Update api.py

api.py

@@ -150,7 +150,7 @@ print("모델 가중치 로드 완료!")
 
 def generate_text_sample(model, prompt, max_len=100, max_gen=98,
                          temperature=0.8, top_k=55, top_p=0.95, min_len=12):
-    model_input = text_to_ids(f"<start> {prompt} <sep>")
+    model_input = text_to_ids(f"<start> {prompt}")
     model_input = model_input[:max_len]
     generated = list(model_input)
 
@@ -234,13 +234,13 @@ def respond(input_text):
         return response
 
     # 메모리 관련 부분 싹 제거하고, 단순 프롬프트 생성
-    full_prompt = f"현재 질문:\n{input_text}\n\n응답:"
+    full_prompt = f"현재 질문:\n{input_text}\n\n<sep>"
 
     for _ in range(3):  # 최대 3번 재시도
         full_response = generate_text_sample(model, full_prompt)
 
         if "응답:" in full_response:
-            response = full_response.split("응답:")[-1].strip()
+            response = full_response.split("<sep>")[-1].strip()
         else:
             response = full_response.strip()