mirror of https://github.com/Mai-with-u/MaiBot.git
Merge branch 'PFC-test' of https://github.com/smartmita/MaiBot into PFC-test
Bakadax
commit
25834f88e7
|
|
@ -4,7 +4,10 @@ from src.plugins.memory_system.Hippocampus import HippocampusManager
|
|||
from src.plugins.knowledge.knowledge_lib import qa_manager
|
||||
from src.common.database import db
|
||||
from src.plugins.chat.utils import get_embedding
|
||||
|
||||
# --- NEW IMPORT ---
|
||||
# 从 heartflow 导入知识检索和数据库查询函数/实例
|
||||
from src.plugins.heartFC_chat.heartflow_prompt_builder import prompt_builder
|
||||
# --- END NEW IMPORT ---
|
||||
# import jieba # 如果需要旧版知识库的回退，可能需要
|
||||
# import re # 如果需要旧版知识库的回退，可能需要
|
||||
from src.common.logger_manager import get_logger
|
||||
|
|
@ -35,7 +38,7 @@ PROMPT_INITIAL_REPLY = """{persona_text}。现在你在参与一场QQ私聊，
|
|||
【上一次行动的详细情况和结果】
|
||||
{last_action_context}
|
||||
【时间和超时提示】
|
||||
{time_since_last_bot_message_info}{timeout_context}
|
||||
{time_since_last_bot_message_info}{timeout_context}
|
||||
【最近的对话记录】(包括你已成功发送的消息 和 新收到的消息)
|
||||
{chat_history_text}
|
||||
【你的的回忆】
|
||||
|
|
@ -58,7 +61,7 @@ block_and_ignore: 更加极端的结束对话方式，直接结束对话并在
|
|||
注意：请严格按照JSON格式输出，不要包含任何其他内容。"""
|
||||
|
||||
# Prompt(2): 上一次成功回复后，决定继续发言时的决策 Prompt
|
||||
PROMPT_FOLLOW_UP = """{persona_text}。现在你在参与一场QQ私聊，刚刚你已经回复了对方，请根据以下【所有信息】审慎且灵活的决策下一步行动，可以继续发送新消息，可以等待，可以倾听，可以调取知识，甚至可以屏蔽对方：
|
||||
PROMPT_FOLLOW_UP = """{persona_text}。现在你在参与一场QQ私聊，刚刚你已经回复了对方，请根据以下【所有信息】审慎且灵活的决策下一步行动，可以继续发送新消息，可以等待，可以倾听，可以调取知识，甚至可以屏蔽对方：
|
||||
|
||||
【当前对话目标】
|
||||
{goals_str}
|
||||
|
|
@ -69,7 +72,7 @@ PROMPT_FOLLOW_UP = """{persona_text}。现在你在参与一场QQ私聊，刚刚
|
|||
【上一次行动的详细情况和结果】
|
||||
{last_action_context}
|
||||
【时间和超时提示】
|
||||
{time_since_last_bot_message_info}{timeout_context}
|
||||
{time_since_last_bot_message_info}{timeout_context}
|
||||
【最近的对话记录】(包括你已成功发送的消息 和 新收到的消息)
|
||||
{chat_history_text}
|
||||
【你的的回忆】
|
||||
|
|
@ -126,6 +129,7 @@ class ActionPlanner:
|
|||
self.private_name = private_name
|
||||
self.chat_observer = ChatObserver.get_instance(stream_id, private_name)
|
||||
|
||||
# _get_memory_info 保持不变
|
||||
async def _get_memory_info(self, text: str) -> str:
|
||||
"""根据文本自动检索相关记忆"""
|
||||
memory_prompt = ""
|
||||
|
|
@ -156,112 +160,9 @@ class ActionPlanner:
|
|||
# memory_prompt = "检索记忆时出错。\n" # 可以选择是否提示错误
|
||||
return memory_prompt
|
||||
|
||||
async def _get_prompt_info_old(self, message: str, threshold: float) -> str:
|
||||
"""
|
||||
旧版的知识检索方法，根据消息文本从旧知识库（knowledges collection）检索。
|
||||
(移植并自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
related_info = ""
|
||||
start_time = time.time()
|
||||
logger.debug(f"[私聊]决策层[{self.private_name}]开始使用旧版知识检索，消息: {message[:30]}...")
