fix:更名和小bug修复

This commit is contained in:
SengokuCola
2025-04-22 21:59:23 +08:00
parent 1482133005
commit 8d50a381e4
21 changed files with 852 additions and 468 deletions

View File

@@ -1,7 +1,5 @@
# 心流系统 (Heart Flow System)
心流系统是一个模拟AI机器人内心思考和情感流动的核心系统。它通过多层次的心流结构使AI能够对外界信息进行观察、思考和情感反应从而产生更自然的对话和行为。
## 系统架构
### 1. 主心流 (Heartflow)
@@ -24,22 +22,6 @@
- 支持多种观察类型(如聊天观察)
- 对信息进行实时总结和更新
## 主要功能
### 思维系统
- 定期进行思维更新
- 维护短期记忆和思维连续性
- 支持多层次的思维处理
### 情感系统
- 情绪状态管理
- 回复意愿判断
- 情感因素影响决策
### 交互系统
- 群聊消息处理
- 多场景并行处理
- 智能回复生成
## 工作流程
@@ -63,11 +45,6 @@ observation = ChattingObservation(chat_id)
subheartflow.add_observation(observation)
```
### 启动心流系统
```python
await heartflow.heartflow_start_working()
```
## 配置说明
系统的主要配置参数:
@@ -81,14 +58,100 @@ await heartflow.heartflow_start_working()
2. 需要合理配置更新间隔以平衡性能和响应速度
3. 观察系统会限制消息处理数量以避免过载
# PFChatting 与主动回复流程说明 (V2)
更新:
把聊天控制移动到心流下吧
首先心流要根据日程以及当前状况判定总体状态MaiStateInfo
本文档描述了 `PFChatting` 类及其在 `heartFC_controler` 模块中实现的主动、基于兴趣的回复流程。
然后根据每个子心流的运行情况给子心流分配聊天资源ChatStateInfoABSENT CHAT 或者 FOCUS
## 1. `PFChatting` 类概述
子心流负责根据状态进行执行
* **目标**: 管理特定聊天流 (`stream_id`) 的主动回复逻辑,使其行为更像人类的自然交流。
* **创建时机**: 当 `HeartFC_Chat` 的兴趣监控任务 (`_interest_monitor_loop`) 检测到某个聊天流的兴趣度 (`InterestChatting`) 达到了触发回复评估的条件 (`should_evaluate_reply`) 时,会为该 `stream_id` 获取或创建唯一的 `PFChatting` 实例 (`_get_or_create_heartFC_chat`)。
* **持有**:
* 对应的 `sub_heartflow` 实例引用 (通过 `heartflow.get_subheartflow(stream_id)`)。
* 对应的 `chat_stream` 实例引用。
*`HeartFC_Chat` 单例的引用 (用于调用发送消息、处理表情等辅助方法)。
* **初始化**: `PFChatting` 实例在创建后会执行异步初始化 (`_initialize`),这可能包括加载必要的上下文或历史信息(*待确认是否实现了读取历史消息*)。
1.将interest.py进行拆分class InterestChatting 将会在 sub_heartflow中声明每个sub_heartflow都会所属一个InterestChatting
class InterestManager 将会在heartflow中声明成为heartflow的一个组件伴随heartflow产生
## 2. 核心回复流程 (由 `HeartFC_Chat` 触发)
`HeartFC_Chat` 调用 `PFChatting` 实例的方法 (例如 `add_time`) 时,会启动内部的回复决策与执行流程:
1. **规划 (Planner):**
* **输入**: 从关联的 `sub_heartflow` 获取观察结果、思考链、记忆片段等上下文信息。
* **决策**:
* 判断当前是否适合进行回复。
* 决定回复的形式(纯文本、带表情包等)。
* 选择合适的回复时机和策略。
* **实现**: *此部分逻辑待详细实现,可能利用 LLM 的工具调用能力来增强决策的灵活性和智能性。需要考虑机器人的个性化设定。*
2. **回复生成 (Replier):**
* **输入**: Planner 的决策结果和必要的上下文。
* **执行**:
* 调用 `ResponseGenerator` (`self.gpt`) 或类似组件生成具体的回复文本内容。
* 可能根据 Planner 的策略生成多个候选回复。
* **并发**: 系统支持同时存在多个思考/生成任务(上限由 `global_config.max_concurrent_thinking_messages` 控制)。
3. **检查 (Checker):**
* **时机**: 在回复生成过程中或生成后、发送前执行。
* **目的**:
* 检查自开始生成回复以来,聊天流中是否出现了新的消息。
* 评估已生成的候选回复在新的上下文下是否仍然合适、相关。
* *需要实现相似度比较逻辑,防止发送与近期消息内容相近或重复的回复。*
* **处理**: 如果检查结果认为回复不合适,则该回复将被**抛弃**。
4. **发送协调:**
* **执行**: 如果 Checker 通过,`PFChatting` 会调用 `HeartFC_Chat` 实例提供的发送接口:
* `_create_thinking_message`: 通知 `MessageManager` 显示"正在思考"状态。
* `_send_response_messages`: 将最终的回复文本交给 `MessageManager` 进行排队和发送。
* `_handle_emoji`: 如果需要发送表情包,调用此方法处理表情包的获取和发送。
* **细节**: 实际的消息发送、排队、间隔控制由 `MessageManager``MessageSender` 负责。
## 3. 与其他模块的交互
* **`HeartFC_Chat`**:
* 创建、管理和触发 `PFChatting` 实例。
* 提供发送消息 (`_send_response_messages`)、处理表情 (`_handle_emoji`)、创建思考消息 (`_create_thinking_message`) 的接口给 `PFChatting` 调用。
* 运行兴趣监控循环 (`_interest_monitor_loop`)。
* **`InterestManager` / `InterestChatting`**:
* `InterestManager` 存储每个 `stream_id``InterestChatting` 实例。
* `InterestChatting` 负责计算兴趣衰减和回复概率。
* `HeartFC_Chat` 查询 `InterestChatting.should_evaluate_reply()` 来决定是否触发 `PFChatting`
* **`heartflow` / `sub_heartflow`**:
* `PFChatting` 从对应的 `sub_heartflow` 获取进行规划所需的核心上下文信息 (观察、思考链等)。
* **`MessageManager` / `MessageSender`**:
* 接收来自 `HeartFC_Chat` 的发送请求 (思考消息、文本消息、表情包消息)。
* 管理消息队列 (`MessageContainer`),处理消息发送间隔和实际发送 (`MessageSender`)。
* **`ResponseGenerator` (`gpt`)**:
*`PFChatting` 的 Replier 部分调用,用于生成回复文本。
* **`MessageStorage`**:
* 存储所有接收和发送的消息。
* **`HippocampusManager`**:
