Update app.py

app.py

@@ -61,7 +61,7 @@ if uploaded_file is not None:
         padding=True,
         return_tensors="pt",
     )
-    inputs = inputs.to("cuda")
+    inputs = inputs.to("cpu")
 
     # Inference: Generation of the output
     generated_ids = model.generate(**inputs, max_new_tokens=128)