Question about OOM

[jizaza]

Sorry to bother you again. When I use a single V100 (16384MiB) to train this model with patch size 389 and full size 1024, I find it can't work. And I also tried the methods you mentioned in Readme.md, it still has the error "out of memory". Do you have any more ideas about this? Thank you very much!!!!!