Lornatang/UGATIT-PyTorch

Cuda out of Memory

Opened this issue · 1 comments

sinagh72 commented

Hi,

I tried to run the code using 1050ti, but I get this error:

File "E:\UGATIT-PyTorch-master2\ugatit_pytorch\model.py", line 31, in forward
out = self.rho.expand(x.shape[0], -1, -1, -1) * out_in + (1 - self.rho.expand(x.shape[0], -1, -1, -1)) * out_ln
RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.44 GiB already allocated; 0 bytes free; 3.50 GiB reserved in total by PyTorch)

Is there any way to solve this issue?

kirill-ionkin commented

Hi,

I tried to run the code using 1050ti, but I get this error:

File "E:\UGATIT-PyTorch-master2\ugatit_pytorch\model.py", line 31, in forward out = self.rho.expand(x.shape[0], -1, -1, -1) * out_in + (1 - self.rho.expand(x.shape[0], -1, -1, -1)) * out_ln RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 3.44 GiB already allocated; 0 bytes free; 3.50 GiB reserved in total by PyTorch)

Is there any way to solve this issue?

@sinagh72
Hi, i'm not an expert. but i think you need more powerful(more gpu memory) gpu on your pc, because forward pass need a lot of resources.

I think, one of most usefull solution is just training on multiple gpu's, but of course you need multiple gpu's to do that.