fix: 增大了默认的maxtoken防止溢出，messagecq改异步get_image防止阻塞
This commit is contained in:
tcmofashi
@@ -14,24 +14,24 @@ from .chat_stream import chat_manager
|
||||
|
||||
class PromptBuilder:
|
||||
def __init__(self):
|
||||
self.prompt_built = ''
|
||||
self.activate_messages = ''
|
||||
self.prompt_built = ""
|
||||
self.activate_messages = ""
|
||||
|
||||
|
||||
|
||||
async def _build_prompt(self,
|
||||
message_txt: str,
|
||||
sender_name: str = "某人",
|
||||
relationship_value: float = 0.0,
|
||||
stream_id: Optional[int] = None) -> tuple[str, str]:
|
||||
async def _build_prompt(
|
||||
self,
|
||||
message_txt: str,
|
||||
sender_name: str = "某人",
|
||||
relationship_value: float = 0.0,
|
||||
stream_id: Optional[int] = None,
|
||||
) -> tuple[str, str]:
|
||||
"""构建prompt
|
||||
|
||||
|
||||
Args:
|
||||
message_txt: 消息文本
|
||||
sender_name: 发送者昵称
|
||||
relationship_value: 关系值
|
||||
group_id: 群组ID
|
||||
|
||||
|
||||
Returns:
|
||||
str: 构建好的prompt
|
||||
"""
|
||||
@@ -56,46 +56,43 @@ class PromptBuilder:
|
||||
current_date = time.strftime("%Y-%m-%d", time.localtime())
|
||||
current_time = time.strftime("%H:%M:%S", time.localtime())
|
||||
bot_schedule_now_time, bot_schedule_now_activity = bot_schedule.get_current_task()
|
||||
prompt_date = f'''今天是{current_date}，现在是{current_time}，你今天的日程是：\n{bot_schedule.today_schedule}\n你现在正在{bot_schedule_now_activity}\n'''
|
||||
prompt_date = f"""今天是{current_date}，现在是{current_time}，你今天的日程是：\n{bot_schedule.today_schedule}\n你现在正在{bot_schedule_now_activity}\n"""
|
||||
|
||||
# 知识构建
|
||||
start_time = time.time()
|
||||
|
||||
prompt_info = ''
|
||||
promt_info_prompt = ''
|
||||
prompt_info = ""
|
||||
promt_info_prompt = ""
|
||||
prompt_info = await self.get_prompt_info(message_txt, threshold=0.5)
|
||||
if prompt_info:
|
||||
prompt_info = f'''你有以下这些[知识]：{prompt_info}请你记住上面的[
|
||||
知识]，之后可能会用到-'''
|
||||
prompt_info = f"""你有以下这些[知识]：{prompt_info}请你记住上面的[
|
||||
知识]，之后可能会用到-"""
|
||||
|
||||
end_time = time.time()
|
||||
logger.debug(f"知识检索耗时: {(end_time - start_time):.3f}秒")
|
||||
|
||||
# 获取聊天上下文
|
||||
chat_in_group=True
|
||||
chat_talking_prompt = ''
|
||||
chat_in_group = True
|
||||
chat_talking_prompt = ""
|
||||
if stream_id:
|
||||
chat_talking_prompt = get_recent_group_detailed_plain_text(stream_id, limit=global_config.MAX_CONTEXT_SIZE,combine = True)
|
||||
chat_stream=chat_manager.get_stream(stream_id)
|
||||
chat_talking_prompt = get_recent_group_detailed_plain_text(
|
||||
stream_id, limit=global_config.MAX_CONTEXT_SIZE, combine=True
|
||||
)
|
||||
chat_stream = chat_manager.get_stream(stream_id)
|
||||
if chat_stream.group_info:
|
||||
chat_talking_prompt = f"以下是群里正在聊天的内容：\n{chat_talking_prompt}"
|
||||
else:
|
||||
chat_in_group=False
|
||||
chat_in_group = False
|
||||
chat_talking_prompt = f"以下是你正在和{sender_name}私聊的内容：\n{chat_talking_prompt}"
|
||||
# print(f"\033[1;34m[调试]\033[0m 已从数据库获取群 {group_id} 的消息记录:{chat_talking_prompt}")
|
||||
|
||||
|
||||
|
||||
|
||||
# 使用新的记忆获取方法
|
||||
memory_prompt = ''
|
||||
memory_prompt = ""
|
||||
start_time = time.time()