|
||||
# --- REMOVED _get_prompt_info_old ---
|
||||
|
||||
# 简化处理：直接使用整个消息进行查询，不再提取主题
|
||||
query_text = message.strip()
|
||||
if not query_text:
|
||||
logger.debug(f"[私聊]决策层[{self.private_name}]旧版知识检索：消息为空，跳过。")
|
||||
return ""
|
||||
|
||||
embedding = None
|
||||
try:
|
||||
embedding = await get_embedding(query_text, request_type="pfc_implicit_knowledge")
|
||||
except Exception as e:
|
||||
logger.error(f"[私聊]决策层[{self.private_name}]旧版知识检索：获取嵌入向量时出错: {str(e)}")
|
||||
|
||||
if not embedding:
|
||||
logger.error(f"[私聊]决策层[{self.private_name}]旧版知识检索：获取嵌入向量失败。")
|
||||
return ""
|
||||
|
||||
# 调用我们之前添加的 get_info_from_db 函数
|
||||
results = get_info_from_db(embedding, limit=5, threshold=threshold, return_raw=True) # 最多查 5 条
|
||||
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}]旧版知识库查询完成，耗时: {time.time() - start_time:.3f}秒，获取{len(results)}条结果"
|
||||
)
|
||||
|
||||
# 去重和格式化
|
||||
unique_contents = set()
|
||||
final_results_content = []
|
||||
for result in results:
|
||||
content = result.get("content", "").strip()
|
||||
# similarity = result.get("similarity", 0.0)
|
||||
if content and content not in unique_contents:
|
||||
unique_contents.add(content)
|
||||
# 可以选择性地加入相似度信息，或者只加内容
|
||||
# final_results_content.append(f"[{similarity:.2f}] {content}")
|
||||
final_results_content.append(content)
|
||||
|
||||
if final_results_content:
|
||||
related_info = "\n".join(final_results_content)
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版知识检索格式化后内容: {related_info[:100]}...")
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版知识检索未找到合适结果或结果为空。")
|
||||
|
||||
logger.info(f"[私聊][{self.private_name}]旧版知识检索总耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
|
||||
async def _get_prompt_info(self, message: str, threshold: float = 0.38) -> str:
|
||||
"""
|
||||
自动检索相关知识的主函数。优先使用 LPMM，失败则回退到旧版。
|
||||
(移植自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
related_info = ""
|
||||
start_time = time.time()
|
||||
message = message.strip()
|
||||
if not message:
|
||||
logger.debug(f"[私聊][{self.private_name}]自动知识检索：输入消息为空。")
|
||||
return ""
|
||||
|
||||
logger.debug(f"[私聊][{self.private_name}]开始自动知识检索，消息: {message[:30]}...")
|
||||
|
||||
# 1. 尝试从 LPMM 知识库获取知识
|
||||
try:
|
||||
found_knowledge_from_lpmm = qa_manager.get_knowledge(message)
|
||||
if found_knowledge_from_lpmm and found_knowledge_from_lpmm.strip():
|
||||
related_info = found_knowledge_from_lpmm.strip()
|
||||
logger.info(f"[私聊][{self.private_name}]从 LPMM 知识库获取到知识，长度: {len(related_info)}")
|
||||
logger.debug(f"[私聊][{self.private_name}]LPMM 知识内容: {related_info[:100]}...")
|
||||
# LPMM 成功获取，直接返回
|
||||
logger.info(f"[私聊][{self.private_name}]自动知识检索(LPMM)耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]LPMM 知识库未返回有效知识，尝试旧版数据库检索。")
|
||||
except Exception as e:
|
||||
logger.error(
|
||||
f"[私聊][{self.private_name}]调用 LPMM 知识库 (qa_manager.get_knowledge) 时发生异常: {str(e)}，尝试旧版数据库检索。"
|
||||
)
|
||||
|
||||
# 2. 如果 LPMM 失败或无结果，尝试旧版数据库
|
||||
try:
|
||||
knowledge_from_old = await self._get_prompt_info_old(message, threshold=threshold)
|
||||
if knowledge_from_old and knowledge_from_old.strip():
|
||||
related_info = knowledge_from_old.strip()
|
||||
logger.info(f"[私聊][{self.private_name}]从旧版数据库检索到知识，长度: {len(related_info)}")
|
||||
