Update app.py

app.py

@@ -1,5 +1,6 @@
 import gradio as gr
 import os
+import time
 from huggingface_hub import InferenceClient
 from huggingface_hub import hf_hub_download
 import chatglm_cpp
@@ -33,11 +34,10 @@ def respond(
         return
 
     response = "..."
-    for _ in range(0, 2)
-        import time
+    for _ in range(0, 3):
+        yield response
         time.sleep(1)
         response += " ..."
-        yield response
     
     generation_kwargs = dict(
         max_length=8192,