sinAshish/Multi-Scale-Attention

out of memory

rocky-DJY opened this issue · 3 comments

Why 22G+ of CUDA memory is occupied during training, when the batchsize is 2

so,adjust the batchsize to the 1 ........and,exactly,i follow this paper github repository by occasion。tomorrow will do a presentation,no any idea。。。

I set batchsize to 1 but I also encountered the question of out of memory... I wonder why this will happen?

I'm just getting started in the field, can you share the steps to run this project?