Merge remote-tracking branch 'upstream/debug' into debug

config/bot_config_template.toml

@@ -30,6 +30,7 @@ enable_pic_translate = false
 model_r1_probability = 0.8 # 麦麦回答时选择R1模型的概率
 model_v3_probability = 0.1 # 麦麦回答时选择V3模型的概率
 model_r1_distill_probability = 0.1 # 麦麦回答时选择R1蒸馏模型的概率
+max_response_length = 1024 # 麦麦回答的最大token数
 
 [memory]
 build_memory_interval = 300 # 记忆构建间隔 单位秒

src/plugins/chat/bot.py

@@ -58,6 +58,7 @@ class ChatBot:
             plain_text=event.get_plaintext(),
             reply_message=event.reply,
         )
+        await message.initialize()
 
         # 过滤词
         for word in global_config.ban_words:

src/plugins/chat/config.py

@@ -35,6 +35,8 @@ class BotConfig:
 
     ban_words = set()
 
+    max_response_length: int = 1024  # 最大回复长度
+    
     # 模型配置
     llm_reasoning: Dict[str, str] = field(default_factory=lambda: {})
     llm_reasoning_minor: Dict[str, str] = field(default_factory=lambda: {})
@@ -117,6 +119,7 @@ class BotConfig:
                 config.MODEL_R1_DISTILL_PROBABILITY = response_config.get("model_r1_distill_probability", config.MODEL_R1_DISTILL_PROBABILITY)
                 config.API_USING = response_config.get("api_using", config.API_USING)
                 config.API_PAID = response_config.get("api_paid", config.API_PAID)
+                config.max_response_length = response_config.get("max_response_length", config.max_response_length)
                 
             # 加载模型配置
             if "model" in toml_dict:

src/plugins/chat/cq_code.py

@@ -10,11 +10,11 @@ from nonebot.adapters.onebot.v11 import Bot
 from .config import global_config
 import time
 import asyncio
-from .utils_image import storage_image,storage_emoji
+from .utils_image import storage_image, storage_emoji
 from .utils_user import get_user_nickname
 from ..models.utils_model import LLM_request
-#解析各种CQ码
+# 解析各种CQ码
-#包含CQ码类
+# 包含CQ码类
 import urllib3
 from urllib3.util import create_urllib3_context
 from nonebot import get_driver
@@ -27,6 +27,7 @@ ctx = create_urllib3_context()
 ctx.load_default_certs()
 ctx.set_ciphers("AES128-GCM-SHA256")
 
+
 class TencentSSLAdapter(requests.adapters.HTTPAdapter):
     def __init__(self, ssl_context=None, **kwargs):
         self.ssl_context = ssl_context
@@ -37,6 +38,7 @@ class TencentSSLAdapter(requests.adapters.HTTPAdapter):
             num_pools=connections, maxsize=maxsize,
             block=block, ssl_context=self.ssl_context)
 
+
 @dataclass
 class CQCode:
     """
@@ -64,15 +66,15 @@ class CQCode:
         """初始化LLM实例"""
         self._llm = LLM_request(model=global_config.vlm, temperature=0.4, max_tokens=300)
 
-    def translate(self):
+    async def translate(self):
         """根据CQ码类型进行相应的翻译处理"""
         if self.type == 'text':
             self.translated_plain_text = self.params.get('text', '')
         elif self.type == 'image':
             if self.params.get('sub_type') == '0':
-                self.translated_plain_text = self.translate_image()
+                self.translated_plain_text = await self.translate_image()
             else:
-                self.translated_plain_text = self.translate_emoji()
+                self.translated_plain_text = await self.translate_emoji()
         elif self.type == 'at':
             user_nickname = get_user_nickname(self.params.get('qq', ''))
             if user_nickname:
@@ -80,13 +82,13 @@ class CQCode:
             else:
                 self.translated_plain_text = f"@某人"
         elif self.type == 'reply':
-            self.translated_plain_text = self.translate_reply()
+            self.translated_plain_text = await self.translate_reply()
         elif self.type == 'face':
             face_id = self.params.get('id', '')
             # self.translated_plain_text = f"[表情{face_id}]"
             self.translated_plain_text = f"[表情]"
         elif self.type == 'forward':
-            self.translated_plain_text = self.translate_forward()
+            self.translated_plain_text = await self.translate_forward()
         else:
             self.translated_plain_text = f"[{self.type}]"
 
@@ -158,7 +160,7 @@ class CQCode:
 
         return None
 
-    def translate_emoji(self) -> str:
+    async def translate_emoji(self) -> str:
         """处理表情包类型的CQ码"""
         if 'url' not in self.params:
             return '[表情包]'
@@ -167,45 +169,46 @@ class CQCode:
             # 将 base64 字符串转换为字节类型
             image_bytes = base64.b64decode(base64_str)
             storage_emoji(image_bytes)
-            return self.get_emoji_description(base64_str)
+            return await self.get_emoji_description(base64_str)
         else:
             return '[表情包]'
 
-    
+    async def translate_image(self) -> str:
-    def translate_image(self) -> str:
         """处理图片类型的CQ码，区分普通图片和表情包"""
-        #没有url，直接返回默认文本
+        # 没有url，直接返回默认文本
         if 'url' not in self.params:
             return '[图片]'
         base64_str = self.get_img()
         if base64_str:
             image_bytes = base64.b64decode(base64_str)
             storage_image(image_bytes)
-            return self.get_image_description(base64_str)
+            return await self.get_image_description(base64_str)
         else:
             return '[图片]'
 
-    def get_emoji_description(self, image_base64: str) -> str:
+    async def get_emoji_description(self, image_base64: str) -> str:
         """调用AI接口获取表情包描述"""
         try:
             prompt = "这是一个表情包，请用简短的中文描述这个表情包传达的情感和含义。最多20个字。"
-            description, _ = self._llm.generate_response_for_image_sync(prompt, image_base64)
+            # description, _ = self._llm.generate_response_for_image_sync(prompt, image_base64)
+            description, _ = await self._llm.generate_response_for_image(prompt, image_base64)
             return f"[表情包：{description}]"
         except Exception as e:
             print(f"\033[1;31m[错误]\033[0m AI接口调用失败: {str(e)}")
             return "[表情包]"
 
