diff --git a/bot/chatgpt/chat_gpt_bot.py b/bot/chatgpt/chat_gpt_bot.py index 657aa6b..d4523cd 100644 --- a/bot/chatgpt/chat_gpt_bot.py +++ b/bot/chatgpt/chat_gpt_bot.py @@ -1,7 +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 @@ -11,7 +13,6 @@ from common.expired_dict import ExpiredDict import openai import time - # OpenAI对话模型API (可用) class ChatGPTBot(Bot,OpenAIImage): def __init__(self): @@ -19,7 +20,7 @@ class ChatGPTBot(Bot,OpenAIImage): 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") + self.sessions = SessionManager(ChatGPTSession, model= conf().get("model") or "gpt-3.5-turbo") if proxy: openai.proxy = proxy if conf().get('rate_limit_chatgpt'): @@ -44,19 +45,19 @@ class ChatGPTBot(Bot,OpenAIImage): 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']) @@ -85,7 +86,7 @@ class ChatGPTBot(Bot,OpenAIImage): "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 @@ -97,7 +98,7 @@ class ChatGPTBot(Bot,OpenAIImage): 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"], @@ -128,7 +129,6 @@ class ChatGPTBot(Bot,OpenAIImage): return {"completion_tokens": 0, "content": "请再问我一次吧"} - class AzureChatGPTBot(ChatGPTBot): def __init__(self): super().__init__() @@ -139,123 +139,4 @@ 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 \ No newline at end of file + return args \ No newline at end of file diff --git a/bot/chatgpt/chat_gpt_session.py b/bot/chatgpt/chat_gpt_session.py new file mode 100644 index 0000000..c50d486 --- /dev/null +++ b/bot/chatgpt/chat_gpt_session.py @@ -0,0 +1,76 @@ +from bot.session_manager import Session +from common.log import logger +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): + if cur_tokens is None: + cur_tokens = num_tokens_from_messages(self.messages, self.model) + 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) + cur_tokens = num_tokens_from_messages(self.messages, self.model) + 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 + try: + cur_tokens = num_tokens_from_messages(self.messages, self.model) + except Exception as e: + logger.debug("Exception when counting tokens precisely for query: {}".format(e)) + 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 \ No newline at end of file diff --git a/bot/session_manager.py b/bot/session_manager.py new file mode 100644 index 0000000..4494a32 --- /dev/null +++ b/bot/session_manager.py @@ -0,0 +1,77 @@ +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): + 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) + print(session.messages) + 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 prompt: {}".format(str(e))) + return session + + def clear_session(self, session_id): + del(self.sessions[session_id]) + + def clear_all_session(self): + self.sessions.clear()