Update app.py

app.py

@@ -22,6 +22,8 @@ model = LlavaQwen2ForCausalLM.from_pretrained(
     torch_dtype=torch.float16,
     trust_remote_code=True)
 
+model.to('cuda')
+
 class KeywordsStoppingCriteria(StoppingCriteria):
     def __init__(self, keywords, tokenizer, input_ids):
         self.keywords = keywords