Losita/mai-bot
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

better：优化hfc逻辑

Browse Source
This commit is contained in:
SengokuCola
2025-04-19 18:48:59 +08:00
parent b1dc34f7b1
commit c9ab9d4935
11 changed files with 227 additions and 475 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
src/plugins/chat_module/heartFC_chat/pf_chatting.py
View File

@@ -13,6 +13,8 @@ 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.plugins.chat.utils import parse_text_timestamps
from src.plugins.person_info.relationship_manager import relationship_manager
# 定义日志配置 (使用 loguru 格式)
interest_log_config = LogConfig(
@@ -102,8 +104,8 @@ class PFChatting:
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.
懒初始化以使用提供的标识符解析chat_stream和sub_hf。
确保实例已准备好处理触发器。
"""
async with self._init_lock:
if self._initialized:
@@ -171,7 +173,7 @@ class PFChatting:
# Start the loop if it wasn't active and timer is positive
if not self._loop_active and self._loop_timer > 0:
logger.info(f"{log_prefix} 麦麦有兴趣！开始聊天")
# logger.info(f"{log_prefix} 麦麦有兴趣！开始聊天")
self._loop_active = True
if self._loop_task and not self._loop_task.done():
logger.warning(f"{log_prefix} 发现意外的循环任务正在进行。取消它。")
@@ -363,9 +365,17 @@ class PFChatting:
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]}
返回:
dict: 包含决策和上下文的字典，结构如下:
{
'action': str, # 执行动作 (不回复/文字回复/表情包)
'reasoning': str, # 决策理由
'emoji_query': str, # 表情包查询词
'current_mind': str, # 当前心理状态
'send_emoji_from_tools': str, # 工具推荐的表情包
'observed_messages': List[dict] # 观察到的消息列表
}
"""
log_prefix = self._get_log_prefix()
observed_messages: List[dict] = []
@@ -376,14 +386,15 @@ class PFChatting:
# --- 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"{log_prefix}[Planner] 调用 observation.observe()...")
observation = self.sub_hf._get_primary_observation() # Call only once
if observation: # Now check if the result is truthy
# logger.debug(f"{log_prefix}[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"{log_prefix}[Planner] 获取到 {len(observed_messages)} 条观察消息。")
logger.debug(f"{log_prefix}[Planner] 观察获取到 {len(observed_messages)} 条消息。")
else:
logger.warning(f"{log_prefix}[Planner] 无法获取 SubHeartflow 或 Observation 来获取消息。")
logger.warning(f"{log_prefix}[Planner] 无法获取 Observation。")
except Exception as e:
logger.error(f"{log_prefix}[Planner] 获取观察信息时出错: {e}")
logger.error(traceback.format_exc())
@@ -396,49 +407,30 @@ class PFChatting:
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"{log_prefix}[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"{log_prefix}[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"{log_prefix}[Planner] Skipping tool use because SubHeartflow is not available.")
observation_context_text = " ".join(context_texts)
# logger.debug(f"{log_prefix}[Planner] Context for tools: {observation_context_text[:100]}...")
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"{log_prefix}[Planner] 规划前工具结果: {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
]
except Exception as e_tool:
logger.error(f"{log_prefix}[Planner] Tool use failed: {e_tool}")
# Continue even if tool use fails
logger.error(f"{log_prefix}[Planner] 规划前工具使用失败: {e_tool}")
# --- 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"{log_prefix}[Planner] SubHeartflow thought: {current_mind}")
else:
logger.warning(f"{log_prefix}[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"{log_prefix}[Planner] SubHeartflow thinking failed: {e_shf}")
logger.error(traceback.format_exc())
current_mind = "[心流思考出错]"
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,
)
# --- 使用 LLM 进行决策 ---
action = "no_reply" # Default action
@@ -448,8 +440,8 @@ class PFChatting:
try:
# 构建提示 (Now includes current_mind)
prompt = self._build_planner_prompt(observed_messages, current_mind)
logger.debug(f"{log_prefix}[Planner] Prompt: {prompt}")
prompt = await self._build_planner_prompt(observed_messages, current_mind)
logger.debug(f"{log_prefix}[Planner] 规划器 Prompt: {prompt}")
# 准备 LLM 请求 Payload
payload = {
@@ -459,7 +451,6 @@ class PFChatting:
"tool_choice": {"type": "function", "function": {"name": "decide_reply_action"}}, # 强制调用此工具
}
logger.debug(f"{log_prefix}[Planner] 发送 Planner LLM 请求...")
# 调用 LLM
response = await self.planner_llm._execute_request(
endpoint="/chat/completions", payload=payload, prompt=prompt
@@ -717,7 +708,7 @@ class PFChatting:
logger.info(f"{self._get_log_prefix()} PFChatting shutdown complete.")
def _build_planner_prompt(self, observed_messages: List[dict], current_mind: Optional[str]) -> str:
async 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 = f"你的名字是 {global_config.BOT_NICKNAME}。是一个女大二学生，正在QQ聊天，正在决定是否以及如何回应当前的聊天。\n"
@@ -749,6 +740,10 @@ class PFChatting:
prompt += "4. 如果你已经回复过消息，也没有人又回复你，选择'no_reply'。"
prompt += "必须调用 'decide_reply_action' 工具并提供 'action' 和 'reasoning'。"
prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt)
prompt = parse_text_timestamps(prompt, mode="lite")
return prompt
# --- 回复器 (Replier) 的定义 --- #
@@ -771,7 +766,7 @@ class PFChatting:
# --- Tool Use and SubHF Thinking are now in _planner ---
# --- Generate Response with LLM ---
logger.debug(f"{log_prefix}[Replier-{thinking_id}] Calling LLM to generate response...")
# logger.debug(f"{log_prefix}[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,
@@ -785,7 +780,7 @@ class PFChatting:
return None # Indicate failure
# --- 准备并返回结果 ---
logger.info(f"{log_prefix}[Replier-{thinking_id}] 成功生成了回复集: {' '.join(response_set)[:50]}...")
logger.info(f"{log_prefix}[Replier-{thinking_id}] 成功生成了回复集: {' '.join(response_set)[:100]}...")
return {
"response_set": response_set,
"send_emoji": send_emoji, # Pass through the emoji determined earlier (usually by tools)