# 旧版成功获取，返回
|
||||
logger.info(f"[私聊][{self.private_name}]自动知识检索(旧版)耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版数据库也未检索到有效知识。")
|
||||
|
||||
except Exception as e2:
|
||||
logger.error(
|
||||
f"[私聊][{self.private_name}]调用旧版知识库检索 (_get_prompt_info_old) 时也发生异常: {str(e2)}"
|
||||
)
|
||||
|
||||
# 如果两种方法都失败或无结果
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}]自动知识检索总耗时: {time.time() - start_time:.3f}秒，未找到任何相关知识。"
|
||||
)
|
||||
return "" # 返回空字符串
|
||||
# --- REMOVED _get_prompt_info ---
|
||||
|
||||
# 修改 plan 方法签名，增加 last_successful_reply_action 参数
|
||||
async def plan(
|
||||
|
|
@ -370,40 +271,6 @@ class ActionPlanner:
|
|||
logger.error(f"[私聊][{self.private_name}]构建对话目标字符串时出错: {e}")
|
||||
goals_str = "- 构建对话目标时出错。\n"
|
||||
|
||||
# --- 知识信息字符串构建开始 ---
|
||||
# knowledge_info_str = "【已获取的相关知识和记忆】\n"
|
||||
# try:
|
||||
# 检查 conversation_info 是否有 knowledge_list 并且不为空
|
||||
# if hasattr(conversation_info, "knowledge_list") and conversation_info.knowledge_list:
|
||||
# 最多只显示最近的 5 条知识，防止 Prompt 过长
|
||||
# recent_knowledge = conversation_info.knowledge_list[-5:]
|
||||
# for i, knowledge_item in enumerate(recent_knowledge):
|
||||
# if isinstance(knowledge_item, dict):
|
||||
# query = knowledge_item.get("query", "未知查询")
|
||||
# knowledge = knowledge_item.get("knowledge", "无知识内容")
|
||||
# source = knowledge_item.get("source", "未知来源")
|
||||
# 只取知识内容的前 2000 个字，避免太长
|
||||
# knowledge_snippet = knowledge[:2000] + "..." if len(knowledge) > 2000 else knowledge
|
||||
# knowledge_info_str += (
|
||||
# f"{i + 1}. 关于 '{query}' 的知识 (来源: {source}):\n {knowledge_snippet}\n"
|
||||
# )
|
||||
# else:
|
||||
# 处理列表里不是字典的异常情况
|
||||
# knowledge_info_str += f"{i + 1}. 发现一条格式不正确的知识记录。\n"
|
||||
|
||||
# if not recent_knowledge: # 如果 knowledge_list 存在但为空
|
||||
# knowledge_info_str += "- 暂无相关知识和记忆。\n"
|
||||
|
||||
# else:
|
||||
# 如果 conversation_info 没有 knowledge_list 属性，或者列表为空
|
||||
# knowledge_info_str += "- 暂无相关知识记忆。\n"
|
||||
# except AttributeError:
|
||||
# logger.warning(f"[私聊][{self.private_name}]ConversationInfo 对象可能缺少 knowledge_list 属性。")
|
||||
# knowledge_info_str += "- 获取知识列表时出错。\n"
|
||||
# except Exception as e:
|
||||
# logger.error(f"[私聊][{self.private_name}]构建知识信息字符串时出错: {e}")
|
||||
# knowledge_info_str += "- 处理知识列表时出错。\n"
|
||||
# --- 知识信息字符串构建结束 ---
|
||||
|
||||
# 获取聊天历史记录 (chat_history_text)
|
||||
try:
|
||||
|
|
@ -512,22 +379,26 @@ class ActionPlanner:
|
|||
last_action_context += f"- 该行动当前状态: {status}\n"
|
||||
# self.last_successful_action_type = None # 非完成状态，清除记录
|
||||
|
||||
retrieved_memory_str_planner = ""
|
||||
retrieved_memory_str_planner = ""
|
||||
retrieved_knowledge_str_planner = ""
|
||||
retrieval_context = chat_history_text # 使用聊天记录作为检索上下文
|
||||
if retrieval_context and retrieval_context != "还没有聊天记录。" and retrieval_context != "[构建聊天记录出错]":
|
||||
try:
|
||||
# 调用本地的 _get_memory_info
|
||||
logger.debug(f"[私聊][{self.private_name}] (ActionPlanner) 开始自动检索记忆...")
|
||||
retrieved_memory_str_planner = await self._get_memory_info(text=retrieval_context)
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}] (ActionPlanner) 自动检索记忆 {'完成' if retrieved_memory_str_planner else '无结果'}。"
|
||||
)
|
||||
|
||||
logger.debug(f"[私聊][{self.private_name}] (ActionPlanner) 开始自动知识检索...")
