better：优化planner，提及时消耗更少，连续no_reply时降低敏感度
This commit is contained in:
SengokuCola
@@ -111,6 +111,9 @@ class HeartFChatting:
|
||||
self.question_probability_multiplier = 1
|
||||
self.questioned = False
|
||||
|
||||
# 跟踪连续 no_reply 次数，用于动态调整阈值
|
||||
self.consecutive_no_reply_count = 0
|
||||
|
||||
# 聊天内容概括器
|
||||
self.chat_history_summarizer = ChatHistorySummarizer(chat_id=self.stream_id)
|
||||
|
||||
@@ -192,43 +195,21 @@ class HeartFChatting:
|
||||
filter_command=True,
|
||||
)
|
||||
|
||||
question_probability = 0
|
||||
if time.time() - self.last_active_time > 7200:
|
||||
question_probability = 0.0003
|
||||
elif time.time() - self.last_active_time > 3600:
|
||||
question_probability = 0.0001
|
||||
|
||||
|
||||
|
||||
# 根据连续 no_reply 次数动态调整阈值
|
||||
# 3次 no_reply 时，阈值调高到 1.5（50%概率为1，50%概率为2）
|
||||
# 5次 no_reply 时，提高到 2（大于等于两条消息的阈值）
|
||||
if self.consecutive_no_reply_count >= 5:
|
||||
threshold = 2
|
||||
elif self.consecutive_no_reply_count >= 3:
|
||||
# 1.5 的含义：50%概率为1，50%概率为2
|
||||
threshold = 2 if random.random() < 0.5 else 1
|
||||
else:
|
||||
question_probability = 0.00003
|
||||
|
||||
question_probability = question_probability * global_config.chat.get_auto_chat_value(self.stream_id) * self.question_probability_multiplier
|
||||
threshold = 1
|
||||
|
||||
#暂时禁用
|
||||
|
||||
# print(f"{self.log_prefix} questioned: {self.questioned},len: {len(global_conflict_tracker.get_questions_by_chat_id(self.stream_id))}")
|
||||
# if question_probability > 0 and not self.questioned and len(global_conflict_tracker.get_questions_by_chat_id(self.stream_id)) == 0: #长久没有回复，可以试试主动发言，提问概率随着时间增加
|
||||
# # logger.info(f"{self.log_prefix} 长久没有回复，可以试试主动发言，概率: {question_probability}")
|
||||
# if random.random() < question_probability: # 30%概率主动发言
|
||||
# try:
|
||||
# self.questioned = True
|
||||
# self.last_active_time = time.time()
|
||||
# # print(f"{self.log_prefix} 长久没有回复，可以试试主动发言，开始生成问题")
|
||||
# logger.info(f"{self.log_prefix} 长久没有回复，可以试试主动发言，开始生成问题")
|
||||
# cycle_timers, thinking_id = self.start_cycle()
|
||||
# question_maker = QuestionMaker(self.stream_id)
|
||||
# question, context,conflict_context = await question_maker.make_question()
|
||||
# if question:
|
||||
# logger.info(f"{self.log_prefix} 问题: {question}")
|
||||
# await global_conflict_tracker.track_conflict(question, conflict_context, True, self.stream_id)
|
||||
# await self._lift_question_reply(question,context,thinking_id)
|
||||
# else:
|
||||
# logger.info(f"{self.log_prefix} 无问题")
|
||||
# # self.end_cycle(cycle_timers, thinking_id)
|
||||
# except Exception as e:
|
||||
# logger.error(f"{self.log_prefix} 主动提问失败: {e}")
|
||||
# print(traceback.format_exc())
|
||||
|
||||
|
||||
if len(recent_messages_list) >= 1:
|
||||
if len(recent_messages_list) >= threshold:
|
||||
# for message in recent_messages_list:
|
||||
# print(message.processed_plain_text)
|
||||
# !处理no_reply_until_call逻辑
|
||||
@@ -332,14 +313,15 @@ class HeartFChatting:
|
||||
available_actions: Dict[str, ActionInfo],
|
||||
cycle_timers: Dict[str, float],
|
||||
) -> List[ActionPlannerInfo]:
|
||||
"""执行planner，但不包含reply动作（用于并行执行场景）"""
|
||||
"""执行planner，但不包含reply动作（用于并行执行场景，提及时使用简化版提示词）"""
|
||||
try:
|
||||
with Timer("规划器", cycle_timers):
|
||||
action_to_use_info = await self.action_planner.plan(
|
||||
loop_start_time=self.last_read_time,
|
||||
available_actions=available_actions,
|
||||
is_mentioned=True, # 标记为提及时，使用简化版提示词
|
||||
)
|
||||
# 过滤掉reply动作
|
||||
# 过滤掉reply动作（虽然提及时不应该有reply，但为了安全还是过滤一下）
|
||||
return [action for action in action_to_use_info if action.action_type != "reply"]
|
||||
except Exception as e:
|
||||
logger.error(f"{self.log_prefix} Planner执行失败: {e}")
|
||||
@@ -356,6 +338,8 @@ class HeartFChatting:
|
||||
"""当被提及时，独立生成回复的任务"""
|
||||
