About dataset used for training and test

[xlbjy]

Hi, nice work! It looks like the dataset is very large, with 1.2TB of raw data and 41TB of processed data. Do I need to use all the data if I want to get the results in the paper? Moreover, which GPUs are you using for training? Is it a single-GPU training? What is the approximate duration of the training? Thank you very much.