🐛 Describe the bug A spurious "Grad strides do not match bucket view strides" warning is shown for 1x1 convolution running on a tensor in channels_last memory format in DistributedDataParallel model ...