From 0804969d3b86b68feaf0c2c4ac93a41068a731ee Mon Sep 17 00:00:00 2001 From: zhayujie Date: Sun, 5 Mar 2023 22:52:49 +0800 Subject: [PATCH] fix: auto release the oldest conversation #78 --- model/chatgpt/chatgpt_model.py | 52 +++++++++++++++++++++++++--------- 1 file changed, 39 insertions(+), 13 deletions(-) diff --git a/model/chatgpt/chatgpt_model.py b/model/chatgpt/chatgpt_model.py index baa3706..d31608f 100644 --- a/model/chatgpt/chatgpt_model.py +++ b/model/chatgpt/chatgpt_model.py @@ -20,23 +20,21 @@ class ChatGPTModel(Model): def reply(self, query, context=None): # acquire reply content if not context or not context.get('type') or context.get('type') == 'TEXT': - log.info("[OPEN_AI] query={}".format(query)) + log.info("[CHATGPT] query={}".format(query)) from_user_id = context['from_user_id'] if query == '#清除记忆': Session.clear_session(from_user_id) return '记忆已清除' new_query = Session.build_session_query(query, from_user_id) - log.debug("[OPEN_AI] session query={}".format(new_query)) + log.debug("[CHATGPT] session query={}".format(new_query)) # if context.get('stream'): # # reply in stream # return self.reply_text_stream(query, new_query, from_user_id) reply_content = self.reply_text(new_query, from_user_id, 0) - log.debug("[OPEN_AI] new_query={}, user={}, reply_cont={}".format(new_query, from_user_id, reply_content)) - if reply_content: - Session.save_session(query, reply_content, from_user_id) + #log.debug("[CHATGPT] new_query={}, user={}, reply_cont={}".format(new_query, from_user_id, reply_content)) return reply_content elif context.get('type', None) == 'IMAGE_CREATE': @@ -48,24 +46,35 @@ class ChatGPTModel(Model): model="gpt-3.5-turbo", # 对话模型的名称 messages=query, temperature=0.9, # 值在[0,1]之间，越大表示回复越具有不确定性 - #max_tokens=1200, # 回复最大的字符数 top_p=1, 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=0.0, # [-2,2]之间，该值越大则更倾向于产生不同的内容 ) - # res_content = response.choices[0]['text'].strip().replace('<|endoftext|>', '') - log.info(response.choices[0]['message']['content']) - # log.info("[OPEN_AI] reply={}".format(res_content)) + reply_content = response.choices[0]['message']['content'] + used_token = response['usage']['total_tokens'] + log.debug(response) + log.info("[CHATGPT] reply={}", reply_content) + if reply_content: + # save conversation + Session.save_session(query, reply_content, user_id, used_token) return response.choices[0]['message']['content'] except openai.error.RateLimitError as e: # rate limit exception log.warn(e) if retry_count < 1: time.sleep(5) - log.warn("[OPEN_AI] RateLimit exceed, 第{}次重试".format(retry_count+1)) + log.warn("[CHATGPT] 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" + except openai.error.APIConnectionError as e: + log.warn(e) + log.warn("[CHATGPT] APIConnection failed") + return "我连接不到你的网络" + except openai.error.Timeout as e: + log.warn(e) + log.warn("[CHATGPT] Timeout") + return "我没有收到你的消息" except Exception as e: # unknown exception log.exception(e) @@ -93,10 +102,18 @@ class ChatGPTModel(Model): log.warn(e) if retry_count < 1: time.sleep(5) - log.warn("[OPEN_AI] RateLimit exceed, 第{}次重试".format(retry_count+1)) - return self.reply_text(query, user_id, retry_count+1) + log.warn("[CHATGPT] RateLimit exceed, 第{}次重试".format(retry_count+1)) + return self.reply_text_stream(query, user_id, retry_count+1) else: return "提问太快啦，请休息一下再问我吧" + except openai.error.APIConnectionError as e: + log.warn(e) + log.warn("[CHATGPT] APIConnection failed") + return "我连接不到你的网络" + except openai.error.Timeout as e: + log.warn(e) + log.warn("[CHATGPT] Timeout") + return "我没有收到你的消息" except Exception as e: # unknown exception log.exception(e) @@ -176,13 +193,22 @@ class Session(object): return session @staticmethod - def save_session(query, answer, user_id): + def save_session(query, answer, user_id, used_tokens=0): + max_tokens = model_conf(const.OPEN_AI).get('conversation_max_tokens') + if not max_tokens or max_tokens > 4000: + # default value + max_tokens = 1000 session = user_session.get(user_id) if session: # append conversation gpt_item = {'role': 'assistant', 'content': answer} session.append(gpt_item) + if used_tokens > max_tokens and len(session) >= 3: + # pop first conversation (TODO: more accurate calculation) + session.pop(1) + session.pop(1) + @staticmethod def clear_session(user_id): user_session[user_id] = []