gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix：优化记忆提取，修复破损的tool信息

Browse Source
This commit is contained in:
SengokuCola
2025-05-27 18:21:05 +08:00
parent 548a583cc7
commit 52f7cc3762
9 changed files with 110 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/chat/focus_chat/expressors/exprssion_learner.py
View File

@@ -61,10 +61,10 @@ class ExpressionLearner:
def __init__(self) -> None:
# TODO: API-Adapter修改标记
self.express_learn_model: LLMRequest = LLMRequest(
model=global_config.model.normal,
model=global_config.model.focus_expressor,
temperature=0.1,
max_tokens=256,
request_type="response_heartflow",
request_type="learn_expression",
)
async def get_expression_by_chat_id(self, chat_id: str) -> Tuple[List[Dict[str, str]], List[Dict[str, str]]]:

4
src/chat/focus_chat/heartFC_chat.py
View File

@@ -19,6 +19,7 @@ from src.chat.focus_chat.info_processors.working_memory_processor import Working
from src.chat.focus_chat.info_processors.action_processor import ActionProcessor
from src.chat.heart_flow.observation.hfcloop_observation import HFCloopObservation
from src.chat.heart_flow.observation.working_observation import WorkingMemoryObservation
from src.chat.heart_flow.observation.structure_observation import StructureObservation
from src.chat.focus_chat.info_processors.tool_processor import ToolProcessor
from src.chat.focus_chat.expressors.default_expressor import DefaultExpressor
from src.chat.focus_chat.memory_activator import MemoryActivator
@@ -97,6 +98,7 @@ class HeartFChatting:
self.log_prefix: str = str(chat_id) # Initial default, will be updated
self.hfcloop_observation = HFCloopObservation(observe_id=self.stream_id)
self.chatting_observation = observations[0]
self.structure_observation = StructureObservation(observe_id=self.stream_id)
self.memory_activator = MemoryActivator()
self.working_memory = WorkingMemory(chat_id=self.stream_id)
@@ -415,10 +417,12 @@ class HeartFChatting:
await self.chatting_observation.observe()
await self.working_observation.observe()
await self.hfcloop_observation.observe()
await self.structure_observation.observe()
observations: List[Observation] = []
observations.append(self.chatting_observation)
observations.append(self.working_observation)
observations.append(self.hfcloop_observation)
observations.append(self.structure_observation)
loop_observation_info = {
"observations": observations,

4
src/chat/focus_chat/info/structured_info.py
View File

@@ -76,7 +76,11 @@ class StructuredInfo:
"""
info_str = ""
# print(f"self.data: {self.data}")
for key, value in self.data.items():
# print(f"key: {key}, value: {value}")
info_str += f"信息类型：{key}，信息内容：{value}\n"
return info_str

13
src/chat/focus_chat/info_processors/tool_processor.py
View File

@@ -75,10 +75,12 @@ class ToolProcessor(BaseProcessor):
result, used_tools, prompt = await self.execute_tools(observation, running_memorys)
# 更新WorkingObservation中的结构化信息
logger.debug(f"工具调用结果: {result}")
for observation in observations:
if isinstance(observation, StructureObservation):
for structured_info in result:
logger.debug(f"{self.log_prefix} 更新WorkingObservation中的结构化信息: {structured_info}")
# logger.debug(f"{self.log_prefix} 更新WorkingObservation中的结构化信息: {structured_info}")
observation.add_structured_info(structured_info)
working_infos = observation.get_observe_info()
@@ -87,7 +89,12 @@ class ToolProcessor(BaseProcessor):
structured_info = StructuredInfo()
if working_infos:
for working_info in working_infos:
structured_info.set_info(working_info.get("type"), working_info.get("content"))
# print(f"working_info: {working_info}")
# print(f"working_info.get('type'): {working_info.get('type')}")
# print(f"working_info.get('content'): {working_info.get('content')}")
structured_info.set_info(key=working_info.get('type'), value=working_info.get('content'))
# info = structured_info.get_processed_info()
# print(f"info: {info}")
return [structured_info]
@@ -155,7 +162,7 @@ class ToolProcessor(BaseProcessor):
)
# 调用LLM，专注于工具使用
# logger.debug(f"开始执行工具调用{prompt}")
logger.debug(f"开始执行工具调用{prompt}")
response, _, tool_calls = await self.llm_model.generate_response_tool_async(prompt=prompt, tools=tools)
logger.debug(f"获取到工具原始输出:\n{tool_calls}")