|
||||
|
||||
# 调用 hippocampus 的 get_relevant_memories 方法
|
||||
relevant_memories = await hippocampus.get_relevant_memories(
|
||||
text=message_txt,
|
||||
max_topics=5,
|
||||
similarity_threshold=0.4,
|
||||
max_memory_num=5
|
||||
text=message_txt, max_topics=5, similarity_threshold=0.4, max_memory_num=5
|
||||
)
|
||||
|
||||
if relevant_memories:
|
||||
@@ -115,56 +112,58 @@ class PromptBuilder:
|
||||
logger.info(f"回忆耗时: {(end_time - start_time):.3f}秒")
|
||||
|
||||
# 激活prompt构建
|
||||
activate_prompt = ''
|
||||
activate_prompt = ""
|
||||
if chat_in_group:
|
||||
activate_prompt = f"以上是群里正在进行的聊天，{memory_prompt} 现在昵称为 '{sender_name}' 的用户说的:{message_txt}。引起了你的注意,你和ta{relation_prompt},{mood_prompt},你想要{relation_prompt_2}。"
|
||||
activate_prompt = f"以上是群里正在进行的聊天，{memory_prompt} 现在昵称为 '{sender_name}' 的用户说的:{message_txt}。引起了你的注意,你和ta{relation_prompt},{mood_prompt},你想要{relation_prompt_2}。"
|
||||
else:
|
||||
activate_prompt = f"以上是你正在和{sender_name}私聊的内容，{memory_prompt} 现在昵称为 '{sender_name}' 的用户说的:{message_txt}。引起了你的注意,你和ta{relation_prompt},{mood_prompt},你想要{relation_prompt_2}。"
|
||||
|
||||
# 关键词检测与反应
|
||||
keywords_reaction_prompt = ''
|
||||
keywords_reaction_prompt = ""
|
||||
for rule in global_config.keywords_reaction_rules:
|
||||
if rule.get("enable", False):
|
||||
if any(keyword in message_txt.lower() for keyword in rule.get("keywords", [])):
|
||||
logger.info(f"检测到以下关键词之一：{rule.get('keywords', [])}，触发反应：{rule.get('reaction', '')}")
|
||||
keywords_reaction_prompt += rule.get("reaction", "") + '，'
|
||||
|
||||
#人格选择
|
||||
personality=global_config.PROMPT_PERSONALITY
|
||||
logger.info(
|
||||
f"检测到以下关键词之一：{rule.get('keywords', [])}，触发反应：{rule.get('reaction', '')}"
|
||||
)
|
||||
keywords_reaction_prompt += rule.get("reaction", "") + "，"
|
||||
|
||||
# 人格选择
|
||||
personality = global_config.PROMPT_PERSONALITY
|
||||
probability_1 = global_config.PERSONALITY_1
|
||||
probability_2 = global_config.PERSONALITY_2
|
||||
probability_3 = global_config.PERSONALITY_3
|
||||
|
||||
prompt_personality = f'{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，你还有很多别名:{"/".join(global_config.BOT_ALIAS_NAMES)}，'
|
||||
|
||||
prompt_personality = f"{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，你还有很多别名:{'/'.join(global_config.BOT_ALIAS_NAMES)}，"
|
||||
personality_choice = random.random()
|
||||
if chat_in_group:
|
||||
prompt_in_group=f"你正在浏览{chat_stream.platform}群"
|
||||
prompt_in_group = f"你正在浏览{chat_stream.platform}群"
|
||||
else:
|
||||
prompt_in_group=f"你正在{chat_stream.platform}上和{sender_name}私聊"
|
||||
prompt_in_group = f"你正在{chat_stream.platform}上和{sender_name}私聊"
|
||||
if personality_choice < probability_1: # 第一种人格
|
||||
prompt_personality += f'''{personality[0]}, 你正在浏览qq群,{promt_info_prompt},
|
||||
prompt_personality += f"""{personality[0]}, 你正在浏览qq群,{promt_info_prompt},
|
||||
现在请你给出日常且口语化的回复，平淡一些，尽量简短一些。{keywords_reaction_prompt}
|
||||
请注意把握群里的聊天内容，不要刻意突出自身学科背景，不要回复的太有条理，可以有个性。'''
|
||||
请注意把握群里的聊天内容，不要刻意突出自身学科背景，不要回复的太有条理，可以有个性。"""
|
||||
elif personality_choice < probability_1 + probability_2: # 第二种人格
