Update app.py

app.py

@@ -41,11 +41,10 @@ histories = []
 
 
 def generate_text(messages):
-    if is_hugging_face:#need everytime initialize for ZeroGPU
-        model = AutoModelForCausalLM.from_pretrained(
+    model = AutoModelForCausalLM.from_pretrained(
                 model_id, token=huggingface_token ,torch_dtype=dtype,device_map=device
             )
-        model.to(device)
+    model.to(device)
     question = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     question = tokenizer(question, return_tensors="pt").to(device)