Update app.py

app.py

@@ -4,6 +4,7 @@ from flask import Flask, request, Response
 from transformers import AutoTokenizer, GPT2LMHeadModel
 from dicttoxml import dicttoxml
 import traceback
+from threading import Lock  # ← 추가
 
 app = Flask(__name__)
 
@@ -22,6 +23,9 @@ except Exception as e:
     print(f"데이터셋 로드 에러: {e}")
     knowledge_list = []
 
+# --- 3. 동시 요청 제한용 Lock ---
+request_lock = Lock()
+
 def find_relevant_context(query, top_n=2):
     """질문과 관련된 지식 데이터 문장 최대 top_n개 반환"""
     query_words = query.replace(" ", "").lower()
@@ -60,7 +64,7 @@ def ask_sayknow(query):
             gen_ids = model.generate(
                 input_ids,
                 attention_mask=attention_mask,
-                max_new_tokens=200,
+                max_new_tokens=100,  # ← 200 -> 100으로 줄여 워커 점유 시간 단축
                 min_length=5,
                 repetition_penalty=1.3,
                 do_sample=True,
@@ -90,14 +94,18 @@ def ask_sayknow(query):
         traceback.print_exc()
         return f"내부 오류: {str(e)}"
 
-# --- 3. API (XML 응답) ---
+# --- 4. API (XML 응답) ---
 @app.route('/chatapi.html', methods=['GET'])
 @app.route('/index.html', methods=['GET'])
 def chat_api():
     query = request.args.get('askdata', '')
     if not query:
         result = {"status": "error", "message": "No data"}
-    else:
+        xml_output = dicttoxml(result, custom_root='SayknowAPI', attr_type=False)
+        return Response(xml_output, mimetype='text/xml')
+
+    # ← Lock으로 요청 순차 처리
+    with request_lock:
         try:
             answer = ask_sayknow(query)
             result = {