* `HeartFC_Processor` 使用它计算传入消息的记忆激活率,作为兴趣度计算的输入之一。
## 4. 原有问题与状态更新
1. **每个 `pfchating` 是否对应一个 `chat_stream`,是否是唯一的?**
* **是**`HeartFC_Chat._get_or_create_heartFC_chat` 确保了每个 `stream_id` 只有一个 `PFChatting` 实例。 (已确认)
2. **`observe_text` 传入进来是纯 str是不是应该传进来 message 构成的 list?**
* **机制已改变**。当前的触发机制是基于 `InterestManager` 的概率判断。`PFChatting` 启动后,应从其关联的 `sub_heartflow` 获取更丰富的上下文信息,而非简单的 `observe_text`
3. **检查失败的回复应该怎么处理?**
* **暂定:抛弃**。这是当前 Checker 逻辑的基础设定。
4. **如何比较相似度?**
* **待实现**。Checker 需要具体的算法来比较候选回复与新消息的相似度。
5. **Planner 怎么写?**
* **待实现**。这是 `PFChatting` 的核心决策逻辑,需要结合 `sub_heartflow` 的输出、LLM 工具调用和个性化配置来设计。
## 6. 未来优化点
* 实现 Checker 中的相似度比较算法。
* 详细设计并实现 Planner 的决策逻辑,包括 LLM 工具调用和个性化。
* 确认并完善 `PFChatting._initialize()` 中的历史消息加载逻辑。
* 探索更优的检查失败回复处理策略(例如:重新规划、修改回复等)。
* 优化 `PFChatting``sub_heartflow` 的信息交互。
BUG:
2.复读可能是planner还未校准好
3.planner还未个性化需要加入bot个性信息且获取的聊天内容有问题

View File

@@ -1,11 +0,0 @@
更新:
把聊天控制移动到心流下吧
首先心流要根据日程以及当前状况判定总体状态MaiStateInfo
然后根据每个子心流的运行情况给子心流分配聊天资源ChatStateInfoABSENT CHAT 或者 FOCUS
子心流负责根据状态进行执行
1.将interest.py进行拆分class InterestChatting 将会在 sub_heartflow中声明每个sub_heartflow都会所属一个InterestChatting
class InterestManager 将会在heartflow中声明成为heartflow的一个组件伴随heartflow产生

View File

@@ -1,4 +1,4 @@
from .sub_heartflow import SubHeartflow, ChattingObservation
from src.heart_flow.sub_heartflow import SubHeartflow, ChattingObservation, ChatState
from src.plugins.moods.moods import MoodManager
from src.plugins.models.utils_model import LLMRequest
from src.config.config import global_config
@@ -9,7 +9,7 @@ from src.common.logger import get_module_logger, LogConfig, HEARTFLOW_STYLE_CONF
from src.individuality.individuality import Individuality
import time
import random
from typing import Dict, Any, Optional
from typing import Dict, Any, Optional, TYPE_CHECKING
import traceback
import enum
import os # 新增
@@ -23,6 +23,9 @@ heartflow_config = LogConfig(
)
logger = get_module_logger("heartflow", config=heartflow_config)
# Type hinting for circular dependency
if TYPE_CHECKING:
from src.heart_flow.sub_heartflow import SubHeartflow, ChatState # Keep SubHeartflow here too
def init_prompt():
prompt = ""
@@ -53,6 +56,13 @@ INACTIVE_THRESHOLD_SECONDS = 1200 # 不活跃时间阈值 (例如20分钟) -
LOG_INTERVAL_SECONDS = 3 # 日志记录间隔 (例如3秒) - 保持与 interest.py 一致
# --- 结束新增常量 ---
# --- 新增:状态更新常量 ---
STATE_UPDATE_INTERVAL_SECONDS = 30 # 状态更新检查间隔(秒)
FIVE_MINUTES = 3 * 60
FIFTEEN_MINUTES = 6 * 60
TWENTY_MINUTES = 9 * 60
# --- 结束新增常量 ---
# 新增 ChatStatus 枚举
class MaiState(enum.Enum):
@@ -92,10 +102,12 @@ class MaiState(enum.Enum):
class MaiStateInfo:
def __init__(self):
self.current_state_info = ""
# 使用枚举类型初始化状态,默认为不在线
self.mai_status: MaiState = MaiState.OFFLINE
self.mai_status_history = [] # 历史状态,包含 状态,最后时间
self.last_status_change_time: float = time.time() # 新增:状态最后改变时间
self.last_5min_check_time: float = time.time() # 新增上次5分钟规则检查时间
self.normal_chatting = []
self.focused_chatting = []
@@ -103,17 +115,21 @@ class MaiStateInfo:
self.mood_manager = MoodManager()
self.mood = self.mood_manager.get_prompt()
def update_current_state_info(self):
self.current_state_info = self.mood_manager.get_current_mood()
# 新增更新聊天状态的方法
def update_mai_status(self, new_status: MaiState):
"""更新聊天状态"""
if isinstance(new_status, MaiState):
if isinstance(new_status, MaiState) and new_status != self.mai_status: # 只有状态实际改变时才更新
self.mai_status = new_status
current_time = time.time()
self.last_status_change_time = current_time # 更新状态改变时间
self.last_5min_check_time = current_time # 重置5分钟检查计时器
# 将新状态和时间戳添加到历史记录
self.mai_status_history.append((new_status, current_time))
logger.info(f"麦麦状态更新为: {self.mai_status.value}")
else:
elif not isinstance(new_status, MaiState):
logger.warning(f"尝试设置无效的麦麦状态: {new_status}")
