Out of memory when using resnet

[Willy0919]

When I used resnet20 on cifar10, after several iter, it raise the error:
grads = torch.autograd.grad(loss_meta, meta_net.params(), create_graph=True)
File "/usr/local/lib/python3.5/dist-packages/torch/autograd/init.py", line 145, in grad
inputs, allow_unused)
RuntimeError: CUDA out of memory. Tried to allocate 14.50 MiB (GPU 0; 10.92 GiB total capacity; 10.04 GiB already allocated; 3.50 MiB free; 127.91 MiB cached)

but baseline runs normally.

[hansbu]

what is the input size of your resnet20?

[Vickeyhw]

I also have the same problem when I use resnet32 on cifar100. Have you solved this problem?