feat:pfc Lite(hearfFC）在群聊初步可用
This commit is contained in:
SengokuCola
@@ -38,7 +38,7 @@ class ResponseGenerator:
|
||||
self.current_model_type = "r1" # 默认使用 R1
|
||||
self.current_model_name = "unknown model"
|
||||
|
||||
async def generate_response(self, message: MessageRecv, thinking_id: str) -> Optional[List[str]]:
|
||||
async def generate_response(self, message: MessageRecv, thinking_id: str,) -> Optional[List[str]]:
|
||||
"""根据当前模型类型选择对应的生成函数"""
|
||||
|
||||
logger.info(
|
||||
|
||||
@@ -1,8 +1,9 @@
|
||||
import time
|
||||
from random import random
|
||||
import traceback
|
||||
from typing import List, Optional
|
||||
from typing import List, Optional, Dict
|
||||
import asyncio
|
||||
from asyncio import Lock
|
||||
from ...moods.moods import MoodManager
|
||||
from ....config.config import global_config
|
||||
from ...chat.emoji_manager import emoji_manager
|
||||
@@ -19,7 +20,8 @@ from ...utils.timer_calculater import Timer
|
||||
from src.do_tool.tool_use import ToolUser
|
||||
from .interest import InterestManager, InterestChatting
|
||||
from src.plugins.chat.chat_stream import chat_manager
|
||||
from src.plugins.chat.message import MessageInfo
|
||||
from src.plugins.chat.message import BaseMessageInfo
|
||||
from .pf_chatting import PFChatting
|
||||
|
||||
# 定义日志配置
|
||||
chat_config = LogConfig(
|
||||
@@ -33,13 +35,32 @@ logger = get_module_logger("heartFC_chat", config=chat_config)
|
||||
INTEREST_MONITOR_INTERVAL_SECONDS = 1
|
||||
|
||||
class HeartFC_Chat:
|
||||
_instance = None # For potential singleton access if needed by MessageManager
|
||||
|
||||
def __init__(self):
|
||||
# --- Updated Init ---
|
||||
if HeartFC_Chat._instance is not None:
|
||||
# Prevent re-initialization if used as a singleton
|
||||
return
|
||||
self.logger = logger # Make logger accessible via self
|
||||
self.gpt = ResponseGenerator()
|
||||
self.mood_manager = MoodManager.get_instance()
|
||||
self.mood_manager.start_mood_update()
|
||||
self.tool_user = ToolUser()
|
||||
self.interest_manager = InterestManager()
|
||||
self._interest_monitor_task: Optional[asyncio.Task] = None
|
||||
# --- New PFChatting Management ---
|
||||
self.pf_chatting_instances: Dict[str, PFChatting] = {}
|
||||
self._pf_chatting_lock = Lock()
|
||||
# --- End New PFChatting Management ---
|
||||
HeartFC_Chat._instance = self # Register instance
|
||||
# --- End Updated Init ---
|
||||
|
||||
# --- Added Class Method for Singleton Access ---
|
||||
@classmethod
|
||||
def get_instance(cls):
|
||||
return cls._instance
|
||||
# --- End Added Class Method ---
|
||||
|
||||
async def start(self):
|
||||
"""启动异步任务,如兴趣监控器"""
|
||||
@@ -61,14 +82,29 @@ class HeartFC_Chat:
|
||||
else:
|
||||
logger.warning("跳过兴趣监控任务创建：任务已存在或正在运行。")
|
||||
|
||||
# --- Added PFChatting Instance Manager ---
|
||||
async def _get_or_create_pf_chatting(self, stream_id: str) -> Optional[PFChatting]:
|
||||
"""获取现有PFChatting实例或创建新实例。"""
|
||||
async with self._pf_chatting_lock:
|
||||
if stream_id not in self.pf_chatting_instances:
|
||||
self.logger.info(f"为流 {stream_id} 创建新的PFChatting实例")
|
||||
# 传递 self (HeartFC_Chat 实例) 进行依赖注入
|
||||
instance = PFChatting(stream_id, self)
|
||||
# 执行异步初始化
|
||||
if not await instance._initialize():
|
||||
self.logger.error(f"为流 {stream_id} 初始化PFChatting失败")
|
||||
return None
|
||||
self.pf_chatting_instances[stream_id] = instance
|
||||
return self.pf_chatting_instances[stream_id]
|
||||
# --- End Added PFChatting Instance Manager ---
|
||||
|
||||
async def _interest_monitor_loop(self):
|
||||
"""后台任务，定期检查兴趣度变化并触发回复"""
|
||||
logger.info("兴趣监控循环开始...")
|
||||
while True:
|
||||
await asyncio.sleep(INTEREST_MONITOR_INTERVAL_SECONDS)
|
||||
try:
|
||||
# --- 修改：遍历 SubHeartflow 并检查触发器 ---
|
||||
active_stream_ids = list(heartflow.get_all_subheartflows_streams_ids()) # 需要 heartflow 提供此方法
|
||||
active_stream_ids = list(heartflow.get_all_subheartflows_streams_ids())
|
||||
logger.trace(f"检查 {len(active_stream_ids)} 个活跃流是否足以开启心流对话...")
|
||||
|
||||
for stream_id in active_stream_ids:
|
||||
@@ -77,26 +113,28 @@ class HeartFC_Chat:
|
||||
logger.warning(f"监控循环: 无法获取活跃流 {stream_id} 的 sub_hf")
|
||||
continue
|
||||
|
||||
# --- 获取 Observation 和消息列表 --- #
|
||||
observation = sub_hf._get_primary_observation()
|
||||
if not observation:
|
||||
logger.warning(f"[{stream_id}] SubHeartflow 没有在观察，无法检查触发器。")
|
||||
continue
|
||||
observed_messages = observation.talking_message # 获取消息字典列表
|
||||
# --- 结束获取 --- #
|
||||
|
||||
should_trigger = False
|
||||
try:
|
||||
# check_reply_trigger 可以选择性地接收 observed_messages 作为参数
|
||||
should_trigger = await sub_hf.check_reply_trigger() # 目前 check_reply_trigger 还不处理这个
|
||||
interest_chatting = self.interest_manager.get_interest_chatting(stream_id)
|
||||
if interest_chatting:
|
||||
should_trigger = interest_chatting.should_evaluate_reply()
|
||||
if should_trigger:
|
||||
logger.info(f"[{stream_id}] 基于兴趣概率决定启动交流模式 (概率: {interest_chatting.current_reply_probability:.4f})。")
|
||||
else:
|
||||
logger.trace(f"[{stream_id}] 没有找到对应的 InterestChatting 实例，跳过基于兴趣的触发检查。")
|
||||
except Exception as e:
|
||||
logger.error(f"错误调用 check_reply_trigger 流 {stream_id}: {e}")
|
||||
logger.error(f"检查兴趣触发器时出错 流 {stream_id}: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
|
||||
if should_trigger:
|
||||
logger.info(f"[{stream_id}] SubHeartflow 决定开启心流对话。")
|
||||
# 调用修改后的处理函数，传递 stream_id 和 observed_messages
|
||||
asyncio.create_task(self._process_triggered_reply(stream_id, observed_messages))
|
||||
logger.info(f"[{stream_id}] 触发条件满足, 委托给PFChatting.")
|
||||
# --- 修改: 获取 PFChatting 实例并调用 add_time (无参数，时间由内部衰减逻辑决定) ---
|
||||
pf_instance = await self._get_or_create_pf_chatting(stream_id)
|
||||
if pf_instance:
|
||||
# 调用 add_time 启动或延长循环，时间由 PFChatting 内部决定
|
||||
asyncio.create_task(pf_instance.add_time())
|
||||
else:
|
||||
logger.error(f"[{stream_id}] 无法获取或创建PFChatting实例。跳过触发。")
|
||||
|
||||
|
||||
except asyncio.CancelledError:
|
||||
@@ -107,32 +145,6 @@ class HeartFC_Chat:
|
||||
logger.error(traceback.format_exc())
|
||||
await asyncio.sleep(5) # 发生错误时等待
|
||||
|
||||
async def _process_triggered_reply(self, stream_id: str, observed_messages: List[dict]):
|
||||
"""Helper coroutine to handle the processing of a triggered reply based on SubHeartflow trigger."""
|
||||
try:
|
||||
logger.info(f"[{stream_id}] SubHeartflow 触发回复...")
|
||||
# 调用修改后的 trigger_reply_generation
|
||||
await self.trigger_reply_generation(stream_id, observed_messages)
|
||||
|
||||
# --- 调整兴趣降低逻辑 ---
|
||||
# 这里的兴趣降低可能不再适用，或者需要基于不同的逻辑
|
||||
# 例如，回复后可以将 SubHeartflow 的某种"回复意愿"状态重置
|
||||
# 暂时注释掉，或根据需要调整
|
||||
# chatting_instance = self.interest_manager.get_interest_chatting(stream_id)
|
||||
# if chatting_instance:
|
||||
# decrease_value = chatting_instance.trigger_threshold / 2 # 使用实例的阈值
|
||||
# self.interest_manager.decrease_interest(stream_id, value=decrease_value)
|
||||
# post_trigger_interest = self.interest_manager.get_interest(stream_id) # 获取更新后的兴趣
|
||||
# logger.info(f"[{stream_id}] Interest decreased by {decrease_value:.2f} (InstanceThreshold/2) after processing triggered reply. Current interest: {post_trigger_interest:.2f}")
|
||||
# else:
|
||||
# logger.warning(f"[{stream_id}] Could not find InterestChatting instance after reply processing to decrease interest.")
|
||||
logger.debug(f"[{stream_id}] Reply processing finished. (Interest decrease logic needs review).")
