update app.py to use cuda

app.py

@@ -4,7 +4,7 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 # 加载模型和tokenizer
 model_name = "StevenChen16/llama3-8b-Lawyer"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name).to('cuda')
 
 def generate_response(input_text):
     inputs = tokenizer(input_text, return_tensors="pt")