Update app.py

app.py

@@ -118,34 +118,41 @@ async def generate_response(message):
         {"role": "user", "content": f"사용자 질문: {user_input}\n\n검색 결과:\n{search_results}"}
     ]
     
-    # 이전 대화 히스토리 추가
-    messages.extend(conversation_history)
+    # 이전 대화 히스토리 추가 (최대 5개까지만)
+    for i in range(min(len(conversation_history), 10), 0, -2):
+        messages.append(conversation_history[-i])
+        if i > 1:
+            messages.append(conversation_history[-i+1])
     
     logging.debug(f'Messages to be sent to the model: {messages}')
     
     loop = asyncio.get_event_loop()
-    response = await loop.run_in_executor(None, lambda: hf_client.chat_completion(
-        messages, max_tokens=1000, stream=True, temperature=0.7, top_p=0.85))
-    
-    full_response = []
-    for part in response:
-        logging.debug(f'Part received from stream: {part}')
-        if part.choices and part.choices[0].delta and part.choices[0].delta.content:
-            full_response.append(part.choices[0].delta.content)
-    
-    full_response_text = ''.join(full_response)
-    logging.debug(f'Full model response: {full_response_text}')
-    
-    # 대화 히스토리 업데이트
-    conversation_history.append({"role": "user", "content": user_input})
-    conversation_history.append({"role": "assistant", "content": full_response_text})
-    
-    # 대화 히스토리 길이 제한 (예: 최근 10개 메시지만 유지)
-    conversation_history = conversation_history[-10:]
-    
-    logging.debug(f'Conversation history updated: {conversation_history}')
-    
-    return f"{user_mention}, {full_response_text}"
+    try:
+        response = await loop.run_in_executor(None, lambda: hf_client.chat_completion(
+            messages, max_tokens=1000, stream=True, temperature=0.7, top_p=0.85))
+        
+        full_response = []
+        for part in response:
+            logging.debug(f'Part received from stream: {part}')
+            if part.choices and part.choices[0].delta and part.choices[0].delta.content:
+                full_response.append(part.choices[0].delta.content)
+        
+        full_response_text = ''.join(full_response)
+        logging.debug(f'Full model response: {full_response_text}')
+        
+        # 대화 히스토리 업데이트
+        conversation_history.append({"role": "user", "content": user_input})
+        conversation_history.append({"role": "assistant", "content": full_response_text})
+        
+        # 대화 히스토리 길이 제한 (최근 10개 메시지만 유지)
+        conversation_history = conversation_history[-10:]
+        
+        logging.debug(f'Conversation history updated: {conversation_history}')
+        
+        return f"{user_mention}, {full_response_text}"
+    except Exception as e:
+        logging.error(f"Error in generate_response: {e}")
+        return f"{user_mention}, 죄송합니다. 응답을 생성하는 중 오류가 발생했습니다. 다시 시도해 주세요."
 
 if __name__ == "__main__":
     # Discord 클라이언트 실행