Update app.py

app.py

@@ -105,8 +105,7 @@ with gr.Blocks() as demo:
                 state['past_key_values'] = outputs.past_key_values
                 state['past_ids'] = outputs.sequences[:, :-1]
                 response = infer.processor.decode(outputs.sequences[0, inputs.input_ids.size(1):], skip_special_tokens=True)
-                print(response)
-                return response, state                
+                yield response, state                
                 
             def gr_chatinterface_chatbot_clear_fn():
                 return {}, {}, 0, 0