🐛 Bug

I am following the new memory_format tutorial:
https://pytorch.org/tutorials/intermediate/memory_format_tutorial.html

When running it on a custom convolutional network, I get:

`conv2d` got channels_last input, but output is not channels_last: torch.Size([8, 32, 256, 512]) (4194304, 131072, 512, 1) cuda:0 torch.float32
`conv2d` inputs are:
 (4194304, 1, 16384, 32) torch.Size([8, 32, 256, 512]) cuda:0 torch.float32
 (25, 1