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Error processing SubHeartflow-triggered reply for stream_id {stream_id}: {e}") # 更新日志信息
|
||||
logger.error(traceback.format_exc())
|
||||
# --- 结束修改 ---
|
||||
|
||||
async def _create_thinking_message(self, anchor_message: Optional[MessageRecv]):
|
||||
"""创建思考消息 (尝试锚定到 anchor_message)"""
|
||||
if not anchor_message or not anchor_message.chat_stream:
|
||||
@@ -270,7 +282,7 @@ class HeartFC_Chat:
|
||||
sub_hf = None
|
||||
anchor_message: Optional[MessageRecv] = None # <--- 重命名，用于锚定回复的消息对象
|
||||
userinfo: Optional[UserInfo] = None
|
||||
messageinfo: Optional[MessageInfo] = None
|
||||
messageinfo: Optional[BaseMessageInfo] = None
|
||||
|
||||
timing_results = {}
|
||||
current_mind = None
|
||||
@@ -295,33 +307,58 @@ class HeartFC_Chat:
|
||||
logger.error(traceback.format_exc())
|
||||
return
|
||||
|
||||
# --- 2. 尝试从 observed_messages 重建最后一条消息作为锚点 --- #
|
||||
# --- 2. 尝试从 observed_messages 重建最后一条消息作为锚点, 失败则创建占位符 --- #
|
||||
try:
|
||||
with Timer("获取最后消息锚点", timing_results):
|
||||
with Timer("获取或创建锚点消息", timing_results):
|
||||
reconstruction_failed = False
|
||||
if observed_messages:
|
||||
last_msg_dict = observed_messages[-1] # 直接从传入列表获取最后一条
|
||||
# 尝试从字典重建 MessageRecv 对象（可能需要调整 MessageRecv 的构造方式或创建一个辅助函数）
|
||||
# 这是一个简化示例，假设 MessageRecv 可以从字典初始化
|
||||
# 你可能需要根据 MessageRecv 的实际 __init__ 来调整
|
||||
try:
|
||||
anchor_message = MessageRecv(last_msg_dict) # 假设 MessageRecv 支持从字典创建
|
||||
last_msg_dict = observed_messages[-1]
|
||||
logger.debug(f"[{stream_id}] Attempting to reconstruct MessageRecv from last observed message.")
|
||||
anchor_message = MessageRecv(last_msg_dict, chat_stream=chat)
|
||||
if not (anchor_message and anchor_message.message_info and anchor_message.message_info.message_id and anchor_message.message_info.user_info):
|
||||
raise ValueError("Reconstructed MessageRecv missing essential info.")
|
||||
userinfo = anchor_message.message_info.user_info
|
||||
messageinfo = anchor_message.message_info
|
||||
logger.debug(f"[{stream_id}] 获取到最后消息作为锚点: ID={messageinfo.message_id}, Sender={userinfo.user_nickname}")
|
||||
except Exception as e_msg:
|
||||
logger.error(f"[{stream_id}] 从字典重建最后消息 MessageRecv 失败: {e_msg}. 字典: {last_msg_dict}")
|
||||
anchor_message = None # 重置以表示失败
|
||||
logger.debug(f"[{stream_id}] Successfully reconstructed anchor message: ID={messageinfo.message_id}, Sender={userinfo.user_nickname}")
|
||||
except Exception as e_reconstruct:
|
||||
logger.warning(f"[{stream_id}] Reconstructing MessageRecv from observed message failed: {e_reconstruct}. Will create placeholder.")
|
||||
reconstruction_failed = True
|
||||
else:
|
||||
logger.warning(f"[{stream_id}] 无法从 Observation 获取最后消息锚点。")
|
||||
except Exception as e:
|
||||
logger.error(f"[{stream_id}] 获取最后消息锚点时出错: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
# 即使没有锚点，也可能继续尝试生成非回复性消息，取决于后续逻辑
|
||||
logger.warning(f"[{stream_id}] observed_messages is empty. Will create placeholder anchor message.")
|
||||
reconstruction_failed = True # Treat empty observed_messages as a failure to reconstruct
|
||||
|
||||
# --- 3. 检查是否能继续 (需要思考消息锚点) ---
|
||||
if not anchor_message:
|
||||
logger.warning(f"[{stream_id}] 没有有效的消息锚点，无法创建思考消息和发送回复。取消回复生成。")
|
||||
return
|
||||
# 如果重建失败或 observed_messages 为空，创建占位符
|
||||
if reconstruction_failed:
|
||||
placeholder_id = f"mid_{int(time.time() * 1000)}" # 使用毫秒时间戳增加唯一性
|
||||
placeholder_user = UserInfo(user_id="system_trigger", user_nickname="系统触发")
|
||||
placeholder_msg_info = BaseMessageInfo(
|
||||
message_id=placeholder_id,
|
||||
platform=chat.platform,
|
||||
group_info=chat.group_info,
|
||||
user_info=placeholder_user,
|
||||
time=time.time()
|
||||
# 其他 BaseMessageInfo 可能需要的字段设为默认值或 None
|
||||
)
|
||||
# 创建 MessageRecv 实例，注意它需要消息字典结构，我们创建一个最小化的
|
||||
placeholder_msg_dict = {
|
||||
"message_info": placeholder_msg_info.to_dict(),
|
||||
"processed_plain_text": "", # 提供空文本
|
||||
"raw_message": "",
|
||||
"time": placeholder_msg_info.time,
|
||||
}
|
||||
# 先只用字典创建实例
|
||||
anchor_message = MessageRecv(placeholder_msg_dict)
|
||||
# 然后调用方法更新 chat_stream
|
||||
anchor_message.update_chat_stream(chat)
|
||||
userinfo = anchor_message.message_info.user_info
|
||||
messageinfo = anchor_message.message_info
|
||||
logger.info(f"[{stream_id}] Created placeholder anchor message: ID={messageinfo.message_id}, Sender={userinfo.user_nickname}")
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"[{stream_id}] 获取或创建锚点消息时出错: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
anchor_message = None # 确保出错时 anchor_message 为 None
|
||||
|
||||
# --- 4. 检查并发思考限制 (使用 anchor_message 简化获取) ---
|
||||
try:
|
||||
@@ -399,6 +436,7 @@ class HeartFC_Chat:
|
||||
with Timer("生成内心想法(SubHF)", timing_results):
|
||||
# 不再传递 message_txt 和 sender_info, SubHeartflow 应基于其内部观察
|
||||
current_mind, past_mind = await sub_hf.do_thinking_before_reply(
|
||||
# sender_info=userinfo,
|
||||
chat_stream=chat,
|
||||
extra_info=tool_result_info,
|
||||
obs_id=get_mid_memory_id,
|
||||
@@ -415,7 +453,8 @@ class HeartFC_Chat:
|
||||
# --- 9. 调用 ResponseGenerator 生成回复 (使用 anchor_message 和 current_mind) ---
|
||||
try:
|
||||
with Timer("生成最终回复(GPT)", timing_results):
|
||||
response_set = await self.gpt.generate_response(anchor_message, thinking_id, current_mind=current_mind)
|
||||
# response_set = await self.gpt.generate_response(anchor_message, thinking_id, current_mind=current_mind)
|
||||
response_set = await self.gpt.generate_response(anchor_message, thinking_id)
|
||||
except Exception as e:
|
||||
logger.error(f"[{stream_id}] GPT 生成回复失败: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
|
||||
@@ -20,11 +20,11 @@ logger = get_module_logger("InterestManager", config=interest_log_config)
|
||||
|
||||
|
||||
# 定义常量
|
||||
DEFAULT_DECAY_RATE_PER_SECOND = 0.95 # 每秒衰减率 (兴趣保留 99%)
|
||||
MAX_INTEREST = 10.0 # 最大兴趣值
|
||||
MIN_INTEREST_THRESHOLD = 0.1 # 低于此值可能被清理 (可选)
|
||||
DEFAULT_DECAY_RATE_PER_SECOND = 0.98 # 每秒衰减率 (兴趣保留 99%)
|
||||
MAX_INTEREST = 15.0 # 最大兴趣值
|
||||
# MIN_INTEREST_THRESHOLD = 0.1 # 低于此值可能被清理 (可选)
|
||||
CLEANUP_INTERVAL_SECONDS = 3600 # 清理任务运行间隔 (例如：1小时)
|
||||
INACTIVE_THRESHOLD_SECONDS = 3600 * 24 # 不活跃时间阈值 (例如：1天)
|
||||
INACTIVE_THRESHOLD_SECONDS = 3600 # 不活跃时间阈值 (例如：1小时)
|
||||
LOG_INTERVAL_SECONDS = 3 # 日志记录间隔 (例如：30秒)
|
||||
LOG_DIRECTORY = "logs/interest" # 日志目录
|
||||
LOG_FILENAME = "interest_log.json" # 快照日志文件名 (保留，以防其他地方用到)
|
||||
@@ -33,11 +33,11 @@ HISTORY_LOG_FILENAME = "interest_history.log" # 新的历史日志文件名
|
||||
# INTEREST_INCREASE_THRESHOLD = 0.5
|
||||
|
||||
# --- 新增：概率回复相关常量 ---
|
||||
REPLY_TRIGGER_THRESHOLD = 5.0 # 触发概率回复的兴趣阈值 (示例值)
|
||||
REPLY_TRIGGER_THRESHOLD = 3.0 # 触发概率回复的兴趣阈值 (示例值)
|
||||
BASE_REPLY_PROBABILITY = 0.05 # 首次超过阈值时的基础回复概率 (示例值)
|
||||
PROBABILITY_INCREASE_RATE_PER_SECOND = 0.02 # 高于阈值时，每秒概率增加量 (线性增长, 示例值)
|
||||
PROBABILITY_DECAY_FACTOR_PER_SECOND = 0.3 # 低于阈值时，每秒概率衰减因子 (指数衰减, 示例值)
|
||||
MAX_REPLY_PROBABILITY = 0.95 # 回复概率上限 (示例值)
|
||||
MAX_REPLY_PROBABILITY = 1 # 回复概率上限 (示例值)
|
||||
# --- 结束：概率回复相关常量 ---
|
||||
|
||||
class InterestChatting:
|
||||
@@ -117,15 +117,15 @@ class InterestChatting:
|
||||
# 持续高于阈值，线性增加概率
|
||||
increase_amount = self.probability_increase_rate * time_delta
|
||||
self.current_reply_probability += increase_amount
|
||||
logger.debug(f"兴趣高于阈值 ({self.trigger_threshold}) 持续 {time_delta:.2f}秒. 概率增加 {increase_amount:.4f} 到 {self.current_reply_probability:.4f}")
|
||||
# logger.debug(f"兴趣高于阈值 ({self.trigger_threshold}) 持续 {time_delta:.2f}秒. 概率增加 {increase_amount:.4f} 到 {self.current_reply_probability:.4f}")
|
||||
|
||||
# 限制概率不超过最大值
|
||||
self.current_reply_probability = min(self.current_reply_probability, self.max_reply_probability)
|
||||
|
||||
else: # 低于阈值
|
||||
if self.is_above_threshold:
|
||||
# 刚低于阈值，开始衰减
|
||||
logger.debug(f"兴趣低于阈值 ({self.trigger_threshold}). 概率衰减开始于 {self.current_reply_probability:.4f}")
|
||||
# if self.is_above_threshold:
|
||||
# # 刚低于阈值，开始衰减
|
||||
# logger.debug(f"兴趣低于阈值 ({self.trigger_threshold}). 概率衰减开始于 {self.current_reply_probability:.4f}")
|
||||
# else: # 持续低于阈值，继续衰减
|
||||
# pass # 不需要特殊处理
|
||||
|
||||
@@ -133,12 +133,12 @@ class InterestChatting:
|
||||
# 检查 decay_factor 是否有效
|
||||
if 0 < self.probability_decay_factor < 1:
|
||||
decay_multiplier = math.pow(self.probability_decay_factor, time_delta)
|
||||
old_prob = self.current_reply_probability
|
||||
# old_prob = self.current_reply_probability
|
||||
self.current_reply_probability *= decay_multiplier
|
||||
# 避免因浮点数精度问题导致概率略微大于0，直接设为0
|
||||
if self.current_reply_probability < 1e-6:
|
||||
self.current_reply_probability = 0.0
|
||||
logger.debug(f"兴趣低于阈值 ({self.trigger_threshold}) 持续 {time_delta:.2f}秒. 概率从 {old_prob:.4f} 衰减到 {self.current_reply_probability:.4f} (因子: {self.probability_decay_factor})")
|
||||
# logger.debug(f"兴趣低于阈值 ({self.trigger_threshold}) 持续 {time_delta:.2f}秒. 概率从 {old_prob:.4f} 衰减到 {self.current_reply_probability:.4f} (因子: {self.probability_decay_factor})")
|
||||
elif self.probability_decay_factor <= 0:
|
||||
# 如果衰减因子无效或为0，直接清零
|
||||
if self.current_reply_probability > 0:
|
||||
@@ -212,19 +212,19 @@ class InterestChatting:
|
||||
# 确保概率是基于最新兴趣值计算的
|
||||
self._update_reply_probability(current_time)
|
||||
# 更新兴趣衰减（如果需要，取决于逻辑，这里保持和 get_interest 一致）
|
||||
self._calculate_decay(current_time)
|
||||
self.last_update_time = current_time # 更新时间戳
|
||||
# self._calculate_decay(current_time)
|
||||
# self.last_update_time = current_time # 更新时间戳
|
||||
|
||||
if self.is_above_threshold and self.current_reply_probability > 0:
|
||||
if self.current_reply_probability > 0:
|
||||
# 只有在阈值之上且概率大于0时才有可能触发
|
||||
trigger = random.random() < self.current_reply_probability
|
||||
if trigger:
|
||||
logger.info(f"Reply evaluation triggered! Probability: {self.current_reply_probability:.4f}, Threshold: {self.trigger_threshold}, Interest: {self.interest_level:.2f}")
|
||||
logger.info(f"回复概率评估触发! 概率: {self.current_reply_probability:.4f}, 阈值: {self.trigger_threshold}, 兴趣: {self.interest_level:.2f}")
|
||||
# 可选：触发后是否重置/降低概率？根据需要决定
|
||||
# self.current_reply_probability = self.base_reply_probability # 例如，触发后降回基础概率
|
||||
# self.current_reply_probability *= 0.5 # 例如，触发后概率减半
|
||||
else:
|
||||
logger.debug(f"Reply evaluation NOT triggered. Probability: {self.current_reply_probability:.4f}, Random value: {trigger + 1e-9:.4f}") # 打印随机值用于调试
|
||||
logger.debug(f"回复概率评估未触发。概率: {self.current_reply_probability:.4f}")
|
||||
return trigger
|
||||
else:
|
||||
# logger.debug(f"Reply evaluation check: Below threshold or zero probability. Probability: {self.current_reply_probability:.4f}")
|
||||
@@ -271,12 +271,12 @@ class InterestManager:
|
||||
except OSError as e:
|
||||
logger.error(f"Error creating log directory '{LOG_DIRECTORY}': {e}")
|
||||
|
||||
async def _periodic_cleanup_task(self, interval_seconds: int, threshold: float, max_age_seconds: int):
|
||||
async def _periodic_cleanup_task(self, interval_seconds: int, max_age_seconds: int):
|
||||
"""后台清理任务的异步函数"""
|
||||
while True:
|
||||
await asyncio.sleep(interval_seconds)
|
||||
logger.info(f"运行定期清理 (间隔: {interval_seconds}秒)...")