97
src/chat/focus_chat/memory_activator.py
View File

@@ -4,24 +4,58 @@ from src.chat.heart_flow.observation.hfcloop_observation import HFCloopObservati
from src.llm_models.utils_model import LLMRequest
from src.config.config import global_config
from src.common.logger_manager import get_logger
from src.chat.utils.prompt_builder import Prompt
from src.chat.utils.prompt_builder import Prompt, global_prompt_manager
from datetime import datetime
from src.chat.memory_system.Hippocampus import HippocampusManager
from typing import List, Dict
import difflib
import json
from json_repair import repair_json
logger = get_logger("memory_activator")
def get_keywords_from_json(json_str):
"""
从JSON字符串中提取关键词列表
Args:
json_str: JSON格式的字符串
Returns:
List[str]: 关键词列表
"""
try:
# 使用repair_json修复JSON格式
fixed_json = repair_json(json_str)
# 如果repair_json返回的是字符串，需要解析为Python对象
if isinstance(fixed_json, str):
result = json.loads(fixed_json)
else:
# 如果repair_json直接返回了字典对象，直接使用
result = fixed_json
# 提取关键词
keywords = result.get("keywords", [])
return keywords
except Exception as e:
logger.error(f"解析关键词JSON失败: {e}")
return []
def init_prompt():
# --- Group Chat Prompt ---
memory_activator_prompt = """
你是一个记忆分析器，你需要根据以下信息来进行会议
你是一个记忆分析器，你需要根据以下信息来进行回忆
以下是一场聊天中的信息，请根据这些信息，总结出几个关键词作为记忆回忆的触发词
{obs_info_text}
历史关键词（请避免重复提取这些关键词）：
{cached_keywords}
请输出一个json格式，包含以下字段：
{{
"keywords": ["关键词1", "关键词2", "关键词3",......]
@@ -39,6 +73,7 @@ class MemoryActivator:
model=global_config.model.memory_summary, temperature=0.7, max_tokens=50, request_type="chat_observation"
)
self.running_memory = []
self.cached_keywords = set() # 用于缓存历史关键词
async def activate_memory(self, observations) -> List[Dict]:
"""
@@ -61,31 +96,47 @@ class MemoryActivator:
elif isinstance(observation, HFCloopObservation):
obs_info_text += observation.get_observe_info()
logger.debug(f"回忆待检索内容：obs_info_text: {obs_info_text}")
# logger.debug(f"回忆待检索内容：obs_info_text: {obs_info_text}")
# prompt = await global_prompt_manager.format_prompt(
# "memory_activator_prompt",
# obs_info_text=obs_info_text,
# )
# 将缓存的关键词转换为字符串，用于prompt
cached_keywords_str = ", ".join(self.cached_keywords) if self.cached_keywords else "暂无历史关键词"
# logger.debug(f"prompt: {prompt}")
# response = await self.summary_model.generate_response(prompt)
# logger.debug(f"response: {response}")
# # 只取response的第一个元素（字符串）
# response_str = response[0]
# keywords = list(get_keywords_from_json(response_str))
# #调用记忆系统获取相关记忆
# related_memory = await HippocampusManager.get_instance().get_memory_from_topic(
# valid_keywords=keywords, max_memory_num=3, max_memory_length=2, max_depth=3
# )
related_memory = await HippocampusManager.get_instance().get_memory_from_text(
text=obs_info_text, max_memory_num=5, max_memory_length=2, max_depth=3, fast_retrieval=True
prompt = await global_prompt_manager.format_prompt(
"memory_activator_prompt",
obs_info_text=obs_info_text,
cached_keywords=cached_keywords_str,
)
logger.debug(f"prompt: {prompt}")
response = await self.summary_model.generate_response(prompt)
logger.debug(f"response: {response}")
# 只取response的第一个元素（字符串）
response_str = response[0]
keywords = list(get_keywords_from_json(response_str))
# 更新关键词缓存
if keywords:
# 限制缓存大小，最多保留10个关键词
if len(self.cached_keywords) > 10:
# 转换为列表，移除最早的关键词
cached_list = list(self.cached_keywords)
self.cached_keywords = set(cached_list[-8:])
# 添加新的关键词到缓存
self.cached_keywords.update(keywords)
logger.debug(f"更新关键词缓存: {self.cached_keywords}")
#调用记忆系统获取相关记忆
related_memory = await HippocampusManager.get_instance().get_memory_from_topic(
valid_keywords=keywords, max_memory_num=3, max_memory_length=2, max_depth=3
)
# related_memory = await HippocampusManager.get_instance().get_memory_from_text(
# text=obs_info_text, max_memory_num=5, max_memory_length=2, max_depth=3, fast_retrieval=False
# )
# logger.debug(f"获取到的记忆: {related_memory}")
# 激活时，所有已有记忆的duration+1，达到3则移除

