Update system prompt of assistant mode to make sure AudioEvals metric is replicable

modeling_minicpmo.py

@@ -717,8 +717,8 @@ class MiniCPMO(MiniCPMOPreTrainedModel):
                 vc_prompt_prefix = "模仿输入音频中的声音特征。"
                 vc_prompt_suffix = "作为助手，你将使用这种声音风格说话。"
             else:
-                vc_prompt_prefix = "Clone the voice in the provided audio prompt."
-                vc_prompt_suffix = "As an assistant, you will speak using this voice style."
+                vc_prompt_prefix = "Use the voice in the audio prompt to synthesize new content."
+                vc_prompt_suffix = "You are a helpful assistant with the above voice style."
 
             if ref_audio is not None:
                 sys_msgs = {"role": "user", "content": [vc_prompt_prefix, ref_audio, vc_prompt_suffix]}