While trying to implement Improved WGANs I bumped into another error. I wanted to replace MaxPool layers with strided convolutions, but it seems there is a problem with CUDNN when computing the gradient penalty. On CPU strides work fine.
/home/hartmank/anaconda2/lib/python2.7/site-packages/torch/autograd/variable.pyc in backward(self, gradient, retain_graph, create_graph, retain_variables)
150 Defaults to False, unless ``gradient`` is a volatile Variable.
151 """
--> 152 torch.autograd.backward(self, gradient, retain_graph, create_graph, retain_variables)
153
154 def register_hook(self, hook):
/home/hartmank/anaconda2/lib/python2.7/site-packages/torch/autograd/__init__.pyc in backward(variables, grad_variables, retain_graph, create_graph, retain_variables)
96
97 Variable._execution_engine.run_backward(
---> 98 variables, grad_variables, retain_graph)
99
100
RuntimeError: CUDNN_STATUS_NOT_SUPPORTED. This error may appear if you passed in a non-contiguous input.
GPU:
CPU:
compiled pytorch from source with CUDA 8.0.61 and CUDNN 6.0.21