gyt/opendan
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Support Text summary based Knowledge System,

Browse Source 
Update Agent Workspace.
This commit is contained in:
Liu Zhicong
2023-11-12 21:59:04 -08:00
parent fb0b88d44a
commit 763360b305
12 changed files with 765 additions and 257 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/aios_kernel/agent.py
+135 -69
View File
@@ -485,12 +485,12 @@ class AIAgent:
summary = self.llm_select_session_summary(msg,chatsession)
prompt.append(AgentPrompt(summary))
known_info_str = "# 已知信息\n"
known_info_str = "# Known information\n"
have_known_info = False
todos_str,todo_count = await workspace.get_todo_tree()
if todo_count > 0:
have_known_info = True
known_info_str += f"## 已有todo\n{todos_str}\n"
known_info_str += f"## todo\n{todos_str}\n"
inner_functions,function_token_len = self._get_inner_functions()
system_prompt_len = prompt.get_prompt_token_len()
input_len = len(msg.body)
@@ -879,40 +879,63 @@ class AIAgent:
# 尝试自我学习，会主动获取、读取资料并进行整理
# LLM的本质能力是处理海量知识，应该让LLM能基于知识把自己的工作处理的更好
def do_self_learn(self) -> None:
async def do_self_learn(self) -> None:
# 不同的workspace是否应该有不同的学习方法？
learn_power = self.get_learn_power()
kb = self.get_knowledge_base()
for item in kb.un_learn_items():
if learn_power <= 0:
workspace = self.get_workspace_by_msg(None)
hash_list = workspace.kb_db.get_knowledge_without_llm_title()
for hash in hash_list:
if self.agent_energy <= 0:
break
match item.type():
case "book":
self.llm_read_book(kb,item)
learn_power -= 1
case "article":
# 可以用vdb 对不同目录的名字进行选择后，先进行一次快速的插入。有时间再慢慢用LLM整理
self.llm_read_article(kb,item)
learn_power -= 1
case "video":
self.llm_watch_video(kb,item)
learn_power -= 1
case "audio":
self.llm_listen_audio(kb,item)
learn_power -= 1
case "code_project":
self.llm_read_code_project(kb,item)
learn_power -= 1
case "image":
self.llm_view_image(kb,item)
learn_power -= 1
case "other":
self.llm_read_other(kb,item)
learn_power -= 1
case _:
self.llm_learn_any(kb,item)
pass
knowledge = workspace.kb_db.get_knowledge_by_hash(hash)
if knowledge is None:
continue
if os.path.exists(knowledge.path) is False:
logger.warning(f"do_self_learn: knowledge {knowledge.path} is not exists!")
continue
#TODO 可以用v-db 对不同目录的名字进行选择后，先进行一次快速的插入。有时间再慢慢用LLM整理
llm_result = await self._llm_read_article(knowledge)
#根据结果更新knowledge
if llm_result is not None:
workspace.kb_db.update_knowledge_by_hash(hash,llm_result)
# 在知识库中创建软链接
self.agent_energy -= 1
# match item.type():
# case "book":
# self.llm_read_book(kb,item)
# learn_power -= 1
# case "article":
#
# self.llm_read_article(kb,item)
# learn_power -= 1
# case "video":
# self.llm_watch_video(kb,item)
# learn_power -= 1
# case "audio":
# self.llm_listen_audio(kb,item)
# learn_power -= 1
# case "code_project":
# self.llm_read_code_project(kb,item)
# learn_power -= 1
# case "image":
# self.llm_view_image(kb,item)
# learn_power -= 1
# case "other":
# self.llm_read_other(kb,item)
# learn_power -= 1
# case _:
# self.llm_learn_any(kb,item)
# pass
async def do_blance_knowledge_base(selft):
# 整理自己的知识库(让分类更平衡，更由于自己以后的工作)，并尝试更新学习目标
current_path = "/"
current_list = kb.get_list(current_path)
@@ -933,48 +956,86 @@ class AIAgent:
def parser_learn_llm_result(self,llm_result:LLMResult):
pass
async def _llm_read_article(self,item:KnowledgeObject) -> ComputeTaskResult:
full_content = item.get_article_full_content()
full_content_len = ComputeKernel.llm_num_tokens_from_text(full_content,self.get_llm_model_name())
async def gen_known_info_for_knowledge_prompt(self,knowledge_item:dict,need_catalogs = False) -> AgentPrompt:
#已知信息：
# 组织的工作总结（如有）待完成
# 现在知识库的结构（注意大小控制）gen_kb_tree_prompt (当为空的时候应该让LLM生成一个合适的初始目录结构)
# 原始路径，现在标题，摘要，目录
workspace =self.get_workspace_by_msg(None)
kb_tree = await workspace.get_knowledege_catalog()
known_obj = {}
title = knowledge_item.get("title")
if title:
known_obj["title"] = title
summary = knowledge_item.get("summary")
if summary:
known_obj["summary"] = summary
tags = knowledge_item.get("tags")
if tags:
known_obj["tags"] = tags
if need_catalogs:
catalogs = knowledge_item.get("catalogs")
if catalogs:
known_obj["catalogs"] = catalogs
org_path = knowledge_item.get("path")
known_obj["orginal_path"] = org_path
know_info_str = f"# Known information\n{json.dumps(known_obj)}\n"
return AgentPrompt(know_info_str)
async def _llm_read_article(self,knowledge_item:dict) -> ComputeTaskResult:
#目标：
# 得到更好的标题，摘要，目录 （如有必要）,tags
# 应放的合适的位置 （结合组织的目标）
#已知信息：
# 整理是为什么目标服务的 learn_prompt
# 组织的工作总结（如有）
# 现在知识库的结构（注意大小控制）gen_kb_tree_prompt (当为空的时候应该让LLM生成一个合适的初始目录结构)
# 原始路径，现在标题，摘要，目录
# 整理长文件（通用技巧）
# 告诉输入的是部分内容，让LLM为任务产生中间结果
# 依次输入内容，在最后一个内容块输入时，LLM得到结果
#full_content = item.get_article_full_content()
workspace = self.get_workspace_by_msg(None)
full_content = await workspace.load_knowledge_content(knowledge_item["hash"])
if full_content is None:
return
full_content_len = self.token_len(full_content)
if full_content_len < self.get_llm_learn_token_limit():
# 短文章不用总结catelog
#path_list,summary = llm_get_summary(summary,full_content)
prompt = self.get_agent_role_prompt()
learn_prompt = self.get_learn_prompt()
cotent_prompt = AgentPrompt(full_content)
prompt.append(learn_prompt)
prompt.append(cotent_prompt)
env_functions = self._get_inner_functions()
#prompt = self.get_agent_role_prompt()
prompt = self.get_learn_prompt()
known_info_prompt = await self.gen_known_info_for_knowledge_prompt(knowledge_item)
prompt.append(known_info_prompt)
content_prompt = AgentPrompt(full_content)
prompt.append(content_prompt)
env_functions = workspace.get_knowledge_base_ai_functions()
task_result:ComputeTaskResult = await self._do_llm_complection(prompt,env_functions)
if task_result.result_code != ComputeTaskResultCode.OK:
return task_result
llm_result = LLMResult.from_str(task_result.result_str)
path_list,summary = self.parser_learn_llm_result(llm_result)
result_obj = {}
result_obj["error_str"] = task_result.error_str
return result_obj
result_obj = json.loads(task_result.result_str)
return result_obj
else:
# 用传统方法对文章进行一些处理，目的是尽可能减少LLM调用的次数
catelog = item.get_articl_catelog()
chunk_content = full_content.read(self.get_llm_learn_token_limit())
summary = kb.try_get_summary(catelog,full_content)
while chunk_content is not None:
#path_list,summarycatelog = llm_get_summary(summary,chunk_content)
#learn_prompt = self.get_learn_prompt_with_summary()
prompt = AgentPrompt("summary")
learn_prompt.append(prompt)
prompt = AgentPrompt(chunk_content)
learn_prompt.append(prompt)
#llm_result = self.do_llm_competion(learn_prompt)
#path_list,summary,catelog = parser_learn_llm_result(llm_result)
#chunk_content = full_content.read(self.get_llm_learn_token_limit())
logger.warning(f"llm_read_article: article {knowledge_item['path']} is too long,just read summary!")
result_obj = {}
result_obj["error_str"] = f"llm_read_article: article {knowledge_item['path']} is too long,just read summary!"
return result_obj
kb.insert_item(path_list,item,catelog,summary)
async def do_self_think(self):
session_id_list = AIChatSession.list_session(self.agent_id,self.chat_db)
@@ -1043,7 +1104,7 @@ class AIAgent:
known_info = ""
if chatsession.summary is not None:
if len(chatsession.summary) > 1:
known_info += f"## 最近交流的总结 \n {chatsession.summary}\n"
known_info += f"## Recent conversation summary \n {chatsession.summary}\n"
result_token_len -= len(chatsession.summary)
have_known_info = True
@@ -1062,7 +1123,7 @@ class AIAgent:
logger.warning(f"_get_prompt_from_session reach limit of token,just read {read_history_msg} history message.")
break
known_info += f"## 最近的沟通记录 \n {histroy_str}\n"
known_info += f"## Recent conversation history \n {histroy_str}\n"
if have_known_info:
return known_info,result_token_len
@@ -1103,6 +1164,8 @@ class AIAgent:
return False
def need_self_learn(self) -> bool:
if self.learn_prompt is not None:
return True
return False
def wake_up(self) -> None:
@@ -1145,6 +1208,9 @@ class AIAgent:
logger.error(f"agent {self.agent_id} on timer error:{e},{tb_str}")
continue
def token_len(self,text:str) -> int:
return ComputeKernel.llm_num_tokens_from_text(text,self.get_llm_model_name())