Update app.py

app.py

@@ -58,9 +58,10 @@ class OrcaChatBot:
             do_sample=True
     )
 
-    response = self.tokenizer.decode(output_ids[0], skip_special_tokens=True)
-    self.update_conversation_history("", response)
-    return response
+        response = self.tokenizer.decode(output_ids[0], skip_special_tokens=True)
+        self.update_conversation_history("", response)
+        return response
+        
 Orca_bot = OrcaChatBot(model, tokenizer)
 
 def gradio_predict(user_message, system_message, max_new_tokens, temperature, top_p, repetition_penalty):