How to split the train/validate/test dataset in the benchmark methods need to trained?

[Fzuerzmj]

Doctor Huang: Thanks for your hard word! I would like to ask how to split the train/validate/test dataset in the benchmark methods that need to be trained for a fair comparison?