|
||||
self.cleanup_inactive_chats(threshold=threshold, max_age_seconds=max_age_seconds)
|
||||
self.cleanup_inactive_chats(max_age_seconds=max_age_seconds)
|
||||
|
||||
async def _periodic_log_task(self, interval_seconds: int):
|
||||
"""后台日志记录任务的异步函数 (记录历史数据，包含 group_name)"""
|
||||
@@ -318,7 +318,7 @@ class InterestManager:
|
||||
# 将每个条目作为单独的 JSON 行写入
|
||||
f.write(json.dumps(log_entry, ensure_ascii=False) + '\n')
|
||||
count += 1
|
||||
logger.debug(f"Successfully appended {count} interest history entries to {self._history_log_file_path}")
|
||||
# logger.debug(f"Successfully appended {count} interest history entries to {self._history_log_file_path}")
|
||||
|
||||
# 注意：不再写入快照文件 interest_log.json
|
||||
# 如果需要快照文件，可以在这里单独写入 self._snapshot_log_file_path
|
||||
@@ -358,7 +358,6 @@ class InterestManager:
|
||||
self._cleanup_task = asyncio.create_task(
|
||||
self._periodic_cleanup_task(
|
||||
interval_seconds=CLEANUP_INTERVAL_SECONDS,
|
||||
threshold=MIN_INTEREST_THRESHOLD,
|
||||
max_age_seconds=INACTIVE_THRESHOLD_SECONDS
|
||||
)
|
||||
)
|
||||
@@ -449,10 +448,9 @@ class InterestManager:
|
||||
else:
|
||||
logger.warning(f"尝试降低不存在的聊天流 {stream_id} 的兴趣度")
|
||||
|
||||
def cleanup_inactive_chats(self, threshold=MIN_INTEREST_THRESHOLD, max_age_seconds=INACTIVE_THRESHOLD_SECONDS):
|
||||
def cleanup_inactive_chats(self, max_age_seconds=INACTIVE_THRESHOLD_SECONDS):
|
||||
"""
|
||||
清理长时间不活跃的聊天流记录
|
||||
threshold: 低于此兴趣度的将被清理
|
||||
max_age_seconds: 超过此时间未更新的将被清理
|
||||
"""
|
||||
current_time = time.time()
|
||||
|
||||
726
src/plugins/chat_module/heartFC_chat/pf_chatting.py
Normal file
726
src/plugins/chat_module/heartFC_chat/pf_chatting.py
Normal file
@@ -0,0 +1,726 @@
|
||||
import asyncio
|
||||
import time
|
||||
import traceback
|
||||
from typing import List, Optional, Dict, Any, Deque, Union, TYPE_CHECKING
|
||||
from collections import deque
|
||||
import json
|
||||
|
||||
from ....config.config import global_config
|
||||
from ...chat.message import MessageRecv, BaseMessageInfo, MessageThinking, MessageSending
|
||||
from ...chat.chat_stream import ChatStream
|
||||
from ...message import UserInfo
|
||||
from src.heart_flow.heartflow import heartflow, SubHeartflow
|
||||
from src.plugins.chat.chat_stream import chat_manager
|
||||
from .messagesender import MessageManager
|
||||
from src.common.logger import get_module_logger, LogConfig, DEFAULT_CONFIG # 引入 DEFAULT_CONFIG
|
||||
from src.plugins.models.utils_model import LLMRequest
|
||||
from src.individuality.individuality import Individuality
|
||||
|
||||
# 定义日志配置 (使用 loguru 格式)
|
||||
interest_log_config = LogConfig(
|
||||
console_format=DEFAULT_CONFIG["console_format"], # 使用默认控制台格式
|
||||
file_format=DEFAULT_CONFIG["file_format"] # 使用默认文件格式
|
||||
)
|
||||
logger = get_module_logger("PFChattingLoop", config=interest_log_config) # Logger Name Changed
|
||||
|
||||
|
||||
# Forward declaration for type hinting
|
||||
if TYPE_CHECKING:
|
||||
from .heartFC_chat import HeartFC_Chat
|
||||
|
||||
PLANNER_TOOL_DEFINITION = [
|
||||
{
|
||||
"type": "function",
|
||||
"function": {
|
||||
"name": "decide_reply_action",
|
||||
"description": "根据当前聊天内容和上下文，决定机器人是否应该回复以及如何回复。",
|
||||
"parameters": {
|
||||
"type": "object",
|
||||
"properties": {
|
||||
"action": {
|
||||
"type": "string",
|
||||
"enum": ["no_reply", "text_reply", "emoji_reply"],
|
||||
"description": "决定采取的行动：'no_reply'(不回复), 'text_reply'(文本回复) 或 'emoji_reply'(表情回复)。"
|
||||
},
|
||||
"reasoning": {
|
||||
"type": "string",
|
||||
"description": "做出此决定的简要理由。"
|
||||
},
|
||||
"emoji_query": {
|
||||
"type": "string",
|
||||
"description": '如果行动是\'emoji_reply\',则指定表情的主题或概念（例如，"开心"、"困惑"）。仅在需要表情回复时提供。'
|
||||
}
|
||||
},
|
||||
"required": ["action", "reasoning"] # 强制要求提供行动和理由
|
||||
}
|
||||
}
|
||||
}
|
||||
]
|
||||
|
||||
class PFChatting:
|
||||
"""
|
||||
Manages a continuous Plan-Filter-Check (now Plan-Replier-Sender) loop
|
||||
for generating replies within a specific chat stream, controlled by a timer.
|
||||
The loop runs as long as the timer > 0.
|
||||
"""
|
||||
def __init__(self, chat_id: str, heartfc_chat_instance: 'HeartFC_Chat'):
|
||||
"""
|
||||
初始化PFChatting实例。
|
||||
|
||||
Args:
|
||||
chat_id: The identifier for the chat stream (e.g., stream_id).
|
||||
heartfc_chat_instance: 访问共享资源和方法的主HeartFC_Chat实例。
|
||||
"""
|
||||
self.heartfc_chat = heartfc_chat_instance # 访问logger, gpt, tool_user, _send_response_messages等。
|
||||
self.stream_id: str = chat_id
|
||||
self.chat_stream: Optional[ChatStream] = None
|
||||
self.sub_hf: Optional[SubHeartflow] = None
|
||||
self._initialized = False
|
||||
self._init_lock = asyncio.Lock() # Ensure initialization happens only once
|
||||
self._processing_lock = asyncio.Lock() # 确保只有一个 Plan-Replier-Sender 周期在运行
|
||||
self._timer_lock = asyncio.Lock() # 用于安全更新计时器
|
||||
|
||||
self.planner_llm = LLMRequest(
|
||||
model=global_config.llm_normal,
|
||||
temperature=global_config.llm_normal["temp"],
|
||||
max_tokens=1000,
|
||||
request_type="action_planning"
|
||||
)
|
||||
|
||||
# Internal state for loop control
|
||||
self._loop_timer: float = 0.0 # Remaining time for the loop in seconds
|
||||
self._loop_active: bool = False # Is the loop currently running?
|
||||
self._loop_task: Optional[asyncio.Task] = None # Stores the main loop task
|
||||
self._trigger_count_this_activation: int = 0 # Counts triggers within an active period
|
||||
|
||||
# Removed pending_replies as processing is now serial within the loop
|
||||
# self.pending_replies: Dict[str, PendingReply] = {}
|
||||
|
||||
|
||||
async def _initialize(self) -> bool:
|
||||
"""
|
||||
Lazy initialization to resolve chat_stream and sub_hf using the provided identifier.
|
||||
Ensures the instance is ready to handle triggers.
|
||||
"""
|
||||
async with self._init_lock:
|
||||
if self._initialized:
|
||||
return True
|
||||
try:
|
||||
self.chat_stream = chat_manager.get_stream(self.stream_id)
|
||||
|
||||
if not self.chat_stream:
|
||||
logger.error(f"PFChatting-{self.stream_id} 获取ChatStream失败。")
|
||||
return False
|
||||
|
||||
# 子心流(SubHeartflow)可能初始不存在但后续会被创建
|
||||
# 在需要它的方法中应优雅处理其可能缺失的情况
|
||||
self.sub_hf = heartflow.get_subheartflow(self.stream_id)
|
||||
if not self.sub_hf:
|
||||
logger.warning(f"PFChatting-{self.stream_id} 获取SubHeartflow失败。一些功能可能受限。")
|
||||
# 决定是否继续初始化。目前允许初始化。
|
||||
|
||||
self._initialized = True
|
||||
logger.info(f"PFChatting-{self.stream_id} 初始化成功。")
|
||||
return True
|
||||
except Exception as e:
|
||||
logger.error(f"PFChatting-{self.stream_id} 初始化失败: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
return False
|
||||
|
||||
async def add_time(self):
|
||||
"""
|
||||
Adds time to the loop timer with decay and starts the loop if it's not active.
|
||||
Called externally (e.g., by HeartFC_Chat) to trigger or extend activity.
|
||||
Durations: 1st trigger = 10s, 2nd = 5s, 3rd+ = 2s.
|
||||
"""
|
||||
if not self._initialized:
|
||||
if not await self._initialize():
|
||||
logger.error(f"PFChatting-{self.stream_id} 无法添加时间: 未初始化。")
|
||||
return
|
||||
|
||||
async with self._timer_lock:
|
||||
duration_to_add: float = 0.0
|
||||
|
||||
if not self._loop_active: # First trigger for this activation cycle
|
||||
duration_to_add = 10.0
|
||||
self._trigger_count_this_activation = 1 # Start counting for this activation
|
||||
logger.info(f"[{self.stream_id}] First trigger in activation. Adding {duration_to_add:.1f}s.")