|
||||
retrieved_knowledge_str_planner = await self._get_prompt_info(message=retrieval_context)
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}] (ActionPlanner) 自动检索知识 {'完成' if retrieved_knowledge_str_planner else '无结果'}。"
|
||||
)
|
||||
# --- MODIFIED KNOWLEDGE RETRIEVAL ---
|
||||
# 调用导入的 prompt_builder.get_prompt_info
|
||||
logger.debug(f"[私聊][{self.private_name}] (ActionPlanner) 开始自动检索知识 (使用导入函数)...")
|
||||
# 使用导入的 prompt_builder 实例及其方法
|
||||
retrieved_knowledge_str_planner = await prompt_builder.get_prompt_info(message=retrieval_context, threshold=0.38)
|
||||
# --- END MODIFIED KNOWLEDGE RETRIEVAL ---
|
||||
logger.info(f"[私聊][{self.private_name}] (ActionPlanner) 自动检索知识 {'完成' if retrieved_knowledge_str_planner else '无结果'}。")
|
||||
|
||||
except Exception as retrieval_err:
|
||||
logger.error(f"[私聊][{self.private_name}] (ActionPlanner) 自动检索时出错: {retrieval_err}")
|
||||
retrieved_memory_str_planner = "检索记忆时出错。\n"
|
||||
|
|
@ -554,7 +425,7 @@ class ActionPlanner:
|
|||
time_since_last_bot_message_info=time_since_last_bot_message_info,
|
||||
timeout_context=timeout_context,
|
||||
chat_history_text=chat_history_text if chat_history_text.strip() else "还没有聊天记录。",
|
||||
# knowledge_info_str=knowledge_info_str,
|
||||
# knowledge_info_str=knowledge_info_str, # 移除了旧知识展示方式
|
||||
retrieved_memory_str=retrieved_memory_str_planner if retrieved_memory_str_planner else "无相关记忆。",
|
||||
retrieved_knowledge_str=retrieved_knowledge_str_planner
|
||||
if retrieved_knowledge_str_planner
|
||||
|
|
@ -660,90 +531,4 @@ class ActionPlanner:
|
|||
except Exception as e:
|
||||
# 外层异常处理保持不变
|
||||
logger.error(f"[私聊][{self.private_name}]规划行动时调用 LLM 或处理结果出错: {str(e)}")
|
||||
return "wait", f"行动规划处理中发生错误，暂时等待: {str(e)}"
|
||||
|
||||
|
||||
def get_info_from_db(
|
||||
query_embedding: list, limit: int = 1, threshold: float = 0.5, return_raw: bool = False
|
||||
) -> Union[str, list]:
|
||||
"""
|
||||
从旧知识库 (knowledges collection) 中根据嵌入向量相似度检索信息。
|
||||
(移植自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
if not query_embedding:
|
||||
return "" if not return_raw else []
|
||||
# 使用余弦相似度计算
|
||||
pipeline = [
|
||||
{
|
||||
"$addFields": {
|
||||
"dotProduct": {
|
||||
"$reduce": {
|
||||
"input": {"$range": [0, {"$size": "$embedding"}]},
|
||||
"initialValue": 0,
|
||||
"in": {
|
||||
"$add": [
|
||||
"$$value",
|
||||
{
|
||||
"$multiply": [
|
||||
{"$arrayElemAt": ["$embedding", "$$this"]},
|
||||
{"$arrayElemAt": [query_embedding, "$$this"]},
|
||||
]
|
||||
},
|
||||
]
|
||||
},
|
||||
}
|
||||
},
|
||||
"magnitude1": {
|
||||
"$sqrt": {
|
||||
"$reduce": {
|
||||
"input": "$embedding",
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
},
|
||||
"magnitude2": {
|
||||
"$sqrt": {
|
||||
"$reduce": {
|
||||
"input": query_embedding,
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
},
|
||||
}
|
||||
},
|
||||
# 防止除以零错误，添加一个小的 epsilon
|
||||
{
|
||||
"$addFields": {
|
||||
"similarity": {
|
||||