-    def get_image_description(self, image_base64: str) -> str:
+    async def get_image_description(self, image_base64: str) -> str:
         """调用AI接口获取普通图片描述"""
         try:
             prompt = "请用中文描述这张图片的内容。如果有文字，请把文字都描述出来。并尝试猜测这个图片的含义。最多200个字。"
-            description, _ = self._llm.generate_response_for_image_sync(prompt, image_base64)
+            # description, _ = self._llm.generate_response_for_image_sync(prompt, image_base64)
+            description, _ = await self._llm.generate_response_for_image(prompt, image_base64)
             return f"[图片：{description}]"
         except Exception as e:
             print(f"\033[1;31m[错误]\033[0m AI接口调用失败: {str(e)}")
             return "[图片]"
 
-    def translate_forward(self) -> str:
+    async def translate_forward(self) -> str:
         """处理转发消息"""
         try:
             if 'content' not in self.params:
@@ -249,6 +252,7 @@ class CQCode:
                                 plain_text=raw_message,
                                 group_id=msg.get('group_id', 0)
                             )
+                            await message_obj.initialize()
                             content = message_obj.processed_plain_text
                         else:
                             content = '[空消息]'
@@ -263,6 +267,7 @@ class CQCode:
                             plain_text=raw_message,
                             group_id=msg.get('group_id', 0)
                         )
+                        await message_obj.initialize()
                         content = message_obj.processed_plain_text
                     else:
                         content = '[空消息]'
@@ -279,7 +284,7 @@ class CQCode:
             print(f"\033[1;31m[错误]\033[0m 处理转发消息失败: {str(e)}")
             return '[转发消息]'
 
-    def translate_reply(self) -> str:
+    async def translate_reply(self) -> str:
         """处理回复类型的CQ码"""
 
         # 创建Message对象
@@ -295,6 +300,7 @@ class CQCode:
                 raw_message=str(self.reply_message.message),
                 group_id=self.group_id
             )
+            await message_obj.initialize()
             if message_obj.user_id == global_config.BOT_QQ:
                 return f"[回复 {global_config.BOT_NICKNAME} 的消息: {message_obj.processed_plain_text}]"
             else:
@@ -308,9 +314,9 @@ class CQCode:
     def unescape(text: str) -> str:
         """反转义CQ码中的特殊字符"""
         return text.replace(',', ',') \
-                  .replace('[', '[') \
+            .replace('[', '[') \
-                  .replace(']', ']') \
+            .replace(']', ']') \
-                  .replace('&', '&')
+            .replace('&', '&')
 
     @staticmethod
     def create_emoji_cq(file_path: str) -> str:
@@ -325,15 +331,16 @@ class CQCode:
         abs_path = os.path.abspath(file_path)
         # 转义特殊字符
         escaped_path = abs_path.replace('&', '&') \
-                             .replace('[', '[') \
+            .replace('[', '[') \
-                             .replace(']', ']') \
+            .replace(']', ']') \
-                             .replace(',', ',')
+            .replace(',', ',')
         # 生成CQ码，设置sub_type=1表示这是表情包
         return f"[CQ:image,file=file:///{escaped_path},sub_type=1]"
 
+
 class CQCode_tool:
     @staticmethod
-    def cq_from_dict_to_class(cq_code: Dict, reply: Optional[Dict] = None) -> CQCode:
+    async def cq_from_dict_to_class(cq_code: Dict, reply: Optional[Dict] = None) -> CQCode:
         """
         将CQ码字典转换为CQCode对象
         
@@ -362,7 +369,7 @@ class CQCode_tool:
         )
 
         # 进行翻译处理
-        instance.translate()
+        await instance.translate()
         return instance
 
     @staticmethod

src/plugins/chat/message.py

@@ -33,52 +33,54 @@ class Message:
 
     user_id: int = None
     user_nickname: str = None  # 用户昵称
-    user_cardname: str=None # 用户群昵称
+    user_cardname: str = None  # 用户群昵称
 
-    raw_message: str = None # 原始消息，包含未解析的cq码
+    raw_message: str = None  # 原始消息，包含未解析的cq码
-    plain_text: str = None # 纯文本
+    plain_text: str = None  # 纯文本
 
+    reply_message: Dict = None  # 存储 回复的 源消息
+
+    # 延迟初始化字段
+    _initialized: bool = False
     message_segments: List[Dict] = None  # 存储解析后的消息片段
     processed_plain_text: str = None  # 用于存储处理后的plain_text
     detailed_plain_text: str = None  # 用于存储详细可读文本
 
-    reply_message: Dict = None  # 存储 回复的 源消息
+    # 状态标志
+    is_emoji: bool = False
+    has_emoji: bool = False
+    translate_cq: bool = True
 
-    is_emoji: bool = False # 是否是表情包
+    async def initialize(self):
-    has_emoji: bool = False # 是否包含表情包
+        """显式异步初始化方法（必须调用）"""
+        if self._initialized:
+            return
 
-    translate_cq: bool = True # 是否翻译cq码
+        # 异步获取补充信息
+        self.group_name = self.group_name or get_groupname(self.group_id)
+        self.user_nickname = self.user_nickname or get_user_nickname(self.user_id)
+        self.user_cardname = self.user_cardname or get_user_cardname(self.user_id)
 
-    def __post_init__(self):
+        # 消息解析
-        if self.time is None:
+        if self.raw_message:
-            self.time = int(time.time())
+            self.message_segments = await self.parse_message_segments(self.raw_message)
+            self.processed_plain_text = ' '.join(
+                seg.translated_plain_text
+                for seg in self.message_segments
+            )
 
-        if not self.group_name:
+        # 构建详细文本
-            self.group_name = get_groupname(self.group_id)
-        
-        if not self.user_nickname:
-            self.user_nickname = get_user_nickname(self.user_id)
-            
-        if not self.user_cardname:
-            self.user_cardname=get_user_cardname(self.user_id)
-        
-        if not self.processed_plain_text:
-            if self.raw_message:
-                self.message_segments = self.parse_message_segments(str(self.raw_message))
-                self.processed_plain_text = ' '.join(
-                    seg.translated_plain_text
-                    for seg in self.message_segments
-                )
-        #将详细翻译为详细可读文本
         time_str = time.strftime("%m-%d %H:%M:%S", time.localtime(self.time))
-        try:
+        name = (
-            name = f"{self.user_nickname}(ta的昵称:{self.user_cardname},ta的id:{self.user_id})"
+            f"{self.user_nickname}(ta的昵称:{self.user_cardname},ta的id:{self.user_id})"
-        except:
+            if self.user_cardname
-            name = self.user_nickname or f"用户{self.user_id}"
+            else f"{self.user_nickname or f'用户{self.user_id}'}"
-        content = self.processed_plain_text
+        )
-        self.detailed_plain_text = f"[{time_str}] {name}: {content}\n"
+        self.detailed_plain_text = f"[{time_str}] {name}: {self.processed_plain_text}\n"
 
