上传

rag2_0/intent_recognition/IntentRecognition.py

@@ -110,6 +110,7 @@ class IntentRecognizer:
         Returns:
             分类结果
         """
+        classification_start_time = time.time()
         classification_parser = PydanticOutputParser(pydantic_object=Classification)
         formatted_prompt = classification_prompt.format(user_input=query, 
                                                         classification_info=classification_info, 
@@ -119,7 +120,9 @@ class IntentRecognizer:
 
         # 调用LLM
         response = self._llm.invoke(formatted_prompt, False)
-        
+        classification_end_time = time.time()
+        classification_time = classification_end_time - classification_start_time
+        logging.info(f"意图分类耗时统计 - 总耗时: {classification_time:.2f}秒")
         # 解析输出
         try:
             # 尝试直接解析JSON响应
@@ -442,9 +445,12 @@ class IntentRecognizer:
         slot_model = self._get_slot_model(classification)
         if not slot_model:
             raise RuntimeError("未找到匹配的槽位模型")
-        
+        fill_slots_start_time = time.time()
         # 使用LLM进行槽位填充
         filled_slots = self._fill_slots_with_llm(query, classification, slot_model, conversation_context, chat_history, previous_slots)
+        fill_slots_end_time = time.time()
+        fill_slots_time = fill_slots_end_time - fill_slots_start_time
+        logging.info(f"槽位填充耗时统计 - 总耗时: {fill_slots_time:.2f}秒")
         
         # 检查必填槽位是否都已填充
         is_complete, missing_slots = filled_slots.check_required_slots()
@@ -594,18 +600,8 @@ class IntentRecognizer:
             if expected_slot_model is None:
                 # 添加容错处理，应对LLM返回错误分类信息,一级分类跟二级分类错乱
                 # 重新分类
-                classify_start_time = time.time()
                 classification = self._classify_intent(user_input, conversation_context, chat_history, previous_slots)
-                classify_end_time = time.time()
-                classify_time = classify_end_time - classify_start_time
-                # logging.info(f"重新分类耗时: {classify_time:.2f}秒")
-                
-                fill_start_time = time.time()
                 fill_slots = self._fill_slots(user_input, classification, conversation_context, chat_history, previous_slots)
-                fill_end_time = time.time()
-                fill_time = fill_end_time - fill_start_time
-                all_time=fill_end_time-llm_start_time
-                logging.info(f"总耗时：{all_time:.2f}秒,首次槽位+分类:{llm_time:.2f}秒, 重新分类耗时: {classify_time:.2f}秒, 重新槽位填充耗时: {fill_time:.2f}秒")
                 
                 result = {
                     "classification": classification.model_dump(),
@@ -615,13 +611,7 @@ class IntentRecognizer:
                 return result
             elif expected_slot_model.__name__ !=  type(slot_filling).__name__:
                 # 添加容错处理，应对LLM槽位与分类不匹配。重新填充槽位
-                fill_start_time = time.time()
-                slot_filling = self._fill_slots(user_input, classification)
-                fill_end_time = time.time()
-                fill_time = fill_end_time - fill_start_time
-                all_time=fill_end_time-llm_start_time
-                logging.info(f"总耗时：{all_time:.2f}秒,首次槽位+分类:{llm_time:.2f}秒, 重新槽位填充耗时: {fill_time:.2f}秒")
-
+                slot_filling = self._fill_slots(user_input, classification, conversation_context, chat_history, previous_slots)
                 result = {
                     "classification": classification.model_dump(),
                     "slot_filling": slot_filling