This model cannot train on GPU? about lernapparat HOT 4 OPEN

@t-vi Is there a codebase for training available as well?

Thanks!

from lernapparat.

Hey! Can you share your training code? That might help debug it.

from lernapparat.

So the place that I immediately saw is the kernel in the blur layer, you could just do .contiguous() before passing it in. You can also use this recipe to find the exact place where the computation fails and then enforce contiguity.

from lernapparat.

Hi,
When I attempt to train is model on GPU, it reports the error below, while it can be trained on CPU.

RuntimeError: Assertion dWeight.isContiguous()' failed. at /opt/conda/conda-bld/pytorch_1532581333611/work/aten/src/THCUNN/generic/SpatialDepthwiseConvolution.cu:140`

Do you have any ideas about it?
pytorch 0.4.1
cuda 9.0

Have you solved this problem?
I also attempt to train the frozen pre-trained stylegan and a new encoder(modiffied d_basic) from scratch, the code like this:
latent = torch.randn(args.batch_size, 512, device=device)
fake_hr_imgs = g_all(latent)
re_latent = d_basic(fake_hr_imgs)

recons_latent_loss = l2_loss(re_latent, latent.detach(), reduct='mean')
recons_latent_loss.backward()

but also meet the issue:
RuntimeError: Assertion dWeight.isContiguous()' failed. at /opt/conda/conda-bld/pytorch_1532581333611/work/aten/src/THCUNN/generic/SpatialDepthwiseConvolution.cu:140`

Meanwhile, I have tried the method "At the very top of your program, before you import anything (and in particular PyTorch), insert

import os
os.environ['CUDA_LAUNCH_BLOCKING'] = "1""

But there is still the same issue like above

pytorch 0.4.1
cuda 9.0

from lernapparat.