-    def parse_message_segments(self, message: str) -> List[CQCode]:
+        self._initialized = True
+    
+    async def parse_message_segments(self, message: str) -> List[CQCode]:
         """
         将消息解析为片段列表，包括纯文本和CQ码
         返回的列表中每个元素都是字典，包含：
@@ -136,7 +138,7 @@ class Message:
         
         #翻译作为字典的CQ码  
         for _code_item in cq_code_dict_list:
-            message_obj = cq_code_tool.cq_from_dict_to_class(_code_item,reply = self.reply_message)
+            message_obj = await cq_code_tool.cq_from_dict_to_class(_code_item,reply = self.reply_message)
             trans_list.append(message_obj)       
         return trans_list
 

src/plugins/chat/prompt_builder.py

@@ -2,7 +2,7 @@ import time
 import random
 from ..schedule.schedule_generator import bot_schedule
 import os
-from .utils import get_embedding, combine_messages, get_recent_group_detailed_plain_text,find_similar_topics
+from .utils import get_embedding, combine_messages, get_recent_group_detailed_plain_text
 from ...common.database import Database
 from .config import global_config
 from .topic_identifier import topic_identifier
@@ -60,7 +60,7 @@ class PromptBuilder:
         
         prompt_info = ''
         promt_info_prompt = ''
-        prompt_info = self.get_prompt_info(message_txt,threshold=0.5)
+        prompt_info = await self.get_prompt_info(message_txt,threshold=0.5)
         if prompt_info:
             prompt_info = f'''\n----------------------------------------------------\n你有以下这些[知识]：\n{prompt_info}\n请你记住上面的[知识]，之后可能会用到\n----------------------------------------------------\n'''
             
@@ -215,10 +215,10 @@ class PromptBuilder:
         return prompt_for_initiative
         
 
-    def get_prompt_info(self,message:str,threshold:float):
+    async def get_prompt_info(self,message:str,threshold:float):
         related_info = ''
         print(f"\033[1;34m[调试]\033[0m 获取知识库内容，元消息：{message[:30]}...，消息长度: {len(message)}")
-        embedding = get_embedding(message)
+        embedding = await get_embedding(message)
         related_info += self.get_info_from_db(embedding,threshold=threshold)
             
         return related_info

src/plugins/chat/utils.py

@@ -38,11 +38,13 @@ def combine_messages(messages: List[Message]) -> str:
 
     return result
 
-def db_message_to_str (message_dict: Dict) -> str:
+
+def db_message_to_str(message_dict: Dict) -> str:
     print(f"message_dict: {message_dict}")
     time_str = time.strftime("%m-%d %H:%M:%S", time.localtime(message_dict["time"]))
     try:
-        name="[(%s)%s]%s" % (message_dict['user_id'],message_dict.get("user_nickname", ""),message_dict.get("user_cardname", ""))
+        name = "[(%s)%s]%s" % (
+        message_dict['user_id'], message_dict.get("user_nickname", ""), message_dict.get("user_cardname", ""))
     except:
         name = message_dict.get("user_nickname", "") or f"用户{message_dict['user_id']}"
     content = message_dict.get("processed_plain_text", "")
@@ -59,6 +61,7 @@ def is_mentioned_bot_in_message(message: Message) -> bool:
             return True
     return False
 
+
 def is_mentioned_bot_in_txt(message: str) -> bool:
     """检查消息是否提到了机器人"""
     keywords = [global_config.BOT_NICKNAME]
@@ -67,10 +70,13 @@ def is_mentioned_bot_in_txt(message: str) -> bool:
             return True
     return False
 
-def get_embedding(text):
+
+async def get_embedding(text):
     """获取文本的embedding向量"""
     llm = LLM_request(model=global_config.embedding)
-    return llm.get_embedding_sync(text)
+    # return llm.get_embedding_sync(text)
+    return await llm.get_embedding(text)
+
 
 def cosine_similarity(v1, v2):
     dot_product = np.dot(v1, v2)
@@ -78,6 +84,7 @@ def cosine_similarity(v1, v2):
     norm2 = np.linalg.norm(v2)
     return dot_product / (norm1 * norm2)
 
+
 def calculate_information_content(text):
     """计算文本的信息量（熵）"""
     char_count = Counter(text)
@@ -90,6 +97,7 @@ def calculate_information_content(text):
 
     return entropy
 
+
 def get_cloest_chat_from_db(db, length: int, timestamp: str):
     """从数据库中获取最接近指定时间戳的聊天记录，并记录读取次数"""
     chat_text = ''
@@ -107,7 +115,7 @@ def get_cloest_chat_from_db(db, length: int, timestamp: str):
         for record in chat_records:
             # 检查当前记录的memorized值
             current_memorized = record.get('memorized', 0)
-            if current_memorized  > 3:
+            if current_memorized > 3:
                 # print(f"消息已读取3次，跳过")
                 return ''
 
@@ -123,7 +131,8 @@ def get_cloest_chat_from_db(db, length: int, timestamp: str):
     # print(f"消息已读取3次，跳过")
     return ''
 
-def get_recent_group_messages(db, group_id: int, limit: int = 12) -> list:
+
+async def get_recent_group_messages(db, group_id: int, limit: int = 12) -> list:
     """从数据库获取群组最近的消息记录
     
     Args:
@@ -135,7 +144,7 @@ def get_recent_group_messages(db, group_id: int, limit: int = 12) -> list:
         list: Message对象列表，按时间正序排列
     """
 
-        # 从数据库获取最近消息
+    # 从数据库获取最近消息
     recent_messages = list(db.db.messages.find(
         {"group_id": group_id},
         # {
@@ -165,6 +174,7 @@ def get_recent_group_messages(db, group_id: int, limit: int = 12) -> list:
                 processed_plain_text=msg_data.get("processed_text", ""),
                 group_id=group_id
             )
+            await msg.initialize()
             message_objects.append(msg)
         except KeyError:
             print("[WARNING] 数据库中存在无效的消息")
@@ -174,7 +184,8 @@ def get_recent_group_messages(db, group_id: int, limit: int = 12) -> list:
     message_objects.reverse()
     return message_objects
 