|
||||
else: # Loop is already active, apply decay
|
||||
self._trigger_count_this_activation += 1
|
||||
if self._trigger_count_this_activation == 2:
|
||||
duration_to_add = 5.0
|
||||
logger.info(f"[{self.stream_id}] 2nd trigger in activation. Adding {duration_to_add:.1f}s.")
|
||||
else: # 3rd trigger or more
|
||||
duration_to_add = 2.0
|
||||
logger.info(f"[{self.stream_id}] {self._trigger_count_this_activation}rd/+ trigger in activation. Adding {duration_to_add:.1f}s.")
|
||||
|
||||
new_timer_value = self._loop_timer + duration_to_add
|
||||
self._loop_timer = max(0, new_timer_value) # Ensure timer doesn't go negative conceptually
|
||||
logger.info(f"[{self.stream_id}] Timer is now {self._loop_timer:.1f}s.")
|
||||
|
||||
if not self._loop_active and self._loop_timer > 0:
|
||||
logger.info(f"[{self.stream_id}] Timer > 0 and loop not active. Starting PF loop.")
|
||||
self._loop_active = True
|
||||
# Cancel previous task just in case (shouldn't happen if logic is correct)
|
||||
if self._loop_task and not self._loop_task.done():
|
||||
logger.warning(f"[{self.stream_id}] Found existing loop task unexpectedly during start. Cancelling it.")
|
||||
self._loop_task.cancel()
|
||||
|
||||
self._loop_task = asyncio.create_task(self._run_pf_loop())
|
||||
# Add callback to reset state if loop finishes or errors out
|
||||
self._loop_task.add_done_callback(self._handle_loop_completion)
|
||||
elif self._loop_active:
|
||||
logger.debug(f"[{self.stream_id}] Loop already active. Timer extended.")
|
||||
|
||||
|
||||
def _handle_loop_completion(self, task: asyncio.Task):
|
||||
"""Callback executed when the _run_pf_loop task finishes."""
|
||||
try:
|
||||
# Check if the task raised an exception
|
||||
exception = task.exception()
|
||||
if exception:
|
||||
logger.error(f"[{self.stream_id}] PF loop task completed with error: {exception}")
|
||||
logger.error(traceback.format_exc())
|
||||
else:
|
||||
logger.info(f"[{self.stream_id}] PF loop task completed normally (timer likely expired or cancelled).")
|
||||
except asyncio.CancelledError:
|
||||
logger.info(f"[{self.stream_id}] PF loop task was cancelled.")
|
||||
finally:
|
||||
# Reset state regardless of how the task finished
|
||||
self._loop_active = False
|
||||
self._loop_task = None
|
||||
# Ensure lock is released if the loop somehow exited while holding it
|
||||
if self._processing_lock.locked():
|
||||
logger.warning(f"[{self.stream_id}] Releasing processing lock after loop task completion.")
|
||||
self._processing_lock.release()
|
||||
logger.info(f"[{self.stream_id}] Loop state reset.")
|
||||
|
||||
|
||||
async def _run_pf_loop(self):
|
||||
"""
|
||||
主循环，当计时器>0时持续进行计划并可能回复消息
|
||||
管理每个循环周期的处理锁
|
||||
"""
|
||||
logger.info(f"[{self.stream_id}] 开始执行PF循环")
|
||||
try:
|
||||
while True:
|
||||
# 使用计时器锁安全地检查当前计时器值
|
||||
async with self._timer_lock:
|
||||
current_timer = self._loop_timer
|
||||
if current_timer <= 0:
|
||||
logger.info(f"[{self.stream_id}] 计时器为零或负数({current_timer:.1f}秒)，退出PF循环")
|
||||
break # 退出条件：计时器到期
|
||||
|
||||
# 记录循环开始时间
|
||||
loop_cycle_start_time = time.monotonic()
|
||||
# 标记本周期是否执行了操作
|
||||
action_taken_this_cycle = False
|
||||
|
||||
# 获取处理锁，确保每个计划-回复-发送周期独占执行
|
||||
acquired_lock = False
|
||||
try:
|
||||
await self._processing_lock.acquire()
|
||||
acquired_lock = True
|
||||
logger.debug(f"[{self.stream_id}] 循环获取到处理锁")
|
||||
|
||||
# --- Planner ---
|
||||
# Planner decides action, reasoning, emoji_query, etc.
|
||||
planner_result = await self._planner() # Modify planner to return decision dict
|
||||
action = planner_result.get("action", "error")
|
||||
reasoning = planner_result.get("reasoning", "Planner did not provide reasoning.")
|
||||
emoji_query = planner_result.get("emoji_query", "")
|
||||
current_mind = planner_result.get("current_mind", "[Mind unavailable]")
|
||||
send_emoji_from_tools = planner_result.get("send_emoji_from_tools", "")
|
||||
observed_messages = planner_result.get("observed_messages", []) # Planner needs to return this
|
||||
|
||||
if action == "text_reply":
|
||||
logger.info(f"[{self.stream_id}] 计划循环决定: 回复文本.")
|
||||
action_taken_this_cycle = True
|
||||
# --- 回复器 ---
|
||||
anchor_message = await self._get_anchor_message(observed_messages)
|
||||
if not anchor_message:
|
||||
logger.error(f"[{self.stream_id}] 循环: 无法获取锚点消息用于回复. 跳过周期.")
|
||||
else:
|
||||
thinking_id = await self.heartfc_chat._create_thinking_message(anchor_message)
|
||||
if not thinking_id:
|
||||
logger.error(f"[{self.stream_id}] 循环: 无法创建思考ID. 跳过周期.")
|
||||
else:
|
||||
replier_result = None
|
||||
try:
|
||||
# 直接 await 回复器工作
|
||||
replier_result = await self._replier_work(
|
||||
observed_messages=observed_messages,
|
||||
anchor_message=anchor_message,
|
||||
thinking_id=thinking_id,
|
||||
current_mind=current_mind,
|
||||
send_emoji=send_emoji_from_tools
|
||||
)
|
||||
except Exception as e_replier:
|
||||
logger.error(f"[{self.stream_id}] 循环: 回复器工作失败: {e_replier}")
|
||||
self._cleanup_thinking_message(thinking_id) # 清理思考消息
|
||||
# 继续循环, 视为非操作周期
|
||||
|
||||
if replier_result:
|
||||
# --- Sender ---
|
||||
try:
|
||||
await self._sender(thinking_id, anchor_message, replier_result)
|
||||
logger.info(f"[{self.stream_id}] 循环: 发送器完成成功.")
|
||||
except Exception as e_sender:
|
||||
logger.error(f"[{self.stream_id}] 循环: 发送器失败: {e_sender}")
|
||||
self._cleanup_thinking_message(thinking_id) # 确保发送失败时清理
|
||||
# 继续循环, 视为非操作周期
|
||||
else:
|
||||
# Replier failed to produce result
|
||||
logger.warning(f"[{self.stream_id}] 循环: 回复器未产生结果. 跳过发送.")
|
||||
self._cleanup_thinking_message(thinking_id) # 清理思考消息
|
||||
|
||||
elif action == "emoji_reply":
|
||||
logger.info(f"[{self.stream_id}] 计划循环决定: 回复表情 ('{emoji_query}').")
|
||||
action_taken_this_cycle = True
|
||||
anchor = await self._get_anchor_message(observed_messages)
|
||||
if anchor:
|
||||
try:
|
||||
await self.heartfc_chat._handle_emoji(anchor, [], emoji_query)
|
||||
except Exception as e_emoji:
|
||||
logger.error(f"[{self.stream_id}] 循环: 发送表情失败: {e_emoji}")
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}] 循环: 无法发送表情, 无法获取锚点.")
|
||||
|
||||
elif action == "no_reply":
|
||||
logger.info(f"[{self.stream_id}] 计划循环决定: 不回复. 原因: {reasoning}")
|
||||
# Do nothing else, action_taken_this_cycle remains False
|
||||
|
||||
elif action == "error":
|
||||
logger.error(f"[{self.stream_id}] 计划循环返回错误或失败. 原因: {reasoning}")
|
||||
# 视为非操作周期
|
||||
|
||||
else: # Unknown action
|
||||
logger.warning(f"[{self.stream_id}] 计划循环返回未知动作: {action}. 视为不回复.")
