Update app.py

app.py

@@ -19,13 +19,16 @@ model = AutoModelForCausalLM.from_pretrained(
 @spaces.GPU(duration=300)
 def generate_response(query):
     # 🚀 generate_response 함수 안에서 매번 로드
-    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-    model = AutoModelForCausalLM.from_pretrained(
-        model_name,
-        torch_dtype=torch.float16,
-        device_map="auto",  # ✅ 중요: 자동으로 GPU 할당
-        trust_remote_code=True,
-    )
+    # tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+    # model = AutoModelForCausalLM.from_pretrained(
+    #     model_name,
+    #     torch_dtype=torch.float16,
+    #     device_map="auto",  # ✅ 중요: 자동으로 GPU 할당
+    #     trust_remote_code=True,
+    # )
+    tokenizer = AutoTokenizer.from_pretrained("dasomaru/gemma-3-4bit-it-demo")
+    model = AutoModelForCausalLM.from_pretrained("dasomaru/gemma-3-4bit-it-demo")
+    model.to("cuda")    
 
     # 1. 검색
     top_k = 5