# else: # 状态未改变,不处理
# pass
class Heartflow:
@@ -125,13 +141,14 @@ class Heartflow:
model=global_config.llm_heartflow, temperature=0.6, max_tokens=1000, request_type="heart_flow"
)
self._subheartflows: Dict[Any, SubHeartflow] = {}
self._subheartflows: Dict[Any, 'SubHeartflow'] = {} # Update type hint
# --- 新增:日志和清理相关属性 (从 InterestManager 移动) ---
self._history_log_file_path = os.path.join(LOG_DIRECTORY, HISTORY_LOG_FILENAME)
self._ensure_log_directory() # 初始化时确保目录存在
self._cleanup_task: Optional[asyncio.Task] = None
self._logging_task: Optional[asyncio.Task] = None
self._state_update_task: Optional[asyncio.Task] = None # 新增:状态更新任务
# 注意:衰减任务 (_decay_task) 不再需要,衰减在 SubHeartflow 的 InterestChatting 内部处理
# --- 结束新增属性 ---
@@ -203,6 +220,107 @@ class Heartflow:
logger.error(f"[Heartflow] Unexpected error during periodic history logging: {e}")
logger.error(traceback.format_exc()) # 记录 traceback
async def _periodic_state_update_task(self):
"""定期检查并更新 Mai 状态"""
while True:
await asyncio.sleep(STATE_UPDATE_INTERVAL_SECONDS)
try:
current_time = time.time()
# 获取更新前的状态
previous_status = self.current_state.mai_status
current_status = self.current_state.mai_status # 保持此行以进行后续逻辑
time_in_current_status = current_time - self.current_state.last_status_change_time
time_since_last_5min_check = current_time - self.current_state.last_5min_check_time
next_state = None # 预设下一状态为 None
# --- 状态转换逻辑 (保持不变) ---
# 1. 通用规则每5分钟检查 (对于非 OFFLINE 状态)
if time_since_last_5min_check >= FIVE_MINUTES:
self.current_state.last_5min_check_time = current_time # 重置5分钟检查计时器无论是否切换
if current_status != MaiState.OFFLINE:
if random.random() < 0.10: # 10% 概率切换到 OFFLINE
logger.debug(f"[Heartflow State] 触发5分钟规则{current_status.value} 切换到 OFFLINE")
next_state = MaiState.OFFLINE # 设置 next_state 而不是直接更新
# self.current_state.update_mai_status(MaiState.OFFLINE)
# continue # 状态已改变,进入下一轮循环
# 2. 状态持续时间规则 (仅在未被5分钟规则覆盖时执行)
if next_state is None: # 仅当5分钟规则未触发切换时检查持续时间
if current_status == MaiState.OFFLINE:
# OFFLINE 状态下检查是否已持续5分钟
if time_in_current_status >= FIVE_MINUTES:
weights = [35, 35, 30]
choices_list = [MaiState.PEEKING, MaiState.NORMAL_CHAT, MaiState.OFFLINE]
next_state_candidate = random.choices(choices_list, weights=weights, k=1)[0]
if next_state_candidate != MaiState.OFFLINE:
next_state = next_state_candidate
logger.debug(f"[Heartflow State] OFFLINE 持续时间达到,切换到 {next_state.value}")
else:
# 保持 OFFLINE重置计时器以开始新的5分钟计时
logger.debug(f"[Heartflow State] OFFLINE 持续时间达到,保持 OFFLINE重置计时器")
self.current_state.last_status_change_time = current_time
self.current_state.last_5min_check_time = current_time # 保持一致
# 显式将 next_state 设为 OFFLINE 以便后续处理
next_state = MaiState.OFFLINE
elif current_status == MaiState.PEEKING:
if time_in_current_status >= FIVE_MINUTES: # PEEKING 最多持续 5 分钟
weights = [50, 30, 20]
choices_list = [MaiState.OFFLINE, MaiState.NORMAL_CHAT, MaiState.FOCUSED_CHAT]
next_state = random.choices(choices_list, weights=weights, k=1)[0]
logger.debug(f"[Heartflow State] PEEKING 持续时间达到,切换到 {next_state.value}")
elif current_status == MaiState.NORMAL_CHAT:
if time_in_current_status >= FIFTEEN_MINUTES: # NORMAL_CHAT 最多持续 15 分钟
weights = [50, 50]
choices_list = [MaiState.OFFLINE, MaiState.FOCUSED_CHAT]
next_state = random.choices(choices_list, weights=weights, k=1)[0]
logger.debug(f"[Heartflow State] NORMAL_CHAT 持续时间达到,切换到 {next_state.value}")
elif current_status == MaiState.FOCUSED_CHAT:
if time_in_current_status >= TWENTY_MINUTES: # FOCUSED_CHAT 最多持续 20 分钟
weights = [80, 20]
choices_list = [MaiState.OFFLINE, MaiState.NORMAL_CHAT]
next_state = random.choices(choices_list, weights=weights, k=1)[0]
logger.debug(f"[Heartflow State] FOCUSED_CHAT 持续时间达到,切换到 {next_state.value}")
