make demo cpu compatible

app.py

@@ -1,8 +1,11 @@
 from typing import Any
 import gradio as gr
 import itertools
+import torch
 from listener import Listener
 
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
 # pragmatic_listener = Listener(
 #     "pragmatic-programs/pragmatic-ft-listener",
 #     {
@@ -18,13 +21,13 @@ listener = Listener(
     "pragmatic-programs/listener-suffix-idx-300k",
     {
         "do_sample": True,
-        "num_return_sequences": 100,
+        "num_return_sequences": 500,
         "num_beams": 1,
         "temperature": 1,
         "top_p": 0.9,
         "max_new_tokens": 128,
     },
-)
+).to(device)
 
 N_EXAMPLES = 3