why does eval report out of memory while n_scale is set the same in train and eval?

NVIDIA / semantic-segmentation

Nvidia Semantic Segmentation monorepo

BSD 3-Clause "New" or "Revised" License

1.76k stars 388 forks source link

why does eval report out of memory while n_scale is set the same in train and eval? #184

Open xu19971109 opened 1 year ago

xu19971109 commented 1 year ago

N_scale is set the same in train and eval, batch 2 or batch 4 can be used in training, but eval will report out of memory error after batch 2 or batch 4 is run.