Closed f-dangel closed 3 years ago
Passing a torch.float64 input to unfoldNd.unfoldNd raises an exception in the underlying convolution, as the one-hot weight is always created in the default data type.
torch.float64
unfoldNd.unfoldNd
Reproduced in 086ed13.
086ed13
Passing a
torch.float64
input tounfoldNd.unfoldNd
raises an exception in the underlying convolution, as the one-hot weight is always created in the default data type.Reproduced in
086ed13
.