Update app.py

app.py

@@ -16,12 +16,19 @@ models = {
 def get_client(model):
     return InferenceClient(model=model, token=hf_token)
 
+MAX_HISTORY_LENGTH = 5  # 히스토리에 유지할 최대 대화 수
+
+def truncate_history(history):
+    return history[-MAX_HISTORY_LENGTH:] if len(history) > MAX_HISTORY_LENGTH else history
+
 def respond(message, history, system_message, max_tokens, temperature, top_p, selected_model):
     stop_event.clear()
     client = InferenceClient(model=selected_model, token=hf_token)
     
+    truncated_history = truncate_history(history)
+    
     messages = [{"role": "system", "content": system_message + "\n사용자의 입력에만 직접적으로 답변하세요. 추가 질문을 생성하거나 사용자의 입력을 확장하지 마세요."}]
-    messages.extend([{"role": "user" if i % 2 == 0 else "assistant", "content": m} for h in history for i, m in enumerate(h) if m])
+    messages.extend([{"role": "user" if i % 2 == 0 else "assistant", "content": m} for h in truncated_history for i, m in enumerate(h) if m])
     messages.append({"role": "user", "content": message})
     
     try:
@@ -37,21 +44,24 @@ def respond(message, history, system_message, max_tokens, temperature, top_p, se
                 break
             if chunk:
                 response += chunk
-                # 응답의 시작 부분에서 사용자 입력의 반복을 제거
                 if response.startswith(message):
                     response = response[len(message):].lstrip()
-                yield history + [(message, response)]
+                yield truncated_history + [(message, response)]
         
     except Exception as e:
-        yield history + [(message, f"오류 발생: {str(e)}")]
+        yield truncated_history + [(message, f"오류 발생: {str(e)}")]
 
-def continue_writing(message, chat_history, system_message, max_tokens, temperature, top_p, selected_model):
-    if not chat_history:
-        return "대화 내역이 없습니다."
-    last_user_message = chat_history[-1][0]
-    last_assistant_message = chat_history[-1][1]
-    prompt = f"이전 대화를 계속 이어서 작성해주세요. 이전 응답: {last_assistant_message}"
-    return respond(prompt, chat_history, system_message, max_tokens, temperature, top_p, selected_model)
+def continue_writing(history, system_message, max_tokens, temperature, top_p, model):
+    if not history:
+        return [("시스템", "대화 내역이 없습니다.")]
+    
+    truncated_history = truncate_history(history)
+    last_user_message = truncated_history[-1][0]
+    last_assistant_message = truncated_history[-1][1]
+    
+    prompt = f"이전 대화를 간단히 요약하고 이어서 작성해주세요. 마지막 응답: {last_assistant_message[:100]}..."
+    
+    return respond(prompt, truncated_history[:-1], system_message, max_tokens, temperature, top_p, model)
 
 def stop_generation():
     stop_event.set()