# --- 状态转换逻辑结束 ---
# --- 更新状态并执行相关操作 --- #
if next_state is not None:
# 检查状态是否真的改变了
if next_state != previous_status:
logger.info(f"[Heartflow] 准备从 {previous_status.value} 转换状态到 {next_state.value}")
self.current_state.update_mai_status(next_state)
# 在状态改变后,强制执行子心流数量限制 (保持)
await self._enforce_subheartflow_limits(next_state)
# --- 新增逻辑:根据状态转换调整子心流 --- #
if previous_status == MaiState.OFFLINE and next_state != MaiState.OFFLINE:
logger.info(f"[Heartflow] 主状态从 OFFLINE 激活,尝试激活子心流到 CHAT 状态。")
await self._activate_random_subflows_to_chat(next_state)
elif next_state == MaiState.OFFLINE and previous_status != MaiState.OFFLINE:
logger.info(f"[Heartflow] 主状态变为 OFFLINE停用所有子心流活动。")
await self._deactivate_all_subflows_on_offline()
# --- 结束新增逻辑 --- #
elif next_state == MaiState.OFFLINE and previous_status == MaiState.OFFLINE:
# 如果决定保持 OFFLINE 状态(例如,因为随机选择或持续时间规则),并且之前已经是 OFFLINE
# 确保计时器被重置 (这在上面的持续时间规则中已处理,但为了清晰再次确认)
if time_in_current_status >= FIVE_MINUTES:
# 确保计时器已在上面重置,这里无需操作,只记录日志
logger.debug(f"[Heartflow State] 保持 OFFLINE 状态,计时器已重置。")
pass # 无需状态转换,也无需调用激活/停用逻辑
# --- 如果没有确定 next_state (即没有触发任何切换规则) --- #
# logger.debug(f"[Heartflow State] 状态未改变,保持 {current_status.value}") # 减少日志噪音
except Exception as e:
logger.error(f"[Heartflow] 状态更新任务出错: {e}")
logger.error(traceback.format_exc())
logger.info(f"当前状态:{self.current_state.mai_status.value}")
def get_all_interest_states(self) -> Dict[str, Dict]: # 新增方法
"""获取所有活跃子心流的当前兴趣状态"""
states = {}
@@ -266,16 +384,6 @@ class Heartflow:
# logger.info(f"[Heartflow] 清理完成。没有流符合移除条件。当前数量: {initial_count}") # 减少日志噪音
pass
async def _sub_heartflow_update(self): # 这个任务目前作用不大,可以考虑移除或赋予新职责
while True:
# 检查是否存在子心流
if not self._subheartflows:
# logger.info("当前没有子心流,等待新的子心流创建...")
await asyncio.sleep(30) # 短暂休眠
continue
# 当前无实际操作,只是等待
await asyncio.sleep(300)
async def heartflow_start_working(self):
# 启动清理任务 (使用新的 periodic_cleanup_task)
@@ -299,8 +407,14 @@ class Heartflow:
else:
logger.warning("[Heartflow] 跳过创建日志任务: 任务已在运行或存在。")
# (可选) 启动旧的子心流更新任务,如果它还有用的话
# asyncio.create_task(self._sub_heartflow_update())
# 新增:启动状态更新任务
if self._state_update_task is None or self._state_update_task.done():
self._state_update_task = asyncio.create_task(self._periodic_state_update_task())
logger.info(f"[Heartflow] 已创建定期状态更新任务。间隔: {STATE_UPDATE_INTERVAL_SECONDS}s")
else:
logger.warning("[Heartflow] 跳过创建状态更新任务: 任务已在运行或存在。")
@staticmethod
async def _update_current_state():
@@ -308,7 +422,6 @@ class Heartflow:
async def do_a_thinking(self):
# logger.debug("麦麦大脑袋转起来了")
self.current_state.update_current_state_info()
# 开始构建prompt
prompt_personality = ""
@@ -426,22 +539,50 @@ class Heartflow:
logger.error(traceback.format_exc())
return "(想法汇总时发生错误...)"
async def create_subheartflow(self, subheartflow_id: Any) -> Optional[SubHeartflow]:
# --- Add helper method to count subflows by state --- #
def count_subflows_by_state(self, target_state: 'ChatState') -> int:
"""Counts the number of subheartflows currently in the specified state."""
count = 0
# Use items() directly for read-only iteration if thread safety isn't a major concern here
# Or create snapshot if modification during iteration is possible elsewhere
items_snapshot = list(self._subheartflows.items())
for _, flow in items_snapshot:
# Check if flow still exists in the main dict in case it was removed concurrently
if flow.subheartflow_id in self._subheartflows and flow.chat_state.chat_status == target_state:
count += 1
return count
# --- End helper method --- #
async def create_subheartflow(self, subheartflow_id: Any) -> Optional['SubHeartflow']:
"""
获取或创建一个新的SubHeartflow实例。
(主要逻辑不变InterestChatting 现在在 SubHeartflow 内部创建)
创建本身不受限因为初始状态是ABSENT。
限制将在状态转换时检查。
"""