"$divide": ["$dotProduct", {"$max": [{"$multiply": ["$magnitude1", "$magnitude2"]}, 1e-9]}]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$match": {
|
||||
"similarity": {"$gte": threshold} # 只保留相似度大于等于阈值的结果
|
||||
}
|
||||
},
|
||||
{"$sort": {"similarity": -1}},
|
||||
{"$limit": limit},
|
||||
{"$project": {"content": 1, "similarity": 1}},
|
||||
]
|
||||
|
||||
try:
|
||||
results = list(db.knowledges.aggregate(pipeline))
|
||||
# 注意：这里的 logger 需要能访问到，或者在这个函数里获取 logger 实例
|
||||
# logger.debug(f"旧知识库查询结果数量: {len(results)}") # 暂时注释掉，避免 logger 未定义
|
||||
except Exception as e:
|
||||
logger.debug(f"执行旧知识库聚合查询时出错: {e}")
|
||||
results = []
|
||||
|
||||
if not results:
|
||||
return "" if not return_raw else []
|
||||
|
||||
if return_raw:
|
||||
return results
|
||||
else:
|
||||
# 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 "wait", f"行动规划处理中发生错误，暂时等待: {str(e)}"
|
||||
|
|
@ -9,7 +9,10 @@ from src.common.database import db
|
|||
|
||||
# 用于获取文本的嵌入向量 (旧知识库需要)
|
||||
from src.plugins.chat.utils import get_embedding
|
||||
|
||||
# --- NEW IMPORT ---
|
||||
# 从 heartflow 导入知识检索和数据库查询函数/实例
|
||||
from src.plugins.heartFC_chat.heartflow_prompt_builder import prompt_builder
|
||||
# --- END NEW IMPORT ---
|
||||
# 可能用于旧知识库提取主题 (如果需要回退到旧方法)
|
||||
# import jieba # 如果报错说找不到 jieba，可能需要安装: pip install jieba
|
||||
# import re # 正则表达式库，通常 Python 自带
|
||||
|
|
@ -59,7 +62,7 @@ PROMPT_DIRECT_REPLY = """{persona_text}。现在你在参与一场QQ私聊，请
|
|||
请直接输出回复内容，不需要任何额外格式。"""
|
||||
|
||||
# Prompt for send_new_message (追问/补充)
|
||||
PROMPT_SEND_NEW_MESSAGE = """{persona_text}。现在你在参与一场QQ私聊，**刚刚你已经发送了一条或多条消息**，现在请根据以下信息再发一条新消息：
|
||||
PROMPT_SEND_NEW_MESSAGE = """{persona_text}。现在你在参与一场QQ私聊，**刚刚你已经发送了一条或多条消息**，现在请根据以下信息再发一条新消息：
|
||||
|
||||
当前对话目标：{goals_str}
|
||||
|
||||
|
|
@ -72,7 +75,7 @@ PROMPT_SEND_NEW_MESSAGE = """{persona_text}。现在你在参与一场QQ私聊
|
|||
|
||||
{retrieved_memory_str}
|
||||
|
||||
请根据上述信息，结合聊天记录，继续发一条新消息（例如对之前消息的补充，深入话题，或追问等等）。该消息应该：
|
||||
请根据上述信息，结合聊天记录，继续发一条新消息（例如对之前消息的补充，深入话题，或追问等等）。该消息应该：
|
||||
1. 符合对话目标，以"你"的角度发言（不要自己与自己对话！）
|
||||
2. 符合你的性格特征和身份细节
|
||||
3. 通俗易懂，自然流畅，像正常聊天一样，简短（通常20字以内，除非特殊情况）
|
||||
|
|
@ -121,6 +124,7 @@ class ReplyGenerator:
|
|||
self.chat_observer = ChatObserver.get_instance(stream_id, private_name)
|
||||
self.reply_checker = ReplyChecker(stream_id, private_name)
|
||||
|
||||
# _get_memory_info 保持不变，因为它不是与 heartflow 重复的部分
|
||||
async def _get_memory_info(self, text: str) -> str:
|
||||
"""根据文本自动检索相关记忆"""
|
||||
memory_prompt = ""
|
||||
|
|
@ -149,111 +153,9 @@ class ReplyGenerator:
|
|||
# memory_prompt = "检索记忆时出错。\n" # 可以选择是否提示错误
|
||||
return memory_prompt
|
||||
|
||||
async def _get_prompt_info_old(self, message: str, threshold: float) -> str:
|
||||
"""
|
||||
旧版的知识检索方法，根据消息文本从旧知识库（knowledges collection）检索。
|
||||
(移植并简化自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
related_info = ""
|
||||
start_time = time.time()
|
||||
logger.debug(f"[私聊][{self.private_name}]开始使用旧版知识检索，消息: {message[:30]}...")
