search

sw-gong / MoNet

Pytorch reproduction of the paper "Gaussian Mixture Model Convolutional Networks" (CVPR 17)
MIT License
55 stars 9 forks source link

RuntimeError: CUDA out of memory. #1

Open ll961546784 opened 4 years ago

ll961546784 commented 4 years ago

Hello, can you tell me the GPU you used? I use the GTX TITAN X, but an error occurred :Tried to allocate 33.95 GiB (GPU 0; 11.78 GiB total capacity; 17.50 MiB already allocated; 10.62 GiB free; 24.00 MiB reserved in total by PyTorch)

hkim716 commented 4 years ago

I have the same problem here. Does it require that much RAM? 33.95 GiB is too much RAM for running a model. I'm also looking forward to answer from the host.

msseibel commented 4 years ago

When I try to run the image example I run into the same problem but with 50.71GiB. It seems to be independent of the batchsize?!

msseibel commented 4 years ago

Just in case someone else stumples about this: There exists a more recent version directly in the pytorch geometric repository under torch_geometric/nn/conv.

EAkeweje commented 3 years ago

Just in case someone else stumples about this: There exists a more recent version directly in the pytorch geometric repository under torch_geometric/nn/conv.

Hey msseibel, thanks for the information. I found the version you talked about. Were you able to work with that?

rrowaida commented 3 years ago

having the same problem. I tried reducing the batch size. But that introduced a new runtime error saying 'The size of tensor a (22065) must match the size of tensor b (25) at non-singleton dimension 1'. Anyone figured how to work around that?

sure7018 commented 3 years ago

有同样的问题。我尝试减少批量大小。但这引入了一个新的运行时错误，指出“张量 a (22065) 的大小必须与非单维 1 处的张量 b (25) 的大小匹配”。有人想出如何解决这个问题吗？

Have you solved this problem and how to solve it