generation always repeat the prompt first while training, dont know why?

[jamestang0219]

while training, i found the model always generate prompt first, then generate the think process and answer, do u have any suggestion? maybe the construction of prompt is not suitable?

[zhangqianqianhzlh]

Remove --max_prompt_length in args. Setting max_prompt_length is currently not supported.