-def get_recent_group_detailed_plain_text(db, group_id: int, limit: int = 12,combine = False):
+
+def get_recent_group_detailed_plain_text(db, group_id: int, limit: int = 12, combine=False):
     recent_messages = list(db.db.messages.find(
         {"group_id": group_id},
         {
@@ -197,7 +208,7 @@ def get_recent_group_detailed_plain_text(db, group_id: int, limit: int = 12,comb
 
     if combine:
         for msg_db_data in recent_messages:
-            message_detailed_plain_text+=str(msg_db_data["detailed_plain_text"])
+            message_detailed_plain_text += str(msg_db_data["detailed_plain_text"])
         return message_detailed_plain_text
     else:
         for msg_db_data in recent_messages:
@@ -205,7 +216,6 @@ def get_recent_group_detailed_plain_text(db, group_id: int, limit: int = 12,comb
         return message_detailed_plain_text_list
 
 
-
 def split_into_sentences_w_remove_punctuation(text: str) -> List[str]:
     """将文本分割成句子，但保持书名号中的内容完整
     Args:
@@ -225,7 +235,7 @@ def split_into_sentences_w_remove_punctuation(text: str) -> List[str]:
         split_strength = 0.7
     else:
         split_strength = 0.9
-    #先移除换行符
+    # 先移除换行符
     # print(f"split_strength: {split_strength}")
 
     # print(f"处理前的文本: {text}")
@@ -239,11 +249,11 @@ def split_into_sentences_w_remove_punctuation(text: str) -> List[str]:
     text_no_1 = ''
     for letter in text:
         # print(f"当前字符: {letter}")
-        if letter in ['!','！','?','？']:
+        if letter in ['!', '！', '?', '？']:
             # print(f"当前字符: {letter}, 随机数: {random.random()}")
             if random.random() < split_strength:
                 letter = ''
-        if letter in ['。','…']:
+        if letter in ['。', '…']:
             # print(f"当前字符: {letter}, 随机数: {random.random()}")
             if random.random() < 1 - split_strength:
                 letter = ''
@@ -277,7 +287,7 @@ def split_into_sentences_w_remove_punctuation(text: str) -> List[str]:
     sentences_done = []
     for sentence in sentences:
         sentence = sentence.rstrip('，,')
-        if random.random() < split_strength*0.5:
+        if random.random() < split_strength * 0.5:
             sentence = sentence.replace('，', '').replace(',', '')
         elif random.random() < split_strength:
             sentence = sentence.replace('，', ' ').replace(',', ' ')
@@ -287,6 +297,7 @@ def split_into_sentences_w_remove_punctuation(text: str) -> List[str]:
     return sentences_done
 
 
+
 def random_remove_punctuation(text: str) -> str:
     """随机处理标点符号，模拟人类打字习惯
     
@@ -314,11 +325,12 @@ def random_remove_punctuation(text: str) -> str:
     return result
 
 
+
 def process_llm_response(text: str) -> List[str]:
     # processed_response = process_text_with_typos(content)
     if len(text) > 300:
-            print(f"回复过长 ({len(text)} 字符)，返回默认回复")
+        print(f"回复过长 ({len(text)} 字符)，返回默认回复")
-            return ['懒得说']
+        return ['懒得说']
     # 处理长消息
     typo_generator = ChineseTypoGenerator(
         error_rate=0.03,
@@ -335,6 +347,7 @@ def process_llm_response(text: str) -> List[str]:
 
     return sentences
 
+
 def calculate_typing_time(input_string: str, chinese_time: float = 0.2, english_time: float = 0.1) -> float:
     """
     计算输入字符串所需的时间，中文和英文字符有不同的输入时间
@@ -351,28 +364,6 @@ def calculate_typing_time(input_string: str, chinese_time: float = 0.2, english_
     return total_time
 
 
-def find_similar_topics(message_txt: str, all_memory_topic: list, top_k: int = 5) -> list:
-    """使用重排序API找出与输入文本最相似的话题
-    
-    Args:
-        message_txt: 输入文本
-        all_memory_topic: 所有记忆主题列表
-        top_k: 返回最相似的话题数量
-        
-    Returns:
-        list: 最相似话题列表及其相似度分数
-    """
-    
-    if not all_memory_topic:
-        return []
-    
-    try:
-        llm = LLM_request(model=global_config.rerank)
-        return llm.rerank_sync(message_txt, all_memory_topic, top_k)
-    except Exception as e:
-        print(f"重排序API调用出错: {str(e)}")
-        return []
-    
 def cosine_similarity(v1, v2):
     """计算余弦相似度"""
     dot_product = np.dot(v1, v2)
@@ -382,6 +373,7 @@ def cosine_similarity(v1, v2):
         return 0
     return dot_product / (norm1 * norm2)
 
+
 def text_to_vector(text):
     """将文本转换为词频向量"""
     # 分词
@@ -390,6 +382,7 @@ def text_to_vector(text):
     word_freq = Counter(words)
     return word_freq
 
+
 def find_similar_topics_simple(text: str, topics: list, top_k: int = 5) -> list:
     """使用简单的余弦相似度计算文本相似度"""
     # 将输入文本转换为词频向量

src/plugins/memory_system/memory.py

@@ -11,7 +11,7 @@ from ..chat.config import global_config
 from ...common.database import Database # 使用正确的导入语法
 from ..models.utils_model import LLM_request
 import math
-from ..chat.utils import calculate_information_content, get_cloest_chat_from_db ,find_similar_topics,text_to_vector,cosine_similarity
+from ..chat.utils import calculate_information_content, get_cloest_chat_from_db ,text_to_vector,cosine_similarity
 
 
 

src/plugins/models/utils_model.py

@@ -25,354 +25,193 @@ class LLM_request:
         self.model_name = model["name"]
         self.params = kwargs
 