# --- 移除创建前的限制检查 --- #
# current_mai_state = self.current_state.mai_status
# normal_limit = current_mai_state.get_normal_chat_max_num()
# focused_limit = current_mai_state.get_focused_chat_max_num()
# total_limit = normal_limit + focused_limit
# current_active_count = 0
# items_snapshot = list(self._subheartflows.items())
# for _, flow in items_snapshot:
# if flow.chat_state.chat_status == ChatState.CHAT or flow.chat_state.chat_status == ChatState.FOCUSED:
# current_active_count += 1
# if current_active_count >= total_limit and total_limit > 0:
# stream_name = chat_manager.get_stream_name(subheartflow_id) or subheartflow_id
# logger.warning(f"[Heartflow Create] Skipped due to limit...")
# return None
# --- 结束移除 --- #
existing_subheartflow = self._subheartflows.get(subheartflow_id)
if existing_subheartflow:
# 如果已存在,确保其 last_active_time 更新 (如果需要的话)
# existing_subheartflow.last_active_time = time.time() # 移除,活跃时间由实际操作更新
# logger.debug(f"[Heartflow] 返回已存在的 subheartflow: {subheartflow_id}")
return existing_subheartflow
logger.info(f"[Heartflow] 尝试创建新的 subheartflow: {subheartflow_id}")
try:
# 创建 SubHeartflow它内部会创建 InterestChatting
subheartflow = SubHeartflow(subheartflow_id)
# --- Pass 'self' (Heartflow instance) to SubHeartflow constructor --- #
subheartflow = SubHeartflow(subheartflow_id, self)
# 创建并初始化观察对象
logger.debug(f"[Heartflow] 为 {subheartflow_id} 创建 observation")
@@ -464,7 +605,7 @@ class Heartflow:
logger.error(traceback.format_exc())
return None
def get_subheartflow(self, observe_chat_id: Any) -> Optional[SubHeartflow]:
def get_subheartflow(self, observe_chat_id: Any) -> Optional['SubHeartflow']:
"""获取指定ID的SubHeartflow实例"""
return self._subheartflows.get(observe_chat_id)
@@ -472,6 +613,182 @@ class Heartflow:
"""获取当前所有活跃的子心流的 ID 列表"""
return list(self._subheartflows.keys())
async def _stop_subheartflow(self, subheartflow_id: Any, reason: str):
"""停止并移除指定的子心流"""
if subheartflow_id in self._subheartflows:
subheartflow = self._subheartflows[subheartflow_id]
stream_name = chat_manager.get_stream_name(subheartflow_id) or subheartflow_id
logger.info(f"[Heartflow Limits] 停止子心流 {stream_name}. 原因: {reason}")
# 标记停止并取消任务
subheartflow.should_stop = True
task_to_cancel = subheartflow.task
if task_to_cancel and not task_to_cancel.done():
task_to_cancel.cancel()
logger.debug(f"[Heartflow Limits] 已取消子心流 {stream_name} 的后台任务")
# TODO: Ensure controller.stop_heartFC_chat is called if needed
from src.plugins.heartFC_chat.heartFC_controler import HeartFCController # Local import to avoid cycle
controller = HeartFCController.get_instance()
if controller and controller.is_heartFC_chat_active(subheartflow_id):
await controller.stop_heartFC_chat(subheartflow_id)
# 从字典移除
del self._subheartflows[subheartflow_id]
logger.debug(f"[Heartflow Limits] 已移除子心流: {stream_name}")
return True
return False
async def _enforce_subheartflow_limits(self, current_mai_state: MaiState):
"""根据当前的 MaiState 强制执行 SubHeartflow 数量限制"""
normal_limit = current_mai_state.get_normal_chat_max_num()
focused_limit = current_mai_state.get_focused_chat_max_num()
logger.debug(f"[Heartflow Limits] 执行限制检查。当前状态: {current_mai_state.value}, Normal上限: {normal_limit}, Focused上限: {focused_limit}")
# 分类并统计当前 subheartflows
normal_flows = []
focused_flows = []
other_flows = [] # e.g., ABSENT
# 创建快照以安全迭代
items_snapshot = list(self._subheartflows.items())
for flow_id, flow in items_snapshot:
# 确保 flow 实例仍然存在 (避免在迭代期间被其他任务移除)
if flow_id not in self._subheartflows:
continue
if flow.chat_state.chat_status == ChatState.CHAT:
normal_flows.append((flow_id, flow.last_active_time))
elif flow.chat_state.chat_status == ChatState.FOCUSED:
focused_flows.append((flow_id, flow.last_active_time))
else:
other_flows.append((flow_id, flow.last_active_time))
logger.debug(f"[Heartflow Limits] 当前计数 - Normal: {len(normal_flows)}, Focused: {len(focused_flows)}, Other: {len(other_flows)}")
stopped_count = 0
# 检查 Normal (CHAT) 限制
if len(normal_flows) > normal_limit:
excess_count = len(normal_flows) - normal_limit
logger.info(f"[Heartflow Limits] 检测到 Normal (CHAT) 状态超额 {excess_count} 个。上限: {normal_limit}")
# 按 last_active_time 升序排序 (最不活跃的在前)
normal_flows.sort(key=lambda item: item[1])
# 停止最不活跃的超额部分
for i in range(excess_count):
flow_id_to_stop = normal_flows[i][0]
if await self._stop_subheartflow(flow_id_to_stop, f"Normal (CHAT) 状态超出上限 ({normal_limit}),停止最不活跃的实例"):
stopped_count += 1
# 重新获取 focused_flows 列表,因为上面的停止操作可能已经改变了状态或移除了实例
focused_flows = []
items_snapshot_after_normal = list(self._subheartflows.items())
for flow_id, flow in items_snapshot_after_normal:
if flow_id not in self._subheartflows: continue # Double check
if flow.chat_state.chat_status == ChatState.FOCUSED:
focused_flows.append((flow_id, flow.last_active_time))
# 检查 Focused (FOCUSED) 限制
if len(focused_flows) > focused_limit:
excess_count = len(focused_flows) - focused_limit
logger.info(f"[Heartflow Limits] 检测到 Focused (FOCUSED) 状态超额 {excess_count} 个。上限: {focused_limit}")
# 按 last_active_time 升序排序
focused_flows.sort(key=lambda item: item[1])
# 停止最不活跃的超额部分
for i in range(excess_count):
flow_id_to_stop = focused_flows[i][0]
if await self._stop_subheartflow(flow_id_to_stop, f"Focused (FOCUSED) 状态超出上限 ({focused_limit}),停止最不活跃的实例"):
stopped_count += 1
if stopped_count > 0:
logger.info(f"[Heartflow Limits] 限制执行完成,共停止了 {stopped_count} 个子心流。当前总数: {len(self._subheartflows)}")
else:
logger.debug(f"[Heartflow Limits] 限制检查完成,无需停止子心流。当前总数: {len(self._subheartflows)}")
# --- 新增方法 --- #
async def _activate_random_subflows_to_chat(self, new_mai_state: MaiState):
"""当主状态从 OFFLINE 激活时,随机选择子心流进入 CHAT 状态"""
limit = new_mai_state.get_normal_chat_max_num()
if limit <= 0:
logger.info("[Heartflow Activate] 当前状态不允许 CHAT 子心流,跳过激活。")
return
# 使用快照进行迭代
all_flows_snapshot = list(self._subheartflows.values())
absent_flows = [flow for flow in all_flows_snapshot if flow.subheartflow_id in self._subheartflows and flow.chat_state.chat_status == ChatState.ABSENT]
num_to_activate = min(limit, len(absent_flows))
if num_to_activate <= 0:
logger.info(f"[Heartflow Activate] 没有处于 ABSENT 状态的子心流可供激活至 CHAT (上限: {limit})。")
return
logger.info(f"[Heartflow Activate] 将随机选择 {num_to_activate} 个 (上限 {limit}) ABSENT 子心流激活至 CHAT 状态。")
selected_flows = random.sample(absent_flows, num_to_activate)
activated_count = 0
for flow in selected_flows:
# 再次检查 flow 是否仍然存在且状态为 ABSENT (以防并发修改)
if flow.subheartflow_id in self._subheartflows and self._subheartflows[flow.subheartflow_id].chat_state.chat_status == ChatState.ABSENT:
stream_name = chat_manager.get_stream_name(flow.subheartflow_id) or flow.subheartflow_id
logger.debug(f"[Heartflow Activate] 正在将子心流 {stream_name} 状态设置为 CHAT。")
# 调用 set_chat_state它内部会处理日志记录
flow.set_chat_state(ChatState.CHAT)
activated_count += 1
else:
stream_name = chat_manager.get_stream_name(flow.subheartflow_id) or flow.subheartflow_id
logger.warning(f"[Heartflow Activate] 跳过激活子心流 {stream_name},因为它不再存在或状态已改变。")
logger.info(f"[Heartflow Activate] 完成激活,成功将 {activated_count} 个子心流设置为 CHAT 状态。")
async def _deactivate_all_subflows_on_offline(self):
"""当主状态变为 OFFLINE 时,停止所有子心流的活动并设置为 ABSENT"""
logger.info("[Heartflow Deactivate] 开始停用所有子心流...")
