Upload app.py

app.py

@@ -82,7 +82,7 @@ class ChatbotArguments:
         },
     )
     max_new_tokens: Optional[int] = field(
-        default=1000,
+        default=200,
         metadata={
             "help": "maximum number of generated tokens"
         },
@@ -117,7 +117,7 @@ def main():
         model_args,
         tune_strategy='none',
         ds_config=ds_config,
-        device=pipeline_args.device,
+        device="cpu",#pipeline_args.device,
     )
 
     # We don't need input data, we will read interactively from stdin