mirror of https://github.com/Mai-with-u/MaiBot.git
118 lines
6.1 KiB
Python
118 lines
6.1 KiB
Python
from typing import List, Dict, Any
|
||
|
||
from src.plugins.PFC.chat_observer import ChatObserver
|
||
from src.common.logger_manager import get_logger
|
||
from src.plugins.models.utils_model import LLMRequest
|
||
from src.plugins.moods.moods import MoodManager # MoodManager 本身是单例
|
||
from src.plugins.utils.chat_message_builder import build_readable_messages
|
||
from src.plugins.PFC.observation_info import ObservationInfo
|
||
from src.plugins.PFC.conversation_info import ConversationInfo
|
||
from src.config.config import global_config # 导入全局配置
|
||
|
||
logger = get_logger("pfc_emotion")
|
||
|
||
|
||
class PfcEmotionUpdater:
|
||
def __init__(self, private_name: str, bot_name: str):
|
||
"""
|
||
初始化情绪更新器。
|
||
"""
|
||
self.private_name = private_name
|
||
self.bot_name = bot_name
|
||
self.mood_mng = MoodManager.get_instance() # 获取 MoodManager 单例
|
||
|
||
# LLM 实例 (根据 global_config.llm_summary 配置)
|
||
llm_config_summary = getattr(global_config, "llm_summary", None)
|
||
if llm_config_summary and isinstance(llm_config_summary, dict):
|
||
logger.debug(f"[私聊][{self.private_name}] 使用 llm_summary 配置初始化情绪判断LLM。")
|
||
self.llm = LLMRequest(
|
||
model=llm_config_summary,
|
||
temperature=llm_config_summary.get(
|
||
"temperature", 0.5
|
||
), # temperature 来自其自身配置或默认0.7，这里用0.5
|
||
max_tokens=llm_config_summary.get("max_tokens", 256), # 情绪词输出不需要很多token
|
||
request_type="pfc_emotion_evaluation",
|
||
)
|
||
else:
|
||
logger.error(f"[私聊][{self.private_name}] 未找到 llm_summary 配置或配置无效！情绪判断功能将受限。")
|
||
self.llm = None # LLM 未初始化
|
||
|
||
self.EMOTION_UPDATE_INTENSITY = getattr(global_config, "pfc_emotion_update_intensity", 0.6)
|
||
self.EMOTION_HISTORY_COUNT = getattr(global_config, "pfc_emotion_history_count", 5)
|
||
|
||
async def update_emotion_based_on_context(
|
||
self,
|
||
conversation_info: ConversationInfo,
|
||
observation_info: ObservationInfo,
|
||
chat_observer_for_history: ChatObserver, # ChatObserver 实例
|
||
event_description: str,
|
||
) -> None:
|
||
if not self.llm:
|
||
logger.error(f"[私聊][{self.private_name}] LLM未初始化，无法进行情绪更新。")
|
||
# 即使LLM失败，也应该更新conversation_info中的情绪文本为MoodManager的当前状态
|
||
if conversation_info and self.mood_mng:
|
||
conversation_info.current_emotion_text = self.mood_mng.get_prompt()
|
||
return
|
||
|
||
if not self.mood_mng or not conversation_info or not observation_info:
|
||
logger.debug(f"[私聊][{self.private_name}] 情绪更新：缺少必要管理器或信息。")
|
||
return
|
||
|
||
recent_messages_for_emotion: List[Dict[str, Any]] = []
|
||
if chat_observer_for_history:
|
||
recent_messages_for_emotion = chat_observer_for_history.get_cached_messages(
|
||
limit=self.EMOTION_HISTORY_COUNT
|
||
)
|
||
elif observation_info.chat_history:
|
||
recent_messages_for_emotion = observation_info.chat_history[-self.EMOTION_HISTORY_COUNT :]
|
||
|
||
readable_recent_history = await build_readable_messages(
|
||
recent_messages_for_emotion, replace_bot_name=True, merge_messages=True, timestamp_mode="none"
|
||
)
|
||
|
||
current_mood_text_from_manager = self.mood_mng.current_mood.text # 从 MoodManager 获取当前情绪文本
|
||
sender_name_for_prompt = self.private_name
|
||
relationship_text_for_prompt = getattr(
|
||
conversation_info, "relationship_text", "关系一般。"
|
||
) # 从 ConversationInfo 获取关系文本
|
||
|
||
emotion_prompt = f"""你是{self.bot_name}。你现在的心情是【{current_mood_text_from_manager}】。
|
||
你正在和用户【{sender_name_for_prompt}】私聊，你们的关系是：【{relationship_text_for_prompt}】。
|
||
最近发生的事件是：【{event_description}】
|
||
最近的对话摘要：
|
||
---
|
||
{readable_recent_history}
|
||
---
|
||
基于以上所有信息，你认为你现在最主要的情绪是什么？请从以下情绪词中选择一个（必须是列表中的一个）：
|
||
[开心, 害羞, 愤怒, 恐惧, 悲伤, 厌恶, 惊讶, 困惑, 平静]
|
||
请只输出一个最符合的情绪词。例如： 开心
|
||
如果难以判断或当前情绪依然合适，请输出： 无变化
|
||
"""
|
||
try:
|
||
logger.debug(f"[私聊][{self.private_name}] 情绪判断Prompt:\n{emotion_prompt}")
|
||
content, _ = await self.llm.generate_response_async(emotion_prompt)
|
||
detected_emotion_word = content.strip().replace('"', "").replace("'", "")
|
||
logger.debug(f"[私聊][{self.private_name}] 情绪判断LLM原始返回: '{detected_emotion_word}'")
|
||
|
||
if (
|
||
detected_emotion_word
|
||
and detected_emotion_word != "无变化"
|
||
and detected_emotion_word in self.mood_mng.emotion_map
|
||
):
|
||
self.mood_mng.update_mood_from_emotion(detected_emotion_word, intensity=self.EMOTION_UPDATE_INTENSITY)
|
||
logger.debug(
|
||
f"[私聊][{self.private_name}] 基于事件 '{event_description}'，情绪已更新为倾向于 '{detected_emotion_word}'。当前心情: {self.mood_mng.current_mood.text}"
|
||
)
|
||
elif detected_emotion_word == "无变化":
|
||
logger.debug(f"[私聊][{self.private_name}] 基于事件 '{event_description}'，LLM判断情绪无显著变化。")
|
||
else:
|
||
logger.warning(
|
||
f"[私聊][{self.private_name}] LLM返回了未知的情绪词 '{detected_emotion_word}' 或未返回有效词，情绪未主动更新。"
|
||
)
|
||
except Exception as e:
|
||
logger.error(f"[私聊][{self.private_name}] 情绪判断LLM调用或处理失败: {e}")
|
||
|
||
# 无论LLM判断如何，都更新conversation_info中的情绪文本以供Prompt使用
|
||
if conversation_info and self.mood_mng: # 确保conversation_info有效
|
||
conversation_info.current_emotion_text = self.mood_mng.get_prompt()
|