-    async def generate_response(self, prompt: str) -> Tuple[str, str]:
+    async def _execute_request(
-        """根据输入的提示生成模型的异步响应"""
+            self,
-        headers = {
+            endpoint: str,
-            "Authorization": f"Bearer {self.api_key}",
+            prompt: str = None,
-            "Content-Type": "application/json"
+            image_base64: str = None,
+            payload: dict = None,
+            retry_policy: dict = None,
+            response_handler: callable = None,
+    ):
+        """统一请求执行入口
+        Args:
+            endpoint: API端点路径 (如 "chat/completions")
+            prompt: prompt文本
+            image_base64: 图片的base64编码
+            payload: 请求体数据
+            is_async: 是否异步
+            retry_policy: 自定义重试策略
+                (示例: {"max_retries":3, "base_wait":15, "retry_codes":[429,500]})
+            response_handler: 自定义响应处理器
+        """
+        # 合并重试策略
+        default_retry = {
+            "max_retries": 3, "base_wait": 15,
+            "retry_codes": [429, 413, 500, 503],
+            "abort_codes": [400, 401, 402, 403]}
+        policy = {**default_retry, **(retry_policy or {})}
+
+        # 常见Error Code Mapping
+        error_code_mapping = {
+            400: "参数不正确",
+            401: "API key 错误，认证失败",
+            402: "账号余额不足",
+            403: "需要实名,或余额不足",
+            404: "Not Found",
+            429: "请求过于频繁，请稍后再试",
+            500: "服务器内部故障",
+            503: "服务器负载过高"
         }
 
+        api_url = f"{self.base_url.rstrip('/')}/{endpoint.lstrip('/')}"
+        logger.info(f"发送请求到URL: {api_url}")
+        logger.info(f"使用模型: {self.model_name}")
+
         # 构建请求体
-        data = {
+        if image_base64:
-            "model": self.model_name,
+            payload = await self._build_payload(prompt, image_base64)
-            "messages": [{"role": "user", "content": prompt}],
+        elif payload is None:
-            **self.params
+            payload = await self._build_payload(prompt)
-        }
 
-        # 发送请求到完整的chat/completions端点
+        for retry in range(policy["max_retries"]):
-        api_url = f"{self.base_url.rstrip('/')}/chat/completions"
-        logger.info(f"发送请求到URL: {api_url}/{self.model_name}")  # 记录请求的URL
-
-        max_retries = 3
-        base_wait_time = 15
-
-        for retry in range(max_retries):
             try:
+                # 使用上下文管理器处理会话
+                headers = await self._build_headers()
+
                 async with aiohttp.ClientSession() as session:
-                    async with session.post(api_url, headers=headers, json=data) as response:
+                    async with session.post(api_url, headers=headers, json=payload) as response:
-                        if response.status == 429:
+                        # 处理需要重试的状态码
-                            wait_time = base_wait_time * (2 ** retry)  # 指数退避
+                        if response.status in policy["retry_codes"]:
-                            logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
+                            wait_time = policy["base_wait"] * (2 ** retry)
+                            logger.warning(f"错误码: {response.status}, 等待 {wait_time}秒后重试")
+                            if response.status == 413:
+                                logger.warning("请求体过大，尝试压缩...")
+                                image_base64 = compress_base64_image_by_scale(image_base64)
+                                payload = await self._build_payload(prompt, image_base64)
+                            elif response.status in [500, 503]:
+                                logger.error(f"错误码: {response.status} - {error_code_mapping.get(response.status)}")
+                                raise RuntimeError("服务器负载过高，模型恢复失败QAQ")
+                            else:
+                                logger.warning(f"请求限制(429)，等待{wait_time}秒后重试...")
+
                             await asyncio.sleep(wait_time)
                             continue
+                        elif response.status in policy["abort_codes"]:
+                            logger.error(f"错误码: {response.status} - {error_code_mapping.get(response.status)}")
+                            raise RuntimeError(f"请求被拒绝: {error_code_mapping.get(response.status)}")
 
-                        if response.status in [500, 503]:
+                        response.raise_for_status()
-                            logger.error(f"服务器错误: {response.status}")
-                            raise RuntimeError("服务器负载过高，模型恢复失败QAQ")
-
-                        response.raise_for_status()  # 检查其他响应状态
-
                         result = await response.json()
-                        if "choices" in result and len(result["choices"]) > 0:
+
-                            message = result["choices"][0]["message"]
+                        # 使用自定义处理器或默认处理
-                            content = message.get("content", "")
+                        return response_handler(result) if response_handler else self._default_response_handler(result)
-                            think_match = None
-                            reasoning_content = message.get("reasoning_content", "")
-                            if not reasoning_content:
-                                think_match = re.search(r'(?:<think>)?(.*?)</think>', content, re.DOTALL)
-                            if think_match:
-                                reasoning_content = think_match.group(1).strip()
-                            content = re.sub(r'(?:<think>)?.*?</think>', '', content, flags=re.DOTALL, count=1).strip()
-                            return content, reasoning_content
-                        return "没有返回结果", ""
 
             except Exception as e:
-                if retry < max_retries - 1:  # 如果还有重试机会
+                if retry < policy["max_retries"] - 1:
-                    wait_time = base_wait_time * (2 ** retry)
+                    wait_time = policy["base_wait"] * (2 ** retry)
-                    logger.error(f"[回复]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
+                    logger.error(f"请求失败，等待{wait_time}秒后重试... 错误: {str(e)}")
                     await asyncio.sleep(wait_time)
                 else:
-                    logger.critical(f"请求失败: {str(e)}", exc_info=True)
+                    logger.critical(f"请求失败: {str(e)}")
-                    logger.critical(f"请求头: {headers} 请求体: {data}")
+                    logger.critical(f"请求头: {await self._build_headers()} 请求体: {payload}")
                     raise RuntimeError(f"API请求失败: {str(e)}")
 
         logger.error("达到最大重试次数，请求仍然失败")
         raise RuntimeError("达到最大重试次数，API请求仍然失败")
 
-    async def generate_response_for_image(self, prompt: str, image_base64: str) -> Tuple[str, str]:
+    async def _build_payload(self, prompt: str, image_base64: str = None) -> dict:
-        """根据输入的提示和图片生成模型的异步响应"""
+        """构建请求体"""
-        headers = {
+        if image_base64:
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-
-        # 构建请求体
-        def build_request_data(img_base64: str):
             return {
                 "model": self.model_name,
                 "messages": [
                     {
                         "role": "user",
                         "content": [
-                            {
+                            {"type": "text", "text": prompt},
-                                "type": "text",
+                            {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_base64}"}}
-                                "text": prompt
-                            },
-                            {
-                                "type": "image_url",
-                                "image_url": {
-                                    "url": f"data:image/jpeg;base64,{img_base64}"
-                                }
-                            }
                         ]
                     }
                 ],
+                "max_tokens": global_config.max_response_length,
+                **self.params
+            }
+        else:
+            return {
+                "model": self.model_name,
+                "messages": [{"role": "user", "content": prompt}],
+                "max_tokens": global_config.max_response_length,
                 **self.params
             }
 
