用户id=\n{user}"
                 elif cmd == "reset":
-                    if bottype == const.CHATGPT:
+                    if bottype in (const.CHATGPT, const.OPEN_AI):
                         bot.sessions.clear_session(session_id)
                         ok, result = True, "会话已重置"
                     else:
@@ -201,7 +201,7 @@ class Godcmd(Plugin):
                             load_config()
                             ok, result = True, "配置已重载"
                         elif cmd == "resetall":
-                            if bottype == const.CHATGPT:
+                            if bottype in (const.CHATGPT, const.OPEN_AI):
                                 bot.sessions.clear_all_session()
                                 ok, result = True, "重置所有会话成功"
                             else:

plugins/role/role.py

@@ -17,15 +17,15 @@ class RolePlay():
         self.sessionid = sessionid
         self.wrapper = wrapper or "%s"  # 用于包装用户输入
         self.desc = desc
+        self.bot.sessions.build_session(self.sessionid, system_prompt=self.desc)
 
     def reset(self):
         self.bot.sessions.clear_session(self.sessionid)
 
     def action(self, user_action):
-        session = self.bot.sessions.build_session(self.sessionid, self.desc)
-        if session[0]['role'] == 'system' and session[0]['content'] != self.desc: # 目前没有触发session过期事件，这里先简单判断，然后重置
-            self.reset()
-            self.bot.sessions.build_session(self.sessionid, self.desc)
+        session = self.bot.sessions.build_session(self.sessionid)
+        if session.system_prompt != self.desc: # 目前没有触发session过期事件，这里先简单判断，然后重置
+            session.set_system_prompt(self.desc)
         prompt = self.wrapper % user_action
         return prompt
 
@@ -74,7 +74,7 @@ class Role(Plugin):
         if e_context['context'].type != ContextType.TEXT:
             return
         bottype = Bridge().get_bot_type("chat")
-        if bottype != const.CHATGPT:
+        if bottype not in (const.CHATGPT, const.OPEN_AI):
             return
         bot = Bridge().get_bot("chat")
         content = e_context['context'].content[:]