Use DEVICE variable for input tensor allocation

app.py

@@ -182,7 +182,7 @@ with gr.Blocks() as demo:
             padding=True,
             return_tensors="pt",
         )
-        inputs = inputs.to("cuda")
+        inputs = inputs.to(DEVICE)
 
         generated_ids = model.generate(**inputs, max_new_tokens=max_new_tokens)
         generated_ids_trimmed = [