+    def _default_response_handler(self, result: dict) -> Tuple:
+        """默认响应解析"""
+        if "choices" in result and result["choices"]:
+            message = result["choices"][0]["message"]
+            content = message.get("content", "")
+            content, reasoning = self._extract_reasoning(content)
+            reasoning_content = message.get("model_extra", {}).get("reasoning_content", "")
+            if not reasoning_content:
+                reasoning_content = reasoning
 
-        # 发送请求到完整的chat/completions端点
+            return content, reasoning_content
-        api_url = f"{self.base_url.rstrip('/')}/chat/completions"
-        logger.info(f"发送请求到URL: {api_url}/{self.model_name}")  # 记录请求的URL
 
-        max_retries = 3
+        return "没有返回结果", ""
-        base_wait_time = 15
 
-        current_image_base64 = image_base64
+    def _extract_reasoning(self, content: str) -> tuple[str, str]:
-        current_image_base64 = compress_base64_image_by_scale(current_image_base64)
+        """CoT思维链提取"""
+        match = re.search(r'(?:<think>)?(.*?)</think>', content, re.DOTALL)
+        content = re.sub(r'(?:<think>)?.*?</think>', '', content, flags=re.DOTALL, count=1).strip()
+        if match:
+            reasoning = match.group(1).strip()
+        else:
+            reasoning = ""
+        return content, reasoning
 
-        for retry in range(max_retries):
+    async def _build_headers(self) -> dict:
-            try:
+        """构建请求头"""
-                data = build_request_data(current_image_base64)
+        return {
-                async with aiohttp.ClientSession() as session:
-                    async with session.post(api_url, headers=headers, json=data) as response:
-                        if response.status == 429:
-                            wait_time = base_wait_time * (2 ** retry)  # 指数退避
-                            logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                            await asyncio.sleep(wait_time)
-                            continue
-
-                        elif response.status == 413:
-                            logger.warning("图片太大(413)，尝试压缩...")
-                            current_image_base64 = compress_base64_image_by_scale(current_image_base64)
-                            continue
-
-                        response.raise_for_status()  # 检查其他响应状态
-
-                        result = await response.json()
-                        if "choices" in result and len(result["choices"]) > 0:
-                            message = result["choices"][0]["message"]
-                            content = message.get("content", "")
-                            think_match = None
-                            reasoning_content = message.get("reasoning_content", "")
-                            if not reasoning_content:
-                                think_match = re.search(r'(?:<think>)?(.*?)</think>', content, re.DOTALL)
-                            if think_match:
-                                reasoning_content = think_match.group(1).strip()
-                            content = re.sub(r'(?:<think>)?.*?</think>', '', content, flags=re.DOTALL, count=1).strip()
-                            return content, reasoning_content
-                        return "没有返回结果", ""
-
-            except Exception as e:
-                if retry < max_retries - 1:  # 如果还有重试机会
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[image回复]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    await asyncio.sleep(wait_time)
-                else:
-                    logger.critical(f"请求失败: {str(e)}", exc_info=True)
-                    logger.critical(f"请求头: {headers} 请求体: {data}")
-                    raise RuntimeError(f"API请求失败: {str(e)}")
-
-        logger.error("达到最大重试次数，请求仍然失败")
-        raise RuntimeError("达到最大重试次数，API请求仍然失败")
-
-    async def generate_response_async(self, prompt: str) -> Union[str, Tuple[str, str]]:
-        """异步方式根据输入的提示生成模型的响应"""
-        headers = {
             "Authorization": f"Bearer {self.api_key}",
             "Content-Type": "application/json"
         }
 
+    async def generate_response(self, prompt: str) -> Tuple[str, str]:
+        """根据输入的提示生成模型的异步响应"""
+
+        content, reasoning_content = await self._execute_request(
+            endpoint="/chat/completions",
+            prompt=prompt
+        )
+        return content, reasoning_content
+
+    async def generate_response_for_image(self, prompt: str, image_base64: str) -> Tuple[str, str]:
+        """根据输入的提示和图片生成模型的异步响应"""
+
+        content, reasoning_content = await self._execute_request(
+            endpoint="/chat/completions",
+            prompt=prompt,
+            image_base64=image_base64
+        )
+        return content, reasoning_content
+
+    async def generate_response_async(self, prompt: str) -> Union[str, Tuple[str, str]]:
+        """异步方式根据输入的提示生成模型的响应"""
         # 构建请求体
         data = {
             "model": self.model_name,
             "messages": [{"role": "user", "content": prompt}],
             "temperature": 0.5,
+            "max_tokens": global_config.max_response_length,
             **self.params
         }
 