try:
from src.plugins.heartFC_chat.heartFC_controler import HeartFCController # 本地导入避免循环依赖
controller = HeartFCController.get_instance()
except ImportError:
logger.warning("[Heartflow Deactivate] 无法导入 HeartFCController将跳过停止 heartFC_chat。")
controller = None
except Exception as e:
logger.error(f"[Heartflow Deactivate] 获取 HeartFCController 实例时出错: {e}")
controller = None
# 使用 ID 快照进行迭代
flow_ids_snapshot = list(self._subheartflows.keys())
deactivated_count = 0
for flow_id in flow_ids_snapshot:
subflow = self._subheartflows.get(flow_id)
if not subflow:
continue # Subflow 可能在迭代过程中被清理
stream_name = chat_manager.get_stream_name(flow_id) or flow_id
try:
# 停止相关聊天进程 (例如 pf_chat)
if controller:
# TODO: 确认是否有 reason_chat 需要停止,并添加相应逻辑
if controller.is_heartFC_chat_active(flow_id):
logger.debug(f"[Heartflow Deactivate] 正在停止子心流 {stream_name} 的 heartFC_chat。")
await controller.stop_heartFC_chat(flow_id)
# 设置状态为 ABSENT
if subflow.chat_state.chat_status != ChatState.ABSENT:
logger.debug(f"[Heartflow Deactivate] 正在将子心流 {stream_name} 状态设置为 ABSENT。")
# 调用 set_chat_state它会处理日志和状态更新
subflow.set_chat_state(ChatState.ABSENT)
deactivated_count += 1
else:
# 如果已经是 ABSENT则无需再次设置但记录一下检查
logger.trace(f"[Heartflow Deactivate] 子心流 {stream_name} 已处于 ABSENT 状态。")
except Exception as e:
logger.error(f"[Heartflow Deactivate] 停用子心流 {stream_name} 时出错: {e}")
logger.error(traceback.format_exc())
logger.info(f"[Heartflow Deactivate] 完成停用,共将 {deactivated_count} 个子心流设置为 ABSENT 状态 (不包括已是 ABSENT 的)。")
# --- 结束新增方法 --- #
init_prompt()
# 创建一个全局的管理器实例

View File

@@ -78,8 +78,6 @@ class ChattingObservation(Observation):
return self.talking_message_str
async def observe(self):
# 查找新消息,最多获取 self.max_now_obs_len 条
print("2222222222222222221111111111111111开始观察")
new_messages_list = get_raw_msg_by_timestamp_with_chat(
chat_id=self.chat_id,
timestamp_start=self.last_observe_time,
@@ -87,8 +85,8 @@ class ChattingObservation(Observation):
limit=self.max_now_obs_len,
limit_mode="latest",
)
print(f"2222222222222222221111111111111111获取到新消息{len(new_messages_list)}")
if new_messages_list: # 检查列表是否为空
last_obs_time_mark = self.last_observe_time
self.last_observe_time = new_messages_list[-1]["time"]
self.talking_message.extend(new_messages_list)
@@ -98,7 +96,11 @@ class ChattingObservation(Observation):
oldest_messages = self.talking_message[:messages_to_remove_count]
self.talking_message = self.talking_message[messages_to_remove_count:] # 保留后半部分,即最新的
oldest_messages_str = await build_readable_messages(oldest_messages)
oldest_messages_str = await build_readable_messages(
messages=oldest_messages,
timestamp_mode="normal",
read_mark=last_obs_time_mark,
)
# 调用 LLM 总结主题
prompt = (
@@ -134,10 +136,7 @@ class ChattingObservation(Observation):
f"距离现在{time_diff}分钟前(聊天记录id:{mid_memory_item['id']}){mid_memory_item['theme']}\n"
)
self.mid_memory_info = mid_memory_str
# except Exception as e: # 将异常处理移至此处以覆盖整个总结过程
# logger.error(f"处理和总结旧消息时出错 for chat {self.chat_id}: {e}")
# traceback.print_exc() # 记录详细堆栈
# print(f"处理后self.talking_message{self.talking_message}")
self.talking_message_str = await build_readable_messages(messages=self.talking_message, timestamp_mode="normal")

View File

@@ -4,7 +4,7 @@ from src.plugins.moods.moods import MoodManager
from src.plugins.models.utils_model import LLMRequest
from src.config.config import global_config
import time
from typing import Optional, List, Dict
from typing import Optional, List, Dict, Callable, TYPE_CHECKING
import traceback
from src.plugins.chat.utils import parse_text_timestamps
import enum
@@ -14,8 +14,14 @@ import random
from src.plugins.person_info.relationship_manager import relationship_manager
from ..plugins.utils.prompt_builder import Prompt, global_prompt_manager
from src.plugins.chat.message import MessageRecv
from src.plugins.chat.chat_stream import chat_manager
import math
# Type hinting for circular dependency
if TYPE_CHECKING:
from .heartflow import Heartflow, MaiState # Import Heartflow for type hinting
from .sub_heartflow import ChatState # Keep ChatState here too?