|
||||
# 视为非操作周期
|
||||
|
||||
except Exception as e_cycle:
|
||||
# Catch errors occurring within the locked section (e.g., planner crash)
|
||||
logger.error(f"[{self.stream_id}] 循环周期执行时发生错误: {e_cycle}")
|
||||
logger.error(traceback.format_exc())
|
||||
# Ensure lock is released if an error occurs before the finally block
|
||||
if acquired_lock and self._processing_lock.locked():
|
||||
self._processing_lock.release()
|
||||
acquired_lock = False # 防止在 finally 块中重复释放
|
||||
logger.warning(f"[{self.stream_id}] 由于循环周期中的错误释放了处理锁.")
|
||||
|
||||
finally:
|
||||
# Ensure the lock is always released after a cycle
|
||||
if acquired_lock:
|
||||
self._processing_lock.release()
|
||||
logger.debug(f"[{self.stream_id}] 循环释放了处理锁.")
|
||||
|
||||
# --- Timer Decrement ---
|
||||
cycle_duration = time.monotonic() - loop_cycle_start_time
|
||||
async with self._timer_lock:
|
||||
self._loop_timer -= cycle_duration
|
||||
logger.debug(f"[{self.stream_id}] 循环周期耗时 {cycle_duration:.2f}s. 计时器剩余: {self._loop_timer:.1f}s.")
|
||||
|
||||
# --- Delay ---
|
||||
# Add a small delay, especially if no action was taken, to prevent busy-waiting
|
||||
try:
|
||||
if not action_taken_this_cycle and cycle_duration < 1.5:
|
||||
# If nothing happened and cycle was fast, wait a bit longer
|
||||
await asyncio.sleep(1.5 - cycle_duration)
|
||||
elif cycle_duration < 0.2: # Minimum delay even if action was taken
|
||||
await asyncio.sleep(0.2)
|
||||
except asyncio.CancelledError:
|
||||
logger.info(f"[{self.stream_id}] Sleep interrupted, likely loop cancellation.")
|
||||
break # Exit loop if cancelled during sleep
|
||||
|
||||
except asyncio.CancelledError:
|
||||
logger.info(f"[{self.stream_id}] PF loop task received cancellation request.")
|
||||
except Exception as e_loop_outer:
|
||||
# Catch errors outside the main cycle lock (should be rare)
|
||||
logger.error(f"[{self.stream_id}] PF loop encountered unexpected outer error: {e_loop_outer}")
|
||||
logger.error(traceback.format_exc())
|
||||
finally:
|
||||
# Reset trigger count when loop finishes
|
||||
async with self._timer_lock:
|
||||
self._trigger_count_this_activation = 0
|
||||
logger.debug(f"[{self.stream_id}] Trigger count reset to 0 as loop finishes.")
|
||||
logger.info(f"[{self.stream_id}] PF loop finished execution run.")
|
||||
# State reset (_loop_active, _loop_task) is handled by _handle_loop_completion callback
|
||||
|
||||
async def _planner(self) -> Dict[str, Any]:
|
||||
"""
|
||||
规划器 (Planner): 使用LLM根据上下文决定是否和如何回复。
|
||||
Returns a dictionary containing the decision and context.
|
||||
{'action': str, 'reasoning': str, 'emoji_query': str, 'current_mind': str,
|
||||
'send_emoji_from_tools': str, 'observed_messages': List[dict]}
|
||||
"""
|
||||
observed_messages: List[dict] = []
|
||||
tool_result_info = {}
|
||||
get_mid_memory_id = []
|
||||
send_emoji_from_tools = "" # Renamed for clarity
|
||||
current_mind: Optional[str] = None
|
||||
|
||||
# --- 获取最新的观察信息 ---
|
||||
try:
|
||||
if self.sub_hf and self.sub_hf._get_primary_observation():
|
||||
observation = self.sub_hf._get_primary_observation()
|
||||
logger.debug(f"[{self.stream_id}][Planner] 调用 observation.observe()...")
|
||||
await observation.observe() # 主动观察以获取最新消息
|
||||
observed_messages = observation.talking_message # 获取更新后的消息列表
|
||||
logger.debug(f"[{self.stream_id}][Planner] 获取到 {len(observed_messages)} 条观察消息。")
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}][Planner] 无法获取 SubHeartflow 或 Observation 来获取消息。")
|
||||
except Exception as e:
|
||||
logger.error(f"[{self.stream_id}][Planner] 获取观察信息时出错: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
# --- 结束获取观察信息 ---
|
||||
|
||||
# --- (Moved from _replier_work) 1. 思考前使用工具 ---
|
||||
try:
|
||||
observation_context_text = ""
|
||||
if observed_messages:
|
||||
context_texts = [msg.get('detailed_plain_text', '') for msg in observed_messages if msg.get('detailed_plain_text')]
|
||||
observation_context_text = "\n".join(context_texts)
|
||||
logger.debug(f"[{self.stream_id}][Planner] Context for tools: {observation_context_text[:100]}...")
|
||||
|
||||
if observation_context_text and self.sub_hf:
|
||||
# Ensure SubHeartflow exists for tool use context
|
||||
tool_result = await self.heartfc_chat.tool_user.use_tool(
|
||||
message_txt=observation_context_text,
|
||||
chat_stream=self.chat_stream,
|
||||
sub_heartflow=self.sub_hf
|
||||
)
|
||||
if tool_result.get("used_tools", False):
|
||||
tool_result_info = tool_result.get("structured_info", {})
|
||||
logger.debug(f"[{self.stream_id}][Planner] Tool results: {tool_result_info}")
|
||||
if "mid_chat_mem" in tool_result_info:
|
||||
get_mid_memory_id = [mem["content"] for mem in tool_result_info["mid_chat_mem"] if "content" in mem]
|
||||
if "send_emoji" in tool_result_info and tool_result_info["send_emoji"]:
|
||||
send_emoji_from_tools = tool_result_info["send_emoji"][0].get("content", "") # Use renamed var
|
||||
elif not self.sub_hf:
|
||||
logger.warning(f"[{self.stream_id}][Planner] Skipping tool use because SubHeartflow is not available.")
|
||||
|
||||
except Exception as e_tool:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Planner] Tool use failed: {e_tool}")
|
||||
# Continue even if tool use fails
|
||||
# --- 结束工具使用 ---
|
||||
|
||||
# 心流思考，然后plan
|
||||
try:
|
||||
if self.sub_hf:
|
||||
# Ensure arguments match the current do_thinking_before_reply signature
|
||||
current_mind, past_mind = await self.sub_hf.do_thinking_before_reply(
|
||||
chat_stream=self.chat_stream,
|
||||
extra_info=tool_result_info,
|
||||
obs_id=get_mid_memory_id,
|
||||
)
|
||||
logger.info(f"[{self.stream_id}][Planner] SubHeartflow thought: {current_mind}")
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}][Planner] Skipping SubHeartflow thinking because it is not available.")
|
||||
current_mind = "[心流思考不可用]" # Set a default/indicator value
|
||||
|
||||
except Exception as e_shf:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Planner] SubHeartflow thinking failed: {e_shf}")
|
||||
logger.error(traceback.format_exc())
|
||||
current_mind = "[心流思考出错]"
|
||||
|
||||
|
||||
# --- 使用 LLM 进行决策 ---
|
||||
action = "no_reply" # Default action
|
||||
emoji_query = ""
|
||||
reasoning = "默认决策或获取决策失败"
|
||||
llm_error = False # Flag for LLM failure
|
||||
|
||||
try:
|
||||
# 构建提示 (Now includes current_mind)
|
||||
prompt = self._build_planner_prompt(observed_messages, current_mind)
|
||||
logger.trace(f"[{self.stream_id}][Planner] Prompt: {prompt}")
|
||||
|
||||
# 准备 LLM 请求 Payload
|
||||
payload = {
|
||||
"model": self.planner_llm.model_name,
|
||||
"messages": [{"role": "user", "content": prompt}],
|
||||
"tools": PLANNER_TOOL_DEFINITION,
|
||||
"tool_choice": {"type": "function", "function": {"name": "decide_reply_action"}}, # 强制调用此工具
|
||||
}
|
||||
|
||||
logger.debug(f"[{self.stream_id}][Planner] 发送 Planner LLM 请求...")
|
||||
# 调用 LLM
|
||||
response = await self.planner_llm._execute_request(
|
||||
endpoint="/chat/completions", payload=payload, prompt=prompt
|
||||
)
|
||||
|
||||
# 解析 LLM 响应
|
||||
if len(response) == 3: # 期望返回 content, reasoning_content, tool_calls
|
||||
_, _, tool_calls = response
|
||||
if tool_calls and isinstance(tool_calls, list) and len(tool_calls) > 0:
|
||||
# 通常强制调用后只会有一个 tool_call
|
||||
tool_call = tool_calls[0]
|
||||
if tool_call.get("type") == "function" and tool_call.get("function", {}).get("name") == "decide_reply_action":
|
||||
try:
|
||||
arguments = json.loads(tool_call["function"]["arguments"])
|
||||
action = arguments.get("action", "no_reply")
|
||||
reasoning = arguments.get("reasoning", "未提供理由")
|
||||
if action == "emoji_reply":
|
||||
# Planner's decision overrides tool's emoji if action is emoji_reply
|
||||
emoji_query = arguments.get("emoji_query", send_emoji_from_tools) # Use tool emoji as default if planner asks for emoji
|
||||
logger.info(f"[{self.stream_id}][Planner] LLM 决策: {action}, 理由: {reasoning}, EmojiQuery: '{emoji_query}'")
|
||||
except json.JSONDecodeError as json_e:
|
||||
logger.error(f"[{self.stream_id}][Planner] 解析工具参数失败: {json_e}. Arguments: {tool_call['function'].get('arguments')}")
|
||||
action = "error"; reasoning = "工具参数解析失败"; llm_error = True
|
||||
except Exception as parse_e:
|
||||
logger.error(f"[{self.stream_id}][Planner] 处理工具参数时出错: {parse_e}")
|
||||
action = "error"; reasoning = "处理工具参数时出错"; llm_error = True
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}][Planner] LLM 未按预期调用 'decide_reply_action' 工具。Tool calls: {tool_calls}")
|
||||
action = "error"; reasoning = "LLM未调用预期工具"; llm_error = True
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}][Planner] LLM 响应中未包含有效的工具调用。Tool calls: {tool_calls}")
|
||||
action = "error"; reasoning = "LLM响应无工具调用"; llm_error = True
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}][Planner] LLM 未返回预期的工具调用响应。Response parts: {len(response)}")
|
||||
action = "error"; reasoning = "LLM响应格式错误"; llm_error = True
|
||||
|
||||
except Exception as llm_e:
|
||||
logger.error(f"[{self.stream_id}][Planner] Planner LLM 调用失败: {llm_e}")
|
||||
logger.error(traceback.format_exc())
|
||||
action = "error"; reasoning = f"LLM 调用失败: {llm_e}"; llm_error = True
|
||||
|
||||
# --- 返回决策结果 ---
|
||||
# Note: Lock release is handled by the loop now
|
||||
return {
|
||||
"action": action,
|
||||
"reasoning": reasoning,
|
||||
"emoji_query": emoji_query, # Specific query if action is emoji_reply
|
||||
"current_mind": current_mind,
|
||||
"send_emoji_from_tools": send_emoji_from_tools, # Emoji suggested by pre-thinking tools
|
||||
"observed_messages": observed_messages,
|
||||
"llm_error": llm_error # Indicate if LLM decision process failed
|
||||
}
|
||||
|
||||
async def _get_anchor_message(self, observed_messages: List[dict]) -> Optional[MessageRecv]:
|
||||
"""
|
||||
重构观察到的最后一条消息作为回复的锚点，
|
||||
如果重构失败或观察为空，则创建一个占位符。
|
||||
"""
|
||||
if not self.chat_stream:
|
||||
logger.error(f"[PFChatting-{self.stream_id}] 无法获取锚点消息: ChatStream 不可用.")
|
||||
return None
|
||||
|
||||
try:
|
||||
last_msg_dict = None
|
||||
if observed_messages:
|
||||
last_msg_dict = observed_messages[-1]
|
||||
|
||||
if last_msg_dict:
|
||||
try:
|
||||
# Attempt reconstruction from the last observed message dictionary
|
||||
anchor_message = MessageRecv(last_msg_dict, chat_stream=self.chat_stream)
|
||||
# Basic validation
|
||||
if not (anchor_message and anchor_message.message_info and anchor_message.message_info.message_id and anchor_message.message_info.user_info):
|
||||
raise ValueError("重构的 MessageRecv 缺少必要信息.")
|
||||
logger.debug(f"[{self.stream_id}] 重构的锚点消息: ID={anchor_message.message_info.message_id}")
|
||||
return anchor_message
|
||||
except Exception as e_reconstruct:
|
||||
logger.warning(f"[{self.stream_id}] 从观察到的消息重构 MessageRecv 失败: {e_reconstruct}. 创建占位符.")
|
||||
else:
|
||||
logger.warning(f"[{self.stream_id}] observed_messages 为空. 创建占位符锚点消息.")
|
||||
|
||||
# --- Create Placeholder ---
|
||||
placeholder_id = f"mid_pf_{int(time.time() * 1000)}"
|
||||
placeholder_user = UserInfo(user_id="system_trigger", user_nickname="System Trigger", platform=self.chat_stream.platform)
|
||||
placeholder_msg_info = BaseMessageInfo(
|
||||
message_id=placeholder_id,
|
||||
platform=self.chat_stream.platform,
|
||||
group_info=self.chat_stream.group_info,
|
||||
user_info=placeholder_user,
|
||||
time=time.time()
|
||||
)
|
||||
placeholder_msg_dict = {
|
||||
"message_info": placeholder_msg_info.to_dict(),
|
||||
"processed_plain_text": "[System Trigger Context]", # Placeholder text
|
||||
"raw_message": "",
|
||||
"time": placeholder_msg_info.time,
|
||||
}
|
||||
anchor_message = MessageRecv(placeholder_msg_dict)
|
||||
anchor_message.update_chat_stream(self.chat_stream) # Associate with the stream
|
||||
logger.info(f"[{self.stream_id}] Created placeholder anchor message: ID={anchor_message.message_info.message_id}")
|
||||
return anchor_message
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"[PFChatting-{self.stream_id}] Error getting/creating anchor message: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
return None
|
||||
|
||||
def _cleanup_thinking_message(self, thinking_id: str):
|
||||
"""Safely removes the thinking message."""
|
||||
try:
|
||||
container = MessageManager().get_container(self.stream_id)
|
||||
container.remove_message(thinking_id, msg_type=MessageThinking)
|
||||
logger.debug(f"[{self.stream_id}] Cleaned up thinking message {thinking_id}.")
|
||||
except Exception as e:
|
||||
logger.error(f"[{self.stream_id}] Error cleaning up thinking message {thinking_id}: {e}")
|
||||
|
||||
|
||||
async def _sender(self, thinking_id: str, anchor_message: MessageRecv, replier_result: Dict[str, Any]):
|
||||
"""
|
||||
发送器 (Sender): 使用HeartFC_Chat的方法发送生成的回复。
|
||||
被 _run_pf_loop 直接调用和 await。
|
||||
也处理相关的操作，如发送表情和更新关系。
|
||||
Raises exception on failure to signal the loop.
|
||||
"""
|
||||
# replier_result should contain 'response_set' and 'send_emoji'
|
||||
response_set = replier_result.get("response_set")
|
||||
send_emoji = replier_result.get("send_emoji", "") # Emoji determined by tools, passed via replier
|
||||
|
||||
if not response_set:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Called with empty response_set.")
|
||||
# Clean up thinking message before raising error
|
||||
self._cleanup_thinking_message(thinking_id)
|
||||
raise ValueError("Sender called with no response_set") # Signal failure to loop
|
||||
|
||||
first_bot_msg: Optional[MessageSending] = None
|
||||
send_success = False
|
||||
try:
|
||||
# --- Send the main text response ---
|
||||
logger.debug(f"[{self.stream_id}][Sender-{thinking_id}] Sending response messages...")
|
||||
# This call implicitly handles replacing the MessageThinking with MessageSending/MessageSet
|
||||
first_bot_msg = await self.heartfc_chat._send_response_messages(anchor_message, response_set, thinking_id)
|
||||
|
||||
if first_bot_msg:
|
||||
send_success = True # Mark success
|
||||
logger.info(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Successfully sent reply.")
|
||||
|
||||
# --- Handle associated emoji (if determined by tools) ---
|
||||
if send_emoji:
|
||||
logger.info(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Sending associated emoji: {send_emoji}")
|
||||
try:
|
||||
# Use first_bot_msg as anchor if available, otherwise fallback to original anchor
|
||||
emoji_anchor = first_bot_msg if first_bot_msg else anchor_message
|
||||
await self.heartfc_chat._handle_emoji(emoji_anchor, response_set, send_emoji)
|
||||
except Exception as e_emoji:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Failed to send associated emoji: {e_emoji}")
|
||||
# Log error but don't fail the whole send process for emoji failure
|
||||
|
||||
# --- Update relationship ---
|
||||
try:
|
||||
await self.heartfc_chat._update_relationship(anchor_message, response_set)
|
||||
logger.debug(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Updated relationship.")