-        # 发送请求到完整的 chat/completions 端点
+        content, reasoning_content = await self._execute_request(
-        api_url = f"{self.base_url.rstrip('/')}/chat/completions"
+            endpoint="/chat/completions",
-        logger.info(f"Request URL: {api_url}")  # 记录请求的 URL
+            payload=data,
-
+            prompt=prompt
-        max_retries = 3
+        )
-        base_wait_time = 15
+        return content, reasoning_content
-
-        async with aiohttp.ClientSession() as session:
-            for retry in range(max_retries):
-                try:
-                    async with session.post(api_url, headers=headers, json=data) as response:
-                        if response.status == 429:
-                            wait_time = base_wait_time * (2 ** retry)  # 指数退避
-                            logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                            await asyncio.sleep(wait_time)
-                            continue
-
-                        response.raise_for_status()  # 检查其他响应状态
-
-                        result = await response.json()
-                        if "choices" in result and len(result["choices"]) > 0:
-                            message = result["choices"][0]["message"]
-                            content = message.get("content", "")
-                            think_match = None
-                            reasoning_content = message.get("reasoning_content", "")
-                            if not reasoning_content:
-                                think_match = re.search(r'(?:<think>)?(.*?)</think>', content, re.DOTALL)
-                            if think_match:
-                                reasoning_content = think_match.group(1).strip()
-                            content = re.sub(r'(?:<think>)?.*?</think>', '', content, flags=re.DOTALL, count=1).strip()
-                            return content, reasoning_content
-                        return "没有返回结果", ""
-
-                except Exception as e:
-                    if retry < max_retries - 1:  # 如果还有重试机会
-                        wait_time = base_wait_time * (2 ** retry)
-                        logger.error(f"[回复]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}")
-                        await asyncio.sleep(wait_time)
-                    else:
-                        logger.error(f"请求失败: {str(e)}")
-                        logger.critical(f"请求头: {headers} 请求体: {data}")
-                        return f"请求失败: {str(e)}", ""
-
-            logger.error("达到最大重试次数，请求仍然失败")
-            return "达到最大重试次数，请求仍然失败", ""
-
-
-
-    def generate_response_for_image_sync(self, prompt: str, image_base64: str) -> Tuple[str, str]:
-        """同步方法：根据输入的提示和图片生成模型的响应"""
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-
-        image_base64=compress_base64_image_by_scale(image_base64)
-
-        # 构建请求体
-        data = {
-            "model": self.model_name,
-            "messages": [
-                {
-                    "role": "user",
-                    "content": [
-                        {
-                            "type": "text",
-                            "text": prompt
-                        },
-                        {
-                            "type": "image_url",
-                            "image_url": {
-                                "url": f"data:image/jpeg;base64,{image_base64}"
-                            }
-                        }
-                    ]
-                }
-            ],
-            **self.params
-        }
-
-        # 发送请求到完整的chat/completions端点
-        api_url = f"{self.base_url.rstrip('/')}/chat/completions"
-        logger.info(f"发送请求到URL: {api_url}/{self.model_name}")  # 记录请求的URL
-
-        max_retries = 2
-        base_wait_time = 6
-
-        for retry in range(max_retries):
-            try:
-                response = requests.post(api_url, headers=headers, json=data, timeout=30)
-
-                if response.status_code == 429:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                    time.sleep(wait_time)
-                    continue
-
-                response.raise_for_status()  # 检查其他响应状态
-
-                result = response.json()
-                if "choices" in result and len(result["choices"]) > 0:
-                    message = result["choices"][0]["message"]
-                    content = message.get("content", "")
-                    think_match = None
-                    reasoning_content = message.get("reasoning_content", "")
-                    if not reasoning_content:
-                        think_match = re.search(r'(?:<think>)?(.*?)</think>', content, re.DOTALL)
-                    if think_match:
-                        reasoning_content = think_match.group(1).strip()
-                    content = re.sub(r'(?:<think>)?.*?</think>', '', content, flags=re.DOTALL, count=1).strip()
-                    return content, reasoning_content
-                return "没有返回结果", ""
-
-            except Exception as e:
-                if retry < max_retries - 1:  # 如果还有重试机会
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[image_sync回复]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    time.sleep(wait_time)
-                else:
-                    logger.critical(f"请求失败: {str(e)}", exc_info=True)
-                    logger.critical(f"请求头: {headers} 请求体: {data}")
-                    raise RuntimeError(f"API请求失败: {str(e)}")
-
-        logger.error("达到最大重试次数，请求仍然失败")
-        raise RuntimeError("达到最大重试次数，API请求仍然失败")
-
-    def get_embedding_sync(self, text: str, model: str = "BAAI/bge-m3") -> Union[list, None]:
-        """同步方法：获取文本的embedding向量
-        
-        Args:
-            text: 需要获取embedding的文本
-            model: 使用的模型名称，默认为"BAAI/bge-m3"
-            
-        Returns:
-            list: embedding向量，如果失败则返回None
-        """
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-
-        data = {
-            "model": model,
-            "input": text,
-            "encoding_format": "float"
-        }
-
-        api_url = f"{self.base_url.rstrip('/')}/embeddings"
-        logger.info(f"发送请求到URL: {api_url}/{self.model_name}")  # 记录请求的URL
-
-        max_retries = 2
-        base_wait_time = 6
-
-        for retry in range(max_retries):
-            try:
-                response = requests.post(api_url, headers=headers, json=data, timeout=30)
-
-                if response.status_code == 429:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                    time.sleep(wait_time)
-                    continue
-
-                response.raise_for_status()
-
-                result = response.json()
-                if 'data' in result and len(result['data']) > 0:
-                    return result['data'][0]['embedding']
-                return None
-
-            except Exception as e:
-                if retry < max_retries - 1:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[embedding_sync]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    time.sleep(wait_time)
-                else:
-                    logger.critical(f"embedding请求失败: {str(e)}", exc_info=True)
-                    logger.critical(f"请求头: {headers} 请求体: {data}")
-                    return None
-
-        logger.error("达到最大重试次数，embedding请求仍然失败")
-        return None
 
     async def get_embedding(self, text: str, model: str = "BAAI/bge-m3") -> Union[list, None]:
         """异步方法：获取文本的embedding向量
@@ -384,245 +223,24 @@ class LLM_request:
         Returns:
             list: embedding向量，如果失败则返回None
         """
-        headers = {
+        def embedding_handler(result):
-            "Authorization": f"Bearer {self.api_key}",
+            """处理响应"""
-            "Content-Type": "application/json"
+            if "data" in result and len(result["data"]) > 0:
-        }
+                return result["data"][0].get("embedding", None)
+            return None
 