23
src/chat/focus_chat/planners/planner.py
View File

@@ -36,9 +36,8 @@ def init_prompt():
{mind_info_block}
{cycle_info_block}
{action_available_block}
请综合分析聊天内容和你看到的新消息，参考聊天规划，选择合适的action:
注意，除了下面动作选项之外，你在群聊里不能做其他任何事情，这是你能力的边界，现在请你选择合适的action:
{action_options_text}
@@ -126,13 +125,6 @@ class ActionPlanner:
action = "no_reply"
reasoning = f"之前选择的动作{action}已被移除，原因: {reason}"
using_actions = self.action_manager.get_using_actions()
action_available_block = ""
for action_name, action_info in using_actions.items():
action_description = action_info["description"]
action_available_block += f"\n你在聊天中可以使用{action_name}，这个动作的描述是{action_description}\n"
action_available_block += "注意，除了上述动作选项之外，你在群聊里不能做其他任何事情，这是你能力的边界\n"
# 继续处理其他信息
for info in all_plan_info:
if isinstance(info, ObsInfo):
@@ -147,7 +139,8 @@ class ActionPlanner:
elif isinstance(info, SelfInfo):
self_info = info.get_processed_info()
elif isinstance(info, StructuredInfo):
_structured_info = info.get_data()
structured_info = info.get_processed_info()
# print(f"structured_info: {structured_info}")
elif not isinstance(info, ActionInfo): # 跳过已处理的ActionInfo
extra_info.append(info.get_processed_info())
@@ -178,11 +171,10 @@ class ActionPlanner:
chat_target_info=None,
observed_messages_str=observed_messages_str, # <-- Pass local variable
current_mind=current_mind, # <-- Pass argument
# structured_info=structured_info, # <-- Pass SubMind info
structured_info=structured_info, # <-- Pass SubMind info
current_available_actions=current_available_actions, # <-- Pass determined actions
cycle_info=cycle_info, # <-- Pass cycle info
extra_info=extra_info,
action_available_block=action_available_block,
)
# --- 调用 LLM (普通文本生成) ---
@@ -268,7 +260,7 @@ class ActionPlanner:
chat_target_info: Optional[dict], # Now passed as argument
observed_messages_str: str,
current_mind: Optional[str],
action_available_block: str,
structured_info: Optional[str],
current_available_actions: Dict[str, ActionInfo],
cycle_info: Optional[str],
extra_info: list[str],
@@ -326,7 +318,8 @@ class ActionPlanner:
action_options_block += using_action_prompt
extra_info_block = "\n".join(extra_info)
if extra_info:
extra_info_block += f"\n{structured_info}"
if extra_info or structured_info:
extra_info_block = f"以下是一些额外的信息，现在请你阅读以下内容，进行决策\n{extra_info_block}\n以上是一些额外的信息，现在请你阅读以下内容，进行决策"
else:
extra_info_block = ""
@@ -343,7 +336,7 @@ class ActionPlanner:
mind_info_block=mind_info_block,
cycle_info_block=cycle_info,
action_options_text=action_options_block,
action_available_block=action_available_block,
# action_available_block=action_available_block,
extra_info_block=extra_info_block,
moderation_prompt=moderation_prompt_block,
)

4
src/chat/memory_system/Hippocampus.py
View File

@@ -526,12 +526,12 @@ class Hippocampus:
if not keywords:
return []
# logger.info(f"提取的关键词: {', '.join(keywords)}")
logger.info(f"提取的关键词: {', '.join(keywords)}")
# 过滤掉不存在于记忆图中的关键词
valid_keywords = [keyword for keyword in keywords if keyword in self.memory_graph.G]
if not valid_keywords:
# logger.info("没有找到有效的关键词节点")
logger.info("没有找到有效的关键词节点")
return []
logger.debug(f"有效的关键词: {', '.join(valid_keywords)}")

4
src/individuality/expression_style.py
View File

@@ -33,10 +33,10 @@ def init_prompt() -> None:
class PersonalityExpression:
def __init__(self):
self.express_learn_model: LLMRequest = LLMRequest(
model=global_config.model.normal,
model=global_config.model.focus_expressor,
temperature=0.1,
max_tokens=256,
request_type="response_heartflow",
request_type="learn_expression",
)
self.meta_file_path = os.path.join("data", "expression", "personality", "expression_style_meta.json")
self.expressions_file_path = os.path.join("data", "expression", "personality", "expressions.json")

3
template/bot_config_template.toml
View File

@@ -255,7 +255,8 @@ provider = "SILICONFLOW"
pri_in = 2
pri_out = 8
#表达器模型，用于生成表达方式
#表达器模型，用于表达麦麦的想法，生成最终回复，对语言风格影响极大
#也用于表达方式学习
[model.focus_expressor]
name = "Pro/deepseek-ai/DeepSeek-V3"
provider = "SILICONFLOW"