# 定义常量 (从 interest.py 移动过来)
MAX_INTEREST = 15.0
@@ -68,9 +74,6 @@ class ChatStateInfo:
self.mood_manager = MoodManager()
self.mood = self.mood_manager.get_prompt()
def update_chat_state_info(self):
self.chat_state_info = self.mood_manager.get_current_mood()
base_reply_probability = 0.05
probability_increase_rate_per_second = 0.08
@@ -87,6 +90,7 @@ class InterestChatting:
increase_rate=probability_increase_rate_per_second,
decay_factor=global_config.probability_decay_factor_per_second,
max_probability=max_reply_probability,
state_change_callback: Optional[Callable[[ChatState], None]] = None
):
self.interest_level: float = 0.0
self.last_update_time: float = time.time()
@@ -101,6 +105,7 @@ class InterestChatting:
self.max_reply_probability: float = max_probability
self.current_reply_probability: float = 0.0
self.is_above_threshold: bool = False
self.state_change_callback = state_change_callback
self.interest_dict: Dict[str, tuple[MessageRecv, float, bool]] = {}
@@ -144,6 +149,7 @@ class InterestChatting:
return
currently_above = self.interest_level >= self.trigger_threshold
previous_is_above = self.is_above_threshold
if currently_above:
if not self.is_above_threshold:
@@ -158,6 +164,13 @@ class InterestChatting:
self.current_reply_probability = min(self.current_reply_probability, self.max_reply_probability)
else:
if previous_is_above:
if self.state_change_callback:
try:
self.state_change_callback(ChatState.ABSENT)
except Exception as e:
interest_logger.error(f"Error calling state_change_callback for ABSENT: {e}")
if 0 < self.probability_decay_factor < 1:
decay_multiplier = math.pow(self.probability_decay_factor, time_delta)
self.current_reply_probability *= decay_multiplier
@@ -216,14 +229,15 @@ class InterestChatting:
class SubHeartflow:
def __init__(self, subheartflow_id):
def __init__(self, subheartflow_id, parent_heartflow: 'Heartflow'):
self.subheartflow_id = subheartflow_id
self.parent_heartflow = parent_heartflow
self.current_mind = "你什么也没想"
self.past_mind = []
self.chat_state: ChatStateInfo = ChatStateInfo()
self.interest_chatting = InterestChatting()
self.interest_chatting = InterestChatting(state_change_callback=self.set_chat_state)
self.llm_model = LLMRequest(
model=global_config.llm_sub_heartflow,
@@ -234,33 +248,58 @@ class SubHeartflow:
self.main_heartflow_info = ""
self.last_active_time = time.time() # 添加最后激活时间
self.should_stop = False # 添加停止标志
self.task: Optional[asyncio.Task] = None # 添加 task 属性
self.last_active_time = time.time()
self.should_stop = False
self.task: Optional[asyncio.Task] = None
self.is_active = False
self.observations: List[ChattingObservation] = [] # 使用 List 类型提示
self.observations: List[ChattingObservation] = []
self.running_knowledges = []
self.bot_name = global_config.BOT_NICKNAME
logger.info(f"SubHeartflow {self.subheartflow_id} created with initial state: {self.chat_state.chat_status.value}")
def set_chat_state(self, new_state: 'ChatState'):
"""更新sub_heartflow的聊天状态"""
current_state = self.chat_state.chat_status
if current_state == new_state:
return # No change needed
log_prefix = f"[{chat_manager.get_stream_name(self.subheartflow_id) or self.subheartflow_id}]"
# --- Limit Check before entering CHAT state --- #
if new_state == ChatState.CHAT:
current_mai_state = self.parent_heartflow.current_state.mai_status
normal_limit = current_mai_state.get_normal_chat_max_num()
current_chat_count = self.parent_heartflow.count_subflows_by_state(ChatState.CHAT)
if current_chat_count >= normal_limit:
logger.debug(f"{log_prefix} 拒绝从 {current_state.value} 转换到 CHAT。原因CHAT 状态已达上限 ({normal_limit})。当前数量: {current_chat_count}")
return # Block the state transition
else:
logger.debug(f"{log_prefix} 允许从 {current_state.value} 转换到 CHAT (上限: {normal_limit}, 当前: {current_chat_count})" )
# 如果检查通过或目标状态不是CHAT则进行状态变更
self.chat_state.chat_status = new_state
# 状态变更时更新最后活跃时间
self.last_active_time = time.time()
logger.info(f"{log_prefix} 聊天状态从 {current_state.value} 变更为 {new_state.value}")
# TODO: 考虑从FOCUSED状态转出时是否需要停止PFChatting
# 这部分逻辑可能更适合放在Heartflow的_stop_subheartflow或HeartFCController的循环中处理
async def subheartflow_start_working(self):
while True:
# --- 调整后台任务逻辑 --- #
# 这个后台循环现在主要负责检查是否需要自我销毁
# 不再主动进行思考或状态更新,这些由 HeartFC_Chat 驱动
# 检查是否被主心流标记为停止
if self.should_stop:
logger.info(f"子心流 {self.subheartflow_id} 被标记为停止,正在退出后台任务...")
