Update app.py

app.py

@@ -11,7 +11,7 @@ from transformers import AutoTokenizer, AutoModelForSequenceClassification, Auto
 
 #initing
 app = Flask(__name__)
-VERSION = '1.0 build115'
+VERSION = '1.0 build117'
 app.config['JSON_AS_ASCII'] = False
 limiter = Limiter(app=app, key_func=get_remote_address, default_limits=["5/minute"], storage_uri="memory://",)
 
@@ -150,7 +150,10 @@ def chitchatRu():
             return {"status": "error", "details": {"error_code": 101, "error_details": "No text provided"}}, 400
         
         inputs = chct_t.encode(text, padding=True, truncation=True, return_tensors="pt")
-        generated_ids = chct_m.generate(input_ids=inputs)
+        generated_ids = chct_m.generate(
+            input_ids=inputs,
+            use_cache=False
+        )
         answer = chct_t.decode(generated_ids[0], skip_special_tokens=True)
         return {"status": "pass", "answer": answer}, 200
     except Exception as e: