thks for your work.
as the figure above, after a few iters (5000) of traning step, the output of the mViT change to 0,0,0,0,...,1.
i can't understand how it happened. it seems like it make the depth split to 2 widths.
did you meet this ? any suggestion would help me a lot,~
thks for your work. as the figure above, after a few iters (5000) of traning step, the output of the mViT change to 0,0,0,0,...,1. i can't understand how it happened. it seems like it make the depth split to 2 widths.
did you meet this ? any suggestion would help me a lot,~