try:
|
||||
self.questioned = False
|
||||
# 重置连续 no_reply 计数
|
||||
self.consecutive_no_reply_count = 0
|
||||
reason = "有人提到了你，进行回复"
|
||||
|
||||
await database_api.store_action_info(
|
||||
@@ -675,76 +659,6 @@ class HeartFChatting:
|
||||
traceback.print_exc()
|
||||
return False, ""
|
||||
|
||||
async def _lift_question_reply(self, question: str, question_context: str, thinking_id: str):
|
||||
reason = f"在聊天中：\n{question_context}\n你对问题\"{question}\"感到好奇，想要和群友讨论"
|
||||
new_msg = get_raw_msg_before_timestamp_with_chat(
|
||||
chat_id=self.stream_id,
|
||||
timestamp=time.time(),
|
||||
limit=1,
|
||||
)
|
||||
|
||||
reply_action_info = ActionPlannerInfo(
|
||||
action_type="reply",
|
||||
reasoning= "",
|
||||
action_data={},
|
||||
action_message=new_msg[0],
|
||||
available_actions=None,
|
||||
loop_start_time=time.time(),
|
||||
action_reasoning=reason)
|
||||
self.action_planner.add_plan_log(reasoning=f"你对问题\"{question}\"感到好奇，想要和群友讨论", actions=[reply_action_info])
|
||||
|
||||
success, llm_response = await generator_api.rewrite_reply(
|
||||
chat_stream=self.chat_stream,
|
||||
reply_data={
|
||||
"raw_reply": f"我对这个问题感到好奇：{question}",
|
||||
"reason": reason,
|
||||
},
|
||||
)
|
||||
|
||||
if not success or not llm_response or not llm_response.reply_set:
|
||||
logger.info("主动提问发言失败")
|
||||
self.action_planner.add_plan_excute_log(result="主动回复生成失败")
|
||||
return {"action_type": "reply", "success": False, "result": "主动回复生成失败", "loop_info": None}
|
||||
|
||||
if success:
|
||||
for reply_seg in llm_response.reply_set.reply_data:
|
||||
send_data = reply_seg.content
|
||||
await send_api.text_to_stream(
|
||||
text=send_data,
|
||||
stream_id=self.stream_id,
|
||||
)
|
||||
|
||||
await database_api.store_action_info(
|
||||
chat_stream=self.chat_stream,
|
||||
action_build_into_prompt=False,
|
||||
action_prompt_display=reason,
|
||||
action_done=True,
|
||||
thinking_id=thinking_id,
|
||||
action_data={"reply_text": llm_response.reply_set.reply_data[0].content},
|
||||
action_name="reply",
|
||||
)
|
||||
|
||||
# 构建循环信息
|
||||
loop_info: Dict[str, Any] = {
|
||||
"loop_plan_info": {
|
||||
"action_result": [reply_action_info],
|
||||
},
|
||||
"loop_action_info": {
|
||||
"action_taken": True,
|
||||
"reply_text": llm_response.reply_set.reply_data[0].content,
|
||||
"command": "",
|
||||
"taken_time": time.time(),
|
||||
},
|
||||
}
|
||||
self.last_active_time = time.time()
|
||||
self.action_planner.add_plan_excute_log(result=f"你提问：{question}")
|
||||
|
||||
return {
|
||||
"action_type": "reply",
|
||||
"success": True,
|
||||
"result": f"你提问：{question}",
|
||||
"loop_info": loop_info,
|
||||
}
|
||||
|
||||
|
||||
async def _send_response(
|
||||
@@ -808,6 +722,9 @@ class HeartFChatting:
|
||||
reason = action_planner_info.reasoning or "选择不回复"
|
||||
# logger.info(f"{self.log_prefix} 选择不回复，原因: {reason}")
|
||||
|
||||
# 增加连续 no_reply 计数
|
||||
self.consecutive_no_reply_count += 1
|
||||
|
||||
await database_api.store_action_info(
|
||||
chat_stream=self.chat_stream,
|
||||
action_build_into_prompt=False,
|
||||
@@ -827,6 +744,8 @@ class HeartFChatting:
|
||||
logger.info(f"{self.log_prefix} 保持沉默，直到有人直接叫的名字")
|
||||
reason = action_planner_info.reasoning or "选择不回复"
|
||||
|
||||
# 增加连续 no_reply 计数
|
||||
self.consecutive_no_reply_count += 1
|
||||
self.no_reply_until_call = True
|
||||
await database_api.store_action_info(
|
||||
chat_stream=self.chat_stream,
|
||||
@@ -844,6 +763,8 @@ class HeartFChatting:
|
||||
# 直接当场执行reply逻辑
|
||||
self.questioned = False
|
||||
# 刷新主动发言状态
|
||||
# 重置连续 no_reply 计数
|
||||
self.consecutive_no_reply_count = 0
|
||||
|
||||
reason = action_planner_info.reasoning or "选择回复"
|
||||
await database_api.store_action_info(
|
||||
|
||||
Reference in New Issue
Block a user