Update app.py

app.py

@@ -56,7 +56,12 @@ def generate(
     outputs = []
     for text in streamer:
         outputs.append(text)
-        yield "".join(outputs)
+        generated_text = "".join(outputs)
+        if "<s>" in generated_text:
+            yield generated_text[:generated_text.index("<s>")+3]
+            break
+        yield generated_text
+
 
 chat_interface = gr.ChatInterface(
     fn=generate,