kimiyoung/ssl_bad_gan

When running the mnist_trainer.py, aCUDA error: out of memory occurred

XiaoHao-Chen opened this issue · 2 comments

XiaoHao-Chen commented

The error occurred when running pixlecnn++, specifically in the loop section.My graphics card is GTX 1060 6GB.Is it because the memory is too small

xuyujie2 commented

The error occurred when running pixlecnn++, specifically in the loop section.My graphics card is GTX 1060 6GB.Is it because the memory is too small

i had the same question.have you solve it?

XiaoHao-Chen commented

The error occurred when running pixlecnn++, specifically in the loop section.My graphics card is GTX 1060 6GB.Is it because the memory is too small

i had the same question.have you solve it?

I'm sorry I didn't solve it either. But I found that the problem was in the part that called pixelcnn, where each for loop in pixelcnn called pixelcnn's own convolution operation, causing the model to be loaded into the GPU in every for loop, so the GPU was overloaded. Although I know the problem, I still don't have an effective way to solve it. Do you have any good idea?