Update app.py

app.py

@@ -50,7 +50,7 @@ def get_tokens_and_labels(prompt):
     """
     Given the prompt (text), return a list of tuples (decoded_token, label)
     """
-    inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
+    inputs = tokenizer([prompt], return_tensors="pt")
     outputs = ref_model.generate(
         **inputs,
         max_new_tokens=1000,
@@ -143,6 +143,7 @@ demo = gr.Interface(
     fn=render_output, 
     inputs='text',
     outputs='html',
+    examples=[['context: <As an AI assistant provide helpful, accurate and detailed answers to user questions> question: <Given the fact that Inhaling, or breathing in, increases the size of the chest, which decreases air pressure inside the lungs. If Mona is done with a race and her chest contracts, what happens to the amount of air pressure in her lungs increases or decreases?>']]
 )
 
 demo.launch()