MemoryScope：为LLM聊天机器人配备的长期记忆系统

# 下载memoryscope仓库git clone https://github.com/modelscope/memoryscope
# 填写DashScope API-KEYvim memoryscope/docker-compose.yml
# 运行！(P.S. 实现双窗口监控实时memory，请参考https://github.com/modelscope/MemoryScope/blob/main/docs/installation_zh.md）docker-compose run memory_scope_main

pip install memoryscope
sudo docker run -p 9200:9200 \-e "discovery.type=single-node" \-e "xpack.security.enabled=false" \-e "xpack.license.self_generated.type=trial" \docker.elastic.co/elasticsearch/elasticsearch:8.13.2

# dashboardexport DASHSCOPE_API_KEY="sk-0000000000"# openaiexport OPENAI_API_KEY="sk-000000000000000000000"# openai base urlexport OPENAI_API_BASE="https://00000.0000.000/v1"

from memoryscope import MemoryScope, Argumentsarguments = Arguments(language="cn",human_name="用户",assistant_name="AI",memory_chat_class="api_memory_chat",generation_backend="dashscope_generation",generation_model="qwen2-72b-instruct",embedding_backend="dashscope_embedding",embedding_model="text-embedding-v2",rank_backend="dashscope_rank",rank_model="gte-rerank",enable_ranker=True,worker_params={"get_reflection_subject": {"reflect_num_questions": 3}})ms = MemoryScope(arguments=arguments)ms.default_memory_service.init_service()

qs = ["I love playing the piano.", "I work at Alibaba.", "I really like eating watermelon.", "Tomorrow I plan to go to Beijing for a business trip.", "What fruit is good to eat this afternoon?", ]for query in qs:# 抽取之前的记忆，并对话response = memory_chat.chat_with_memory(query)print(query, '\n', response.message.content)# 巩固对新产生的对话的记忆result = memory_service.consolidate_memory()# 列举当前存储的记忆print(memory_chat.run_service_operation("list_memory"))# 记忆的反思与再巩固result = memory_service.reflect_and_reconsolidate()

class MemoryScopeAgent(ConversableAgent):def __init__(self,name: str = "assistant",system_message: Optional[str] = "",human_input_mode: Literal["ALWAYS", "NEVER", "TERMINATE"] = "NEVER",llm_config: Optional[Union[Dict, bool]] = None,arguments: Arguments = None,**kwargs,):super().__init__(name=name,system_message=system_message,human_input_mode=human_input_mode,llm_config=llm_config,**kwargs,)
# Create a memory client in MemoryScopeself.memory_scope = MemoryScope(arguments=arguments)self.memory_chat = self.memory_scope.default_memory_chat
self.register_reply([Agent, None], MemoryScopeAgent.generate_reply_with_memory, remove_other_reply_funcs=True)
def generate_reply_with_memory(self,messages: Optional[List[Dict]] = None,sender: Optional[Agent] = None,config: Optional[Any] = None,) -> Tuple[bool, Union[str, Dict, None]]:# Generate response
contents = []for message in messages:if message.get("role") != self.name:contents.append(message.get("content", ""))
query = contents[-1]response = self.memory_chat.chat_with_memory(query=query)return True, response.message.content
def close(self):self.memory_scope.close()

from typing import Optional, Union, Literal, Dict, List, Any, Tuple
from autogen import Agent, ConversableAgent, UserProxyAgent
from memoryscope import MemoryScope, Arguments
from where_you_define_the_new_agent import MemoryScopeAgent
def main():# Create the agent of MemoryScopearguments = Arguments(language="cn",human_name="用户",assistant_name="AI",memory_chat_class="api_memory_chat",generation_backend="dashscope_generation",generation_model="qwen-max",embedding_backend="dashscope_embedding",embedding_model="text-embedding-v2",rank_backend="dashscope_rank",rank_model="gte-rerank")
assistant = MemoryScopeAgent("assistant", arguments=arguments)
# Create the agent that represents the user in the conversation.user_proxy = UserProxyAgent("user", code_execution_config=False)
# Let the assistant start the conversation.It will end when the user types exit.assistant.initiate_chat(user_proxy, message="有什么需要帮忙的吗？")assistant.close()

if __name__ == "__main__":main()

@software{MemoryScope,author = {Li Yu and Tiancheng Qin andQingxu Fu andSen Huang andXianzhe Xu andZhaoyang Liu andBoyin Liu},month = {09},title = {{MemoryScope}},url = {https://github.com/modelscope/MemoryScope},year = {2024}}