|
||||
prompt_personality += f'''{personality[1]}, 你正在浏览qq群，{promt_info_prompt},
|
||||
prompt_personality += f"""{personality[1]}, 你正在浏览qq群，{promt_info_prompt},
|
||||
现在请你给出日常且口语化的回复，请表现你自己的见解，不要一昧迎合，尽量简短一些。{keywords_reaction_prompt}
|
||||
请你表达自己的见解和观点。可以有个性。'''
|
||||
请你表达自己的见解和观点。可以有个性。"""
|
||||
else: # 第三种人格
|
||||
prompt_personality += f'''{personality[2]}, 你正在浏览qq群，{promt_info_prompt},
|
||||
prompt_personality += f"""{personality[2]}, 你正在浏览qq群，{promt_info_prompt},
|
||||
现在请你给出日常且口语化的回复，请表现你自己的见解，不要一昧迎合，尽量简短一些。{keywords_reaction_prompt}
|
||||
请你表达自己的见解和观点。可以有个性。'''
|
||||
请你表达自己的见解和观点。可以有个性。"""
|
||||
|
||||
# 中文高手(新加的好玩功能)
|
||||
prompt_ger = ''
|
||||
prompt_ger = ""
|
||||
if random.random() < 0.04:
|
||||
prompt_ger += '你喜欢用倒装句'
|
||||
prompt_ger += "你喜欢用倒装句"
|
||||
if random.random() < 0.02:
|
||||
prompt_ger += '你喜欢用反问句'
|
||||
prompt_ger += "你喜欢用反问句"
|
||||
if random.random() < 0.01:
|
||||
prompt_ger += '你喜欢用文言文'
|
||||
prompt_ger += "你喜欢用文言文"
|
||||
|
||||
# 额外信息要求
|
||||
extra_info = '''但是记得回复平淡一些，简短一些，尤其注意在没明确提到时不要过多提及自身的背景, 不要直接回复别人发的表情包，记住不要输出多余内容(包括前后缀，冒号和引号，括号，表情等)，只需要输出回复内容就好，不要输出其他任何内容'''
|
||||
extra_info = """但是记得回复平淡一些，简短一些，尤其注意在没明确提到时不要过多提及自身的背景, 不要直接回复别人发的表情包，记住不要输出多余内容(包括前后缀，冒号和引号，括号，表情等)，只需要输出回复内容就好，不要输出其他任何内容"""
|
||||
|
||||
# 合并prompt
|
||||
prompt = ""
|
||||
@@ -175,16 +174,16 @@ class PromptBuilder:
|
||||
prompt += f"{prompt_ger}\n"
|
||||
prompt += f"{extra_info}\n"
|
||||
|
||||
'''读空气prompt处理'''
|
||||
"""读空气prompt处理"""
|
||||
activate_prompt_check = f"以上是群里正在进行的聊天，昵称为 '{sender_name}' 的用户说的:{message_txt}。引起了你的注意,你和他{relation_prompt}，你想要{relation_prompt_2}，但是这不一定是合适的时机，请你决定是否要回应这条消息。"
|
||||
prompt_personality_check = ''
|
||||
prompt_personality_check = ""
|
||||
extra_check_info = f"请注意把握群里的聊天内容的基础上，综合群内的氛围，例如，和{global_config.BOT_NICKNAME}相关的话题要积极回复,如果是at自己的消息一定要回复，如果自己正在和别人聊天一定要回复，其他话题如果合适搭话也可以回复，如果认为应该回复请输出yes，否则输出no，请注意是决定是否需要回复，而不是编写回复内容，除了yes和no不要输出任何回复内容。"
|
||||
if personality_choice < probability_1: # 第一种人格
|
||||
prompt_personality_check = f'''你的网名叫{global_config.BOT_NICKNAME}，{personality[0]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}'''
|
||||
prompt_personality_check = f"""你的网名叫{global_config.BOT_NICKNAME}，{personality[0]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}"""
|
||||
elif personality_choice < probability_1 + probability_2: # 第二种人格
|
||||
prompt_personality_check = f'''你的网名叫{global_config.BOT_NICKNAME}，{personality[1]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}'''
|
||||
prompt_personality_check = f"""你的网名叫{global_config.BOT_NICKNAME}，{personality[1]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}"""
|
||||
else: # 第三种人格
|
||||
prompt_personality_check = f'''你的网名叫{global_config.BOT_NICKNAME}，{personality[2]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}'''
|
||||
prompt_personality_check = f"""你的网名叫{global_config.BOT_NICKNAME}，{personality[2]}, 你正在浏览qq群，{promt_info_prompt} {activate_prompt_check} {extra_check_info}"""
|
||||
|
||||
prompt_check_if_response = f"{prompt_info}\n{prompt_date}\n{chat_talking_prompt}\n{prompt_personality_check}"