|
||||
except Exception as e_rel:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Failed to update relationship: {e_rel}")
|
||||
# Log error but don't fail the whole send process for relationship update failure
|
||||
|
||||
else:
|
||||
# Sending failed (e.g., _send_response_messages found thinking message already gone)
|
||||
send_success = False
|
||||
logger.warning(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Failed to send reply (maybe thinking message expired or was removed?).")
|
||||
# No need to clean up thinking message here, _send_response_messages implies it's gone or handled
|
||||
raise RuntimeError("Sending reply failed, _send_response_messages returned None.") # Signal failure
|
||||
|
||||
|
||||
except Exception as e:
|
||||
# Catch potential errors during sending or post-send actions
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Sender-{thinking_id}] Error during sending process: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
# Ensure thinking message is cleaned up if send failed mid-way and wasn't handled
|
||||
if not send_success:
|
||||
self._cleanup_thinking_message(thinking_id)
|
||||
raise # Re-raise the exception to signal failure to the loop
|
||||
|
||||
# No finally block needed for lock management
|
||||
|
||||
|
||||
async def shutdown(self):
|
||||
"""
|
||||
Gracefully shuts down the PFChatting instance by cancelling the active loop task.
|
||||
"""
|
||||
logger.info(f"[{self.stream_id}] Shutting down PFChatting...")
|
||||
if self._loop_task and not self._loop_task.done():
|
||||
logger.info(f"[{self.stream_id}] Cancelling active PF loop task.")
|
||||
self._loop_task.cancel()
|
||||
try:
|
||||
# Wait briefly for the task to acknowledge cancellation
|
||||
await asyncio.wait_for(self._loop_task, timeout=5.0)
|
||||
except asyncio.CancelledError:
|
||||
logger.info(f"[{self.stream_id}] PF loop task cancelled successfully.")
|
||||
except asyncio.TimeoutError:
|
||||
logger.warning(f"[{self.stream_id}] Timeout waiting for PF loop task cancellation.")
|
||||
except Exception as e:
|
||||
logger.error(f"[{self.stream_id}] Error during loop task cancellation: {e}")
|
||||
else:
|
||||
logger.info(f"[{self.stream_id}] No active PF loop task found to cancel.")
|
||||
|
||||
# Ensure loop state is reset even if task wasn't running or cancellation failed
|
||||
self._loop_active = False
|
||||
self._loop_task = None
|
||||
|
||||
# Double-check lock state (should be released by loop completion/cancellation handler)
|
||||
if self._processing_lock.locked():
|
||||
logger.warning(f"[{self.stream_id}] Releasing processing lock during shutdown.")
|
||||
self._processing_lock.release()
|
||||
|
||||
logger.info(f"[{self.stream_id}] PFChatting shutdown complete.")
|
||||
|
||||
def _build_planner_prompt(self, observed_messages: List[dict], current_mind: Optional[str]) -> str:
|
||||
"""构建 Planner LLM 的提示词 (现在包含 current_mind)"""
|
||||
prompt = "你是一个聊天机器人助手，正在决定是否以及如何回应当前的聊天。\n"
|
||||
prompt += f"你的名字是 {global_config.BOT_NICKNAME}。\n"
|
||||
|
||||
# Add current mind state if available
|
||||
if current_mind:
|
||||
prompt += f"\n你当前的内部想法是：\n---\n{current_mind}\n---\n\n"
|
||||
else:
|
||||
prompt += "\n你当前没有特别的内部想法。\n"
|
||||
|
||||
if observed_messages:
|
||||
context_text = "\n".join([msg.get('detailed_plain_text', '') for msg in observed_messages if msg.get('detailed_plain_text')])
|
||||
prompt += "观察到的最新聊天内容如下：\n---\n"
|
||||
prompt += context_text[:1500] # Limit context length
|
||||
prompt += "\n---\n"
|
||||
else:
|
||||
prompt += "当前没有观察到新的聊天内容。\n"
|
||||
|
||||
prompt += "\n请结合你的内部想法和观察到的聊天内容，分析情况并使用 'decide_reply_action' 工具来决定你的最终行动。\n"
|
||||
prompt += "决策依据：\n"
|
||||
prompt += "1. 如果聊天内容无聊、与你无关、或者你的内部想法认为不适合回复，选择 'no_reply'。\n"
|
||||
prompt += "2. 如果聊天内容值得回应，且适合用文字表达（参考你的内部想法），选择 'text_reply'。\n"
|
||||
prompt += "3. 如果聊天内容或你的内部想法适合用一个表情来回应，选择 'emoji_reply' 并提供表情主题 'emoji_query'。\n"
|
||||
prompt += "必须调用 'decide_reply_action' 工具并提供 'action' 和 'reasoning'。"
|
||||
|
||||
return prompt
|
||||
|
||||
# --- 回复器 (Replier) 的定义 --- #
|
||||
async def _replier_work(self, observed_messages: List[dict], anchor_message: MessageRecv, thinking_id: str, current_mind: Optional[str], send_emoji: str) -> Optional[Dict[str, Any]]:
|
||||
"""
|
||||
回复器 (Replier): 核心逻辑用于生成回复。
|
||||
被 _run_pf_loop 直接调用和 await。
|
||||
Returns dict with 'response_set' and 'send_emoji' or None on failure.
|
||||
"""
|
||||
response_set: Optional[List[str]] = None
|
||||
try:
|
||||
# --- Tool Use and SubHF Thinking are now in _planner ---
|
||||
|
||||
# --- Generate Response with LLM ---
|
||||
logger.debug(f"[{self.stream_id}][Replier-{thinking_id}] Calling LLM to generate response...")
|
||||
# 注意：实际的生成调用是在 self.heartfc_chat.gpt.generate_response 中
|
||||
response_set = await self.heartfc_chat.gpt.generate_response(
|
||||
anchor_message,
|
||||
thinking_id
|
||||
# current_mind 不再直接传递给 gpt.generate_response，
|
||||
# 因为 generate_response 内部会通过 thinking_id 或其他方式获取所需上下文
|
||||
)
|
||||
|
||||
if not response_set:
|
||||
logger.warning(f"[{self.stream_id}][Replier-{thinking_id}] LLM生成了一个空回复集。")
|
||||
return None # Indicate failure
|
||||
|
||||
# --- 准备并返回结果 ---
|
||||
logger.info(f"[{self.stream_id}][Replier-{thinking_id}] 成功生成了回复集: {' '.join(response_set)[:50]}...")
|
||||
return {
|
||||
"response_set": response_set,
|
||||
"send_emoji": send_emoji, # Pass through the emoji determined earlier (usually by tools)
|
||||
}
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"[PFChatting-{self.stream_id}][Replier-{thinking_id}] Unexpected error in replier_work: {e}")
|
||||
logger.error(traceback.format_exc())
|
||||
return None # Indicate failure
|
||||
22
src/plugins/chat_module/heartFC_chat/pfchating.md
Normal file
22
src/plugins/chat_module/heartFC_chat/pfchating.md
Normal file
@@ -0,0 +1,22 @@
|
||||
新写一个类，叫做pfchating
|
||||
这个类初始化时会输入一个chat_stream或者stream_id
|
||||
这个类会包含对应的sub_hearflow和一个chat_stream
|
||||
|
||||
pfchating有以下几个组成部分：
|
||||
规划器：决定是否要进行回复（根据sub_heartflow中的observe内容），可以选择不回复，回复文字或者回复表情包，你可以使用llm的工具调用来实现
|
||||
回复器：可以根据信息产生回复，这部分代码将大部分与trigger_reply_generation(stream_id, observed_messages)一模一样
|
||||
（回复器可能同时运行多个(0-3个)，这些回复器会根据不同时刻的规划器产生不同回复
|
||||
检查器：由于生成回复需要时间，检查器会检查在有了新的消息内容之后，回复是否还适合，如果合适就转给发送器
|
||||
如果一条消息被发送了，其他回复在检查时也要增加这条消息的信息，防止重复发送内容相近的回复
|
||||
发送器，将回复发送到聊天，这部分主体不需要再pfcchating中实现，只需要使用原有的self._send_response_messages(anchor_message, response_set, thinking_id)
|
||||
|
||||
|
||||
当_process_triggered_reply(self, stream_id: str, observed_messages: List[dict]):触发时，并不会单独进行一次回复
|
||||
|
||||
|
||||
问题：
|
||||
1.每个pfchating是否对应一个caht_stream，是否是唯一的？(fix)
|
||||
2.observe_text传入进来是纯str，是不是应该传进来message构成的list?(fix)
|
||||
3.检查失败的回复应该怎么处理？(先抛弃)
|
||||
4.如何比较相似度？
|
||||
5.planner怎么写？（好像可以先不加入这部分）
|
||||
Reference in New Issue
Block a user