|
||||
# --- REMOVED _get_prompt_info_old ---
|
||||
|
||||
# 简化处理：直接使用整个消息进行查询，不再提取主题
|
||||
query_text = message.strip()
|
||||
if not query_text:
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版知识检索：消息为空，跳过。")
|
||||
return ""
|
||||
|
||||
embedding = None
|
||||
try:
|
||||
embedding = await get_embedding(query_text, request_type="pfc_implicit_knowledge")
|
||||
except Exception as e:
|
||||
logger.error(f"[私聊][{self.private_name}]旧版知识检索：获取嵌入向量时出错: {str(e)}")
|
||||
|
||||
if not embedding:
|
||||
logger.error(f"[私聊][{self.private_name}]旧版知识检索：获取嵌入向量失败。")
|
||||
return ""
|
||||
|
||||
# 调用我们之前添加的 get_info_from_db 函数
|
||||
results = get_info_from_db(embedding, limit=5, threshold=threshold, return_raw=True) # 最多查 5 条
|
||||
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}]旧版知识库查询完成，耗时: {time.time() - start_time:.3f}秒，获取{len(results)}条结果"
|
||||
)
|
||||
|
||||
# 去重和格式化
|
||||
unique_contents = set()
|
||||
final_results_content = []
|
||||
for result in results:
|
||||
content = result.get("content", "").strip()
|
||||
# similarity = result.get("similarity", 0.0)
|
||||
if content and content not in unique_contents:
|
||||
unique_contents.add(content)
|
||||
# 可以选择性地加入相似度信息，或者只加内容
|
||||
# final_results_content.append(f"[{similarity:.2f}] {content}")
|
||||
final_results_content.append(content)
|
||||
|
||||
if final_results_content:
|
||||
related_info = "\n".join(final_results_content)
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版知识检索格式化后内容: {related_info[:100]}...")
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版知识检索未找到合适结果或结果为空。")
|
||||
|
||||
logger.info(f"[私聊][{self.private_name}]旧版知识检索总耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
|
||||
async def _get_prompt_info(self, message: str, threshold: float = 0.38) -> str:
|
||||
"""
|
||||
自动检索相关知识的主函数。优先使用 LPMM，失败则回退到旧版。
|
||||
(移植自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
related_info = ""
|
||||
start_time = time.time()
|
||||
message = message.strip()
|
||||
if not message:
|
||||
logger.debug(f"[私聊][{self.private_name}]自动知识检索：输入消息为空。")
|
||||
return ""
|
||||
|
||||
logger.debug(f"[私聊][{self.private_name}]开始自动知识检索，消息: {message[:30]}...")
|
||||
|
||||
# 1. 尝试从 LPMM 知识库获取知识
|
||||
try:
|
||||
found_knowledge_from_lpmm = qa_manager.get_knowledge(message)
|
||||
if found_knowledge_from_lpmm and found_knowledge_from_lpmm.strip():
|
||||
related_info = found_knowledge_from_lpmm.strip()
|
||||
logger.info(f"[私聊][{self.private_name}]从 LPMM 知识库获取到知识，长度: {len(related_info)}")
|
||||
logger.debug(f"[私聊][{self.private_name}]LPMM 知识内容: {related_info[:100]}...")
|
||||
# LPMM 成功获取，直接返回
|
||||
logger.info(f"[私聊][{self.private_name}]自动知识检索(LPMM)耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]LPMM 知识库未返回有效知识，尝试旧版数据库检索。")
|
||||
except Exception as e:
|
||||
logger.error(
|
||||
f"[私聊][{self.private_name}]调用 LPMM 知识库 (qa_manager.get_knowledge) 时发生异常: {str(e)}，尝试旧版数据库检索。"
|
||||
)
|
||||
|
||||
# 2. 如果 LPMM 失败或无结果，尝试旧版数据库
|
||||
try:
|
||||
knowledge_from_old = await self._get_prompt_info_old(message, threshold=threshold)
|
||||
if knowledge_from_old and knowledge_from_old.strip():
|
||||
related_info = knowledge_from_old.strip()
|
||||
logger.info(f"[私聊][{self.private_name}]从旧版数据库检索到知识，长度: {len(related_info)}")
