Dear author,

Thank you for your work. I would like to inquire why the model's computation is placed on device 1. When I set two cards to be visible, for example, gpus=0,1, and I set the batch_size to 1, I noticed a strange occurrence: both card 0 and card 1 are running simultaneously. May I ask you for guidance on how to modify this so that the data and model for a batch_size are all on one card using DataParallel?