diff --git a/toolbench/inference/LLM/tool_llama_lora_model.py b/toolbench/inference/LLM/tool_llama_lora_model.py
index a04b1fc..2c5df1b 100644
--- a/toolbench/inference/LLM/tool_llama_lora_model.py
+++ b/toolbench/inference/LLM/tool_llama_lora_model.py
@@ -93,13 +93,15 @@ def parse(self,functions,process_id,**args):
         self.time = time.time()
         conversation_history = self.conversation_history
         prompt = ''
+        prompt_parts = []
         for message in conversation_history:
             role = roles[message['role']]
             content = message['content']
             if role == "System" and functions != []:
                 content = process_system_message(content, functions)
-            prompt += f"{role}: {content}\n"
-        prompt += "Assistant:\n"
+            prompt_parts.append(f"{role}: {content}")
+        prompt = '\n'.join(prompt_parts) + "\nAssistant:\n"
+
         if functions != []:
             predictions = self.prediction(prompt)
         else:
@@ -138,4 +140,4 @@ def parse(self,functions,process_id,**args):
 
     llm.change_messages(messages)
     output = llm.parse(functions=functions)
-    print(output)
\ No newline at end of file
+    print(output)