|
||||
# 旧版成功获取，返回
|
||||
logger.info(f"[私聊][{self.private_name}]自动知识检索(旧版)耗时: {time.time() - start_time:.3f}秒")
|
||||
return related_info
|
||||
else:
|
||||
logger.debug(f"[私聊][{self.private_name}]旧版数据库也未检索到有效知识。")
|
||||
|
||||
except Exception as e2:
|
||||
logger.error(
|
||||
f"[私聊][{self.private_name}]调用旧版知识库检索 (_get_prompt_info_old) 时也发生异常: {str(e2)}"
|
||||
)
|
||||
# 如果两种方法都失败或无结果
|
||||
logger.info(
|
||||
f"[私聊][{self.private_name}]自动知识检索总耗时: {time.time() - start_time:.3f}秒，未找到任何相关知识。"
|
||||
)
|
||||
return "" # 返回空字符串
|
||||
# --- REMOVED _get_prompt_info ---
|
||||
|
||||
# 修改 generate 方法签名，增加 action_type 参数
|
||||
async def generate(
|
||||
|
|
@ -294,37 +196,6 @@ class ReplyGenerator:
|
|||
else:
|
||||
goals_str = "- 目前没有明确对话目标\n" # 简化无目标情况
|
||||
|
||||
# --- 新增：构建知识信息字符串 ---
|
||||
# knowledge_info_str = "【供参考的相关知识和记忆】\n" # 稍微改下标题，表明是供参考
|
||||
# try:
|
||||
# 检查 conversation_info 是否有 knowledge_list 并且不为空
|
||||
# if hasattr(conversation_info, "knowledge_list") and conversation_info.knowledge_list:
|
||||
# 最多只显示最近的 5 条知识
|
||||
# recent_knowledge = conversation_info.knowledge_list[-5:]
|
||||
# for i, knowledge_item in enumerate(recent_knowledge):
|
||||
# if isinstance(knowledge_item, dict):
|
||||
# query = knowledge_item.get("query", "未知查询")
|
||||
# knowledge = knowledge_item.get("knowledge", "无知识内容")
|
||||
# source = knowledge_item.get("source", "未知来源")
|
||||
# 只取知识内容的前 2000 个字
|
||||
# knowledge_snippet = knowledge[:2000] + "..." if len(knowledge) > 2000 else knowledge
|
||||
# knowledge_info_str += (
|
||||
# f"{i + 1}. 关于 '{query}' (来源: {source}): {knowledge_snippet}\n" # 格式微调，更简洁
|
||||
# )
|
||||
# else:
|
||||
# knowledge_info_str += f"{i + 1}. 发现一条格式不正确的知识记录。\n"
|
||||
|
||||
# if not recent_knowledge:
|
||||
# knowledge_info_str += "- 暂无。\n" # 更简洁的提示
|
||||
|
||||
# else:
|
||||
# knowledge_info_str += "- 暂无。\n"
|
||||
# except AttributeError:
|
||||
# logger.warning(f"[私聊][{self.private_name}]ConversationInfo 对象可能缺少 knowledge_list 属性。")
|
||||
# knowledge_info_str += "- 获取知识列表时出错。\n"
|
||||
# except Exception as e:
|
||||
# logger.error(f"[私聊][{self.private_name}]构建知识信息字符串时出错: {e}")
|
||||
# knowledge_info_str += "- 处理知识列表时出错。\n"
|
||||
|
||||
# 获取聊天历史记录 (chat_history_text)
|
||||
chat_history_text = observation_info.chat_history_str
|
||||
|
|
@ -349,7 +220,7 @@ class ReplyGenerator:
|
|||
retrieval_context = chat_history_text
|
||||
if retrieval_context and retrieval_context != "还没有聊天记录。" and retrieval_context != "[构建聊天记录出错]":
|
||||
try:
|
||||
# 提取记忆
|
||||
# 提取记忆 (调用本地的 _get_memory_info)
|
||||
logger.debug(f"[私聊][{self.private_name}]开始自动检索记忆...")
|
||||
retrieved_memory_str = await self._get_memory_info(text=retrieval_context)
|
||||
if retrieved_memory_str:
|
||||
|
|
@ -357,9 +228,13 @@ class ReplyGenerator:
|
|||
else:
|
||||
logger.info(f"[私聊][{self.private_name}]未自动检索到相关记忆。")
|
||||
|
||||
# 提取知识
|
||||
logger.debug(f"[私聊][{self.private_name}]开始自动检索知识...")
|
||||
retrieved_knowledge_str = await self._get_prompt_info(message=retrieval_context)
|
||||
# --- MODIFIED KNOWLEDGE RETRIEVAL ---
|
||||
# 提取知识 (调用导入的 prompt_builder.get_prompt_info)
|
||||
logger.debug(f"[私聊][{self.private_name}]开始自动检索知识 (使用导入函数)...")