-        data = {
+        embedding = await self._execute_request(
-            "model": model,
+            endpoint="/embeddings",
-            "input": text,
+            prompt=text,
-            "encoding_format": "float"
+            payload={
-        }
+                "model": model,
-
+                "input": text,
-        api_url = f"{self.base_url.rstrip('/')}/embeddings"
+                "encoding_format": "float"
-        logger.info(f"发送请求到URL: {api_url}/{self.model_name}")  # 记录请求的URL
+            },
-
+            retry_policy={
-        max_retries = 3
+                "max_retries": 2,
-        base_wait_time = 15
+                "base_wait": 6
-
+            },
-        for retry in range(max_retries):
+            response_handler=embedding_handler
-            try:
+        )
-                async with aiohttp.ClientSession() as session:
+        return embedding
-                    async with session.post(api_url, headers=headers, json=data) as response:
-                        if response.status == 429:
-                            wait_time = base_wait_time * (2 ** retry)
-                            logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                            await asyncio.sleep(wait_time)
-                            continue
-
-                        response.raise_for_status()
-
-                        result = await response.json()
-                        if 'data' in result and len(result['data']) > 0:
-                            return result['data'][0]['embedding']
-                        return None
-
-            except Exception as e:
-                if retry < max_retries - 1:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[embedding]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    await asyncio.sleep(wait_time)
-                else:
-                    logger.critical(f"embedding请求失败: {str(e)}", exc_info=True)
-                    logger.critical(f"请求头: {headers} 请求体: {data}")
-                    return None
-
-        logger.error("达到最大重试次数，embedding请求仍然失败")
-        return None
-
-    def rerank_sync(self, query: str, documents: list, top_k: int = 5) -> list:
-        """同步方法：使用重排序API对文档进行排序
-        
-        Args:
-            query: 查询文本
-            documents: 待排序的文档列表
-            top_k: 返回前k个结果
-            
-        Returns:
-            list: [(document, score), ...] 格式的结果列表
-        """
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-
-        data = {
-            "model": self.model_name,
-            "query": query,
-            "documents": documents,
-            "top_n": top_k,
-            "return_documents": True,
-        }
-
-        api_url = f"{self.base_url.rstrip('/')}/rerank"
-        logger.info(f"发送请求到URL: {api_url}")
-
-        max_retries = 2
-        base_wait_time = 6
-
-        for retry in range(max_retries):
-            try:
-                response = requests.post(api_url, headers=headers, json=data, timeout=30)
-
-                if response.status_code == 429:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                    time.sleep(wait_time)
-                    continue
-                
-                if response.status_code in [500, 503]:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"服务器错误({response.status_code})，等待{wait_time}秒后重试...")
-                    if retry < max_retries - 1:
-                        time.sleep(wait_time)
-                        continue
-                    else:
-                        # 如果是最后一次重试，尝试使用chat/completions作为备选方案
-                        return self._fallback_rerank_with_chat(query, documents, top_k)
-
-                response.raise_for_status()
-
-                result = response.json()
-                if 'results' in result:
-                    return [(item["document"], item["score"]) for item in result["results"]]
-                return []
-
-            except Exception as e:
-                if retry < max_retries - 1:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[rerank_sync]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    time.sleep(wait_time)
-                else:
-                    logger.critical(f"重排序请求失败: {str(e)}", exc_info=True)
-
-        logger.error("达到最大重试次数，重排序请求仍然失败")
-        return []
-
-    async def rerank(self, query: str, documents: list, top_k: int = 5) -> list:
-        """异步方法：使用重排序API对文档进行排序
-        
-        Args:
-            query: 查询文本
-            documents: 待排序的文档列表
-            top_k: 返回前k个结果
-            
-        Returns:
-            list: [(document, score), ...] 格式的结果列表
-        """
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json"
-        }
-
-        data = {
-            "model": self.model_name,
-            "query": query,
-            "documents": documents,
-            "top_n": top_k,
-            "return_documents": True,
-        }
-
-        api_url = f"{self.base_url.rstrip('/')}/v1/rerank"
-        logger.info(f"发送请求到URL: {api_url}")
-
-        max_retries = 3
-        base_wait_time = 15
-
-        for retry in range(max_retries):
-            try:
-                async with aiohttp.ClientSession() as session:
-                    async with session.post(api_url, headers=headers, json=data) as response:
-                        if response.status == 429:
-                            wait_time = base_wait_time * (2 ** retry)
-                            logger.warning(f"遇到请求限制(429)，等待{wait_time}秒后重试...")
-                            await asyncio.sleep(wait_time)
-                            continue
-
-                        if response.status in [500, 503]:
-                            wait_time = base_wait_time * (2 ** retry)
-                            logger.error(f"服务器错误({response.status})，等待{wait_time}秒后重试...")
-                            if retry < max_retries - 1:
-                                await asyncio.sleep(wait_time)
-                                continue
-                            else:
-                                # 如果是最后一次重试，尝试使用chat/completions作为备选方案
-                                return await self._fallback_rerank_with_chat_async(query, documents, top_k)
-
-                        response.raise_for_status()
-
-                        result = await response.json()
-                        if 'results' in result:
-                            return [(item["document"], item["score"]) for item in result["results"]]
-                        return []
-
-            except Exception as e:
-                if retry < max_retries - 1:
-                    wait_time = base_wait_time * (2 ** retry)
-                    logger.error(f"[rerank]请求失败，等待{wait_time}秒后重试... 错误: {str(e)}", exc_info=True)
-                    await asyncio.sleep(wait_time)
-                else:
-                    logger.critical(f"重排序请求失败: {str(e)}", exc_info=True)
-                    # 作为最后的备选方案，尝试使用chat/completions
-                    return await self._fallback_rerank_with_chat_async(query, documents, top_k)
-
-        logger.error("达到最大重试次数，重排序请求仍然失败")
-        return []
-
-    async def _fallback_rerank_with_chat_async(self, query: str, documents: list, top_k: int = 5) -> list:
-        """当rerank API失败时的备选方案，使用chat/completions异步实现重排序
-        
-        Args:
-            query: 查询文本
-            documents: 待排序的文档列表
-            top_k: 返回前k个结果
-            
-        Returns:
-            list: [(document, score), ...] 格式的结果列表
-        """
-        try:
-            logger.info("使用chat/completions作为重排序的备选方案")
-            
-            # 构建提示词
-            prompt = f"""请对以下文档列表进行重排序，按照与查询的相关性从高到低排序。
-查询: {query}
-
-文档列表:
-{documents}
-
-请以JSON格式返回排序结果，格式为：
-[{{"document": "文档内容", "score": 相关性分数}}, ...]
-只返回JSON，不要其他任何文字。"""
-
-            headers = {
-                "Authorization": f"Bearer {self.api_key}",
-                "Content-Type": "application/json"
-            }
-
-            data = {
-                "model": self.model_name,
-                "messages": [{"role": "user", "content": prompt}],
-                **self.params
-            }
-
-            api_url = f"{self.base_url.rstrip('/')}/v1/chat/completions"
-            
-            async with aiohttp.ClientSession() as session:
-                async with session.post(api_url, headers=headers, json=data) as response:
-                    response.raise_for_status()
-                    result = await response.json()
-                    
-                    if "choices" in result and len(result["choices"]) > 0:
-                        message = result["choices"][0]["message"]
-                        content = message.get("content", "")
-                        try:
-                            import json
-                            parsed_content = json.loads(content)
-                            if isinstance(parsed_content, list):
-                                return [(item["document"], item["score"]) for item in parsed_content]
-                        except:
-                            pass
-            return []
-        except Exception as e:
-            logger.error(f"备选方案也失败了: {str(e)}")
-            return []