Could you please provide the training end eval instruction code to reproduce the result of the paper? for more detail in the readme page,thx