|
||||
# 使用导入的 prompt_builder 实例及其方法
|
||||
retrieved_knowledge_str = await prompt_builder.get_prompt_info(message=retrieval_context, threshold=0.38)
|
||||
# --- END MODIFIED KNOWLEDGE RETRIEVAL ---
|
||||
|
||||
if retrieved_knowledge_str:
|
||||
logger.info(f"[私聊][{self.private_name}]自动检索到相关知识。")
|
||||
else:
|
||||
|
|
@ -390,11 +265,9 @@ class ReplyGenerator:
|
|||
persona_text=persona_text,
|
||||
goals_str=goals_str,
|
||||
chat_history_text=chat_history_text,
|
||||
# knowledge_info_str=knowledge_info_str,
|
||||
retrieved_memory_str=retrieved_memory_str if retrieved_memory_str else "无相关记忆。", # 如果为空则提示无
|
||||
retrieved_knowledge_str=retrieved_knowledge_str
|
||||
if retrieved_knowledge_str
|
||||
else "无相关知识。", # 如果为空则提示无
|
||||
# knowledge_info_str=knowledge_info_str, # 移除了这个旧的知识展示方式
|
||||
retrieved_memory_str=retrieved_memory_str if retrieved_memory_str else "无相关记忆。", # 如果为空则提示无
|
||||
retrieved_knowledge_str=retrieved_knowledge_str if retrieved_knowledge_str else "无相关知识。" # 如果为空则提示无
|
||||
)
|
||||
|
||||
# --- 调用 LLM 生成 ---
|
||||
|
|
@ -417,89 +290,3 @@ class ReplyGenerator:
|
|||
(此方法逻辑保持不变)
|
||||
"""
|
||||
return await self.reply_checker.check(reply, goal, chat_history, chat_history_str, retry_count)
|
||||
|
||||
|
||||
def get_info_from_db(
|
||||
query_embedding: list, limit: int = 1, threshold: float = 0.5, return_raw: bool = False
|
||||
) -> Union[str, list]:
|
||||
"""
|
||||
从旧知识库 (knowledges collection) 中根据嵌入向量相似度检索信息。
|
||||
(移植自 heartflow_prompt_builder.py)
|
||||
"""
|
||||
if not query_embedding:
|
||||
return "" if not return_raw else []
|
||||
# 使用余弦相似度计算
|
||||
pipeline = [
|
||||
{
|
||||
"$addFields": {
|
||||
"dotProduct": {
|
||||
"$reduce": {
|
||||
"input": {"$range": [0, {"$size": "$embedding"}]},
|
||||
"initialValue": 0,
|
||||
"in": {
|
||||
"$add": [
|
||||
"$$value",
|
||||
{
|
||||
"$multiply": [
|
||||
{"$arrayElemAt": ["$embedding", "$$this"]},
|
||||
{"$arrayElemAt": [query_embedding, "$$this"]},
|
||||
]
|
||||
},
|
||||
]
|
||||
},
|
||||
}
|
||||
},
|
||||
"magnitude1": {
|
||||
"$sqrt": {
|
||||
"$reduce": {
|
||||
"input": "$embedding",
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
},
|
||||
"magnitude2": {
|
||||
"$sqrt": {
|
||||
"$reduce": {
|
||||
"input": query_embedding,
|
||||
"initialValue": 0,
|
||||
"in": {"$add": ["$$value", {"$multiply": ["$$this", "$$this"]}]},
|
||||
}
|
||||
}
|
||||
},
|
||||
}
|
||||
},
|
||||
# 防止除以零错误，添加一个小的 epsilon
|
||||
{
|
||||
"$addFields": {
|
||||
"similarity": {
|
||||
"$divide": ["$dotProduct", {"$max": [{"$multiply": ["$magnitude1", "$magnitude2"]}, 1e-9]}]
|
||||
}
|
||||
}
|
||||
},
|
||||
{
|
||||
"$match": {
|
||||
"similarity": {"$gte": threshold} # 只保留相似度大于等于阈值的结果
|
||||
}
|
||||
},
|
||||
{"$sort": {"similarity": -1}},
|
||||
{"$limit": limit},
|
||||
{"$project": {"content": 1, "similarity": 1}},
|
||||
]
|
||||
|
||||
try:
|
||||
results = list(db.knowledges.aggregate(pipeline))
|
||||
# 注意：这里的 logger 需要能访问到，或者在这个函数里获取 logger 实例
|
||||
# logger.debug(f"旧知识库查询结果数量: {len(results)}") # 暂时注释掉，避免 logger 未定义
|
||||
except Exception as e:
|
||||
logger.debug(f"执行旧知识库聚合查询时出错: {e}")
|
||||
results = []
|
||||
|
||||
if not results:
|
||||
return "" if not return_raw else []
|
||||
|
||||
if return_raw:
|
||||
return results
|
||||
else:
|
||||
# 返回所有找到的内容，用换行分隔
|
||||
return "\n".join(str(result["content"]) for result in results)
|
||||
|
|
|
|||
Loading…
Reference in New Issue