better 更好的llm使用统计

src/plugins/chat/config.py

@@ -56,7 +56,6 @@ class BotConfig:
     llm_reasoning: Dict[str, str] = field(default_factory=lambda: {})
     llm_reasoning_minor: Dict[str, str] = field(default_factory=lambda: {})
     llm_normal: Dict[str, str] = field(default_factory=lambda: {})
-    llm_normal_minor: Dict[str, str] = field(default_factory=lambda: {})
     llm_topic_judge: Dict[str, str] = field(default_factory=lambda: {})
     llm_summary_by_topic: Dict[str, str] = field(default_factory=lambda: {})
     llm_emotion_judge: Dict[str, str] = field(default_factory=lambda: {})
@@ -235,7 +234,6 @@ class BotConfig:
                 "llm_reasoning",
                 "llm_reasoning_minor",
                 "llm_normal",
-                "llm_normal_minor",
                 "llm_topic_judge",
                 "llm_summary_by_topic",
                 "llm_emotion_judge",

src/plugins/chat/emoji_manager.py

@@ -38,9 +38,9 @@ class EmojiManager:
 
     def __init__(self):
         self._scan_task = None
-        self.vlm = LLM_request(model=global_config.vlm, temperature=0.3, max_tokens=1000, request_type="image")
+        self.vlm = LLM_request(model=global_config.vlm, temperature=0.3, max_tokens=1000, request_type="emoji")
         self.llm_emotion_judge = LLM_request(
-            model=global_config.llm_emotion_judge, max_tokens=600, temperature=0.8, request_type="image"
+            model=global_config.llm_emotion_judge, max_tokens=600, temperature=0.8, request_type="emoji"
         )  # 更高的温度，更少的token（后续可以根据情绪来调整温度）
 
     def _ensure_emoji_dir(self):
@@ -111,7 +111,7 @@ class EmojiManager:
             if not text_for_search:
                 logger.error("无法获取文本的情绪")
                 return None
-            text_embedding = await get_embedding(text_for_search)
+            text_embedding = await get_embedding(text_for_search, request_type="emoji")
             if not text_embedding:
                 logger.error("无法获取文本的embedding")
                 return None
@@ -310,7 +310,7 @@ class EmojiManager:
                     logger.info(f"[检查] 表情包检查通过: {check}")
 
                 if description is not None:
-                    embedding = await get_embedding(description)
+                    embedding = await get_embedding(description, request_type="emoji")
                     # 准备数据库记录
                     emoji_record = {
                         "filename": filename,

src/plugins/chat/llm_generator.py

@@ -32,10 +32,17 @@ class ResponseGenerator:
             temperature=0.7,
             max_tokens=1000,
             stream=True,
+            request_type="response",
+        )
+        self.model_v3 = LLM_request(
+            model=global_config.llm_normal, temperature=0.7, max_tokens=3000, request_type="response"
+        )
+        self.model_r1_distill = LLM_request(
+            model=global_config.llm_reasoning_minor, temperature=0.7, max_tokens=3000, request_type="response"
+        )
+        self.model_sum = LLM_request(
+            model=global_config.llm_summary_by_topic, temperature=0.7, max_tokens=3000, request_type="relation"
         )
-        self.model_v3 = LLM_request(model=global_config.llm_normal, temperature=0.7, max_tokens=3000)
-        self.model_r1_distill = LLM_request(model=global_config.llm_reasoning_minor, temperature=0.7, max_tokens=3000)
-        self.model_v25 = LLM_request(model=global_config.llm_normal_minor, temperature=0.7, max_tokens=3000)
         self.current_model_type = "r1"  # 默认使用 R1
         self.current_model_name = "unknown model"
 
@@ -175,7 +182,7 @@ class ResponseGenerator:
             """
 
             # 调用模型生成结果
-            result, _, _ = await self.model_v25.generate_response(prompt)
+            result, _, _ = await self.model_sum.generate_response(prompt)
             result = result.strip()
 
             # 解析模型输出的结果

src/plugins/chat/utils.py

@@ -55,9 +55,9 @@ def is_mentioned_bot_in_message(message: MessageRecv) -> bool:
     return False
 
 
-async def get_embedding(text):
+async def get_embedding(text, request_type="embedding"):
     """获取文本的embedding向量"""
-    llm = LLM_request(model=global_config.embedding, request_type="embedding")
+    llm = LLM_request(model=global_config.embedding, request_type=request_type)
     # return llm.get_embedding_sync(text)
     return await llm.get_embedding(text)