|
||||
|
||||
@@ -194,38 +193,38 @@ class PromptBuilder:
|
||||
current_date = time.strftime("%Y-%m-%d", time.localtime())
|
||||
current_time = time.strftime("%H:%M:%S", time.localtime())
|
||||
bot_schedule_now_time, bot_schedule_now_activity = bot_schedule.get_current_task()
|
||||
prompt_date = f'''今天是{current_date}，现在是{current_time}，你今天的日程是：\n{bot_schedule.today_schedule}\n你现在正在{bot_schedule_now_activity}\n'''
|
||||
prompt_date = f"""今天是{current_date}，现在是{current_time}，你今天的日程是：\n{bot_schedule.today_schedule}\n你现在正在{bot_schedule_now_activity}\n"""
|
||||
|
||||
chat_talking_prompt = ''
|
||||
chat_talking_prompt = ""
|
||||
if group_id:
|
||||
chat_talking_prompt = get_recent_group_detailed_plain_text(group_id,
|
||||
limit=global_config.MAX_CONTEXT_SIZE,
|
||||
combine=True)
|
||||
chat_talking_prompt = get_recent_group_detailed_plain_text(
|
||||
group_id, limit=global_config.MAX_CONTEXT_SIZE, combine=True
|
||||
)
|
||||
|
||||
chat_talking_prompt = f"以下是群里正在聊天的内容：\n{chat_talking_prompt}"
|
||||
# print(f"\033[1;34m[调试]\033[0m 已从数据库获取群 {group_id} 的消息记录:{chat_talking_prompt}")
|
||||
|
||||
# 获取主动发言的话题
|
||||
all_nodes = memory_graph.dots
|
||||
all_nodes = filter(lambda dot: len(dot[1]['memory_items']) > 3, all_nodes)
|
||||
all_nodes = filter(lambda dot: len(dot[1]["memory_items"]) > 3, all_nodes)
|
||||
nodes_for_select = random.sample(all_nodes, 5)
|
||||
topics = [info[0] for info in nodes_for_select]
|
||||
infos = [info[1] for info in nodes_for_select]
|
||||
|
||||
# 激活prompt构建
|
||||
activate_prompt = ''
|
||||
activate_prompt = ""
|
||||
activate_prompt = "以上是群里正在进行的聊天。"
|
||||
personality = global_config.PROMPT_PERSONALITY
|
||||
prompt_personality = ''
|
||||
prompt_personality = ""
|
||||
personality_choice = random.random()
|
||||
if personality_choice < probability_1: # 第一种人格
|
||||
prompt_personality = f'''{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[0]}'''
|
||||
prompt_personality = f"""{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[0]}"""
|
||||
elif personality_choice < probability_1 + probability_2: # 第二种人格
|
||||
prompt_personality = f'''{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[1]}'''
|
||||
prompt_personality = f"""{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[1]}"""
|
||||
else: # 第三种人格
|
||||
prompt_personality = f'''{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[2]}'''
|
||||
prompt_personality = f"""{activate_prompt}你的网名叫{global_config.BOT_NICKNAME}，{personality[2]}"""
|
||||
|
||||
topics_str = ','.join(f"\"{topics}\"")
|
||||
topics_str = ",".join(f'"{topics}"')
|
||||
prompt_for_select = f"你现在想在群里发言，回忆了一下，想到几个话题，分别是{topics_str}，综合当前状态以及群内气氛，请你在其中选择一个合适的话题，注意只需要输出话题，除了话题什么也不要输出(双引号也不要输出)"
|
||||
|
||||
prompt_initiative_select = f"{prompt_date}\n{prompt_personality}\n{prompt_for_select}"
|
||||
@@ -234,8 +233,8 @@ class PromptBuilder:
|
||||
return prompt_initiative_select, nodes_for_select, prompt_regular
|
||||
|
||||
def _build_initiative_prompt_check(self, selected_node, prompt_regular):
|
||||