break # 退出循环以停止任务
break
await asyncio.sleep(global_config.sub_heart_flow_update_interval) # 定期检查销毁条件
await asyncio.sleep(global_config.sub_heart_flow_update_interval)
async def ensure_observed(self):
"""确保在思考前执行了观察"""
observation = self._get_primary_observation()
if observation:
try:
@@ -273,18 +312,14 @@ class SubHeartflow:
async def do_thinking_before_reply(
self,
extra_info: str,
obs_id: list[str] = None, # 修改 obs_id 类型为 list[str]
obs_id: list[str] = None,
):
# --- 在思考前确保观察已执行 --- #
# await self.ensure_observed()
self.last_active_time = time.time() # 更新最后激活时间戳
self.last_active_time = time.time()
current_thinking_info = self.current_mind
mood_info = self.chat_state.mood
observation = self._get_primary_observation()
# --- 获取观察信息 --- #
chat_observe_info = ""
if obs_id:
try:
@@ -294,12 +329,11 @@ class SubHeartflow:
logger.error(
f"[{self.subheartflow_id}] Error getting observe info with IDs {obs_id}: {e}. Falling back."
)
chat_observe_info = observation.get_observe_info() # 出错时回退到默认观察
chat_observe_info = observation.get_observe_info()
else:
chat_observe_info = observation.get_observe_info()
logger.debug(f"[{self.subheartflow_id}] Using default observation info.")
# logger.debug(f"[{self.subheartflow_id}] Using default observation info.")
# --- 构建 Prompt (基本逻辑不变) --- #
extra_info_prompt = ""
if extra_info:
for tool_name, tool_data in extra_info.items():
@@ -307,28 +341,25 @@ class SubHeartflow:
for item in tool_data:
extra_info_prompt += f"- {item['name']}: {item['content']}\n"
else:
extra_info_prompt = "无工具信息。\n" # 提供默认值
extra_info_prompt = "无工具信息。\n"
individuality = Individuality.get_instance()
prompt_personality = f"你的名字是{self.bot_name},你"
prompt_personality += individuality.personality.personality_core
# 添加随机性格侧面
if individuality.personality.personality_sides:
random_side = random.choice(individuality.personality.personality_sides)
prompt_personality += f"{random_side}"
# 添加随机身份细节
if individuality.identity.identity_detail:
random_detail = random.choice(individuality.identity.identity_detail)
prompt_personality += f"{random_detail}"
time_now = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime())
# 创建局部Random对象避免影响全局随机状态
local_random = random.Random()
current_minute = int(time.strftime("%M"))
local_random.seed(current_minute) # 用分钟作为种子确保每分钟内选择一致
local_random.seed(current_minute)
hf_options = [
("继续生成你在这个聊天中的想法,在原来想法的基础上继续思考", 0.7),
@@ -343,7 +374,6 @@ class SubHeartflow:
prompt = (await global_prompt_manager.get_prompt_async("sub_heartflow_prompt_before")).format(
extra_info=extra_info_prompt,
# relation_prompt_all=relation_prompt_all,
prompt_personality=prompt_personality,
bot_name=self.bot_name,
current_thinking_info=current_thinking_info,
@@ -351,8 +381,6 @@ class SubHeartflow:
chat_observe_info=chat_observe_info,
mood_info=mood_info,
hf_do_next=hf_do_next,
# sender_name=sender_name_sign,
# message_txt=message_txt,
)
prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt)
@@ -365,18 +393,15 @@ class SubHeartflow:
logger.debug(f"[{self.subheartflow_id}] 心流思考结果:\n{response}\n")
if not response: # 如果 LLM 返回空,给一个默认想法
if not response:
response = "(不知道该想些什么...)"
logger.warning(f"[{self.subheartflow_id}] LLM 返回空结果,思考失败。")
except Exception as e:
logger.error(f"[{self.subheartflow_id}] 内心独白获取失败: {e}")
response = "(思考时发生错误...)" # 错误时的默认想法
response = "(思考时发生错误...)"
self.update_current_mind(response)
# self.current_mind 已经在 update_current_mind 中更新
# logger.info(f"[{self.subheartflow_id}] 思考前脑内状态:{self.current_mind}")
return self.current_mind, self.past_mind
def update_current_mind(self, response):
@@ -384,55 +409,41 @@ class SubHeartflow:
self.current_mind = response
def add_observation(self, observation: Observation):
"""添加一个新的observation对象到列表中如果已存在相同id的observation则不添加"""
# 查找是否存在相同id的observation
for existing_obs in self.observations:
if existing_obs.observe_id == observation.observe_id:
# 如果找到相同id的observation直接返回
return
# 如果没有找到相同id的observation则添加新的
self.observations.append(observation)
def remove_observation(self, observation: Observation):
"""从列表中移除一个observation对象"""
if observation in self.observations:
self.observations.remove(observation)
def get_all_observations(self) -> list[Observation]:
"""获取所有observation对象"""
return self.observations
def clear_observations(self):
"""清空所有observation对象"""
self.observations.clear()
def _get_primary_observation(self) -> Optional[ChattingObservation]:
"""获取主要的通常是第一个ChattingObservation实例"""
if self.observations and isinstance(self.observations[0], ChattingObservation):
return self.observations[0]
logger.warning(f"SubHeartflow {self.subheartflow_id} 没有找到有效的 ChattingObservation")
return None
def get_interest_state(self) -> dict:
"""获取当前兴趣状态"""
return self.interest_chatting.get_state()
def get_interest_level(self) -> float:
"""获取当前兴趣等级"""
return self.interest_chatting.get_interest()
def should_evaluate_reply(self) -> bool:
"""判断是否应该评估回复"""
return self.interest_chatting.should_evaluate_reply()
def add_interest_dict_entry(self, message: MessageRecv, interest_value: float, is_mentioned: bool):
"""添加兴趣字典条目"""
self.interest_chatting.add_interest_dict(message, interest_value, is_mentioned)
def get_interest_dict(self) -> Dict[str, tuple[MessageRecv, float, bool]]:
"""获取兴趣字典"""
return self.interest_chatting.interest_dict
init_prompt()
# subheartflow = SubHeartflow()