memory = random.sample(selected_node['memory_items'], 3)
|
||||
memory = '\n'.join(memory)
|
||||
memory = random.sample(selected_node["memory_items"], 3)
|
||||
memory = "\n".join(memory)
|
||||
prompt_for_check = f"{prompt_regular}你现在想在群里发言，回忆了一下，想到一个话题,是{selected_node['concept']}，关于这个话题的记忆有\n{memory}\n，以这个作为主题发言合适吗？请在把握群里的聊天内容的基础上，综合群内的氛围，如果认为应该发言请输出yes，否则输出no，请注意是决定是否需要发言，而不是编写回复内容，除了yes和no不要输出任何回复内容。"
|
||||
return prompt_for_check, memory
|
||||
|
||||
@@ -244,7 +243,7 @@ class PromptBuilder:
|
||||
return prompt_for_initiative
|
||||
|
||||
async def get_prompt_info(self, message: str, threshold: float):
|
||||
related_info = ''
|
||||
related_info = ""
|
||||
logger.debug(f"获取知识库内容，元消息：{message[:30]}...，消息长度: {len(message)}")
|
||||
embedding = await get_embedding(message)
|
||||
related_info += self.get_info_from_db(embedding, threshold=threshold)
|
||||
@@ -253,7 +252,7 @@ class PromptBuilder:
|
||||
|
||||
def get_info_from_db(self, query_embedding: list, limit: int = 1, threshold: float = 0.5) -> str:
|
||||
if not query_embedding:
|
||||
return ''
|
||||
return ""
|
||||
# 使用余弦相似度计算
|
||||
pipeline = [
|
||||
{
|
||||
@@ -265,12 +264,14 @@ class PromptBuilder:
|
||||
"in": {
|
||||
"$add": [
|
||||
"$$value",
|
||||
{"$multiply": [
|
||||
{"$arrayElemAt": ["$embedding", "$$this"]},
|
||||
{"$arrayElemAt": [query_embedding, "$$this"]}
|
||||
]}
|
||||
{
|
||||
"$multiply": [
|
||||
{"$arrayElemAt": ["$embedding", "$$this"]},
|
||||
{"$arrayElemAt": [query_embedding, "$$this"]},
|
||||
]
|
||||
},
|
||||
]
|
||||
}
|
||||
},
|
||||
}
|
||||
},
|
||||
"magnitude1": {
|
||||
@@ -278,7 +279,7 @@ class PromptBuilder:
|
||||
"$reduce": {
|
||||
"input": "$embedding",
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]}
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
},
|
||||
@@ -287,19 +288,13 @@ class PromptBuilder:
|
||||
"$reduce": {
|
||||
"input": query_embedding,
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]}
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$addFields": {
|
||||
"similarity": {
|
||||
"$divide": ["$dotProduct", {"$multiply": ["$magnitude1", "$magnitude2"]}]
|
||||
}
|
||||
},
|
||||
}
|
||||
},
|
||||
{"$addFields": {"similarity": {"$divide": ["$dotProduct", {"$multiply": ["$magnitude1", "$magnitude2"]}]}}},
|
||||
{
|
||||
"$match": {
|
||||
"similarity": {"$gte": threshold} # 只保留相似度大于等于阈值的结果
|
||||
@@ -307,17 +302,17 @@ class PromptBuilder:
|
||||
},
|
||||
{"$sort": {"similarity": -1}},
|
||||
{"$limit": limit},
|
||||
{"$project": {"content": 1, "similarity": 1}}
|
||||
{"$project": {"content": 1, "similarity": 1}},
|
||||
]
|
||||
|
||||
results = list(db.knowledges.aggregate(pipeline))
|
||||
# print(f"\033[1;34m[调试]\033[0m获取知识库内容结果: {results}")
|
||||
|
||||
if not results:
|
||||
return ''
|
||||
return ""
|
||||
|
||||
# 返回所有找到的内容，用换行分隔
|
||||
return '\n'.join(str(result['content']) for result in results)
|
||||
return "\n".join(str(result["content"]) for result in results)
|
||||
|
||||
|
||||
prompt_builder = PromptBuilder()
|
||||
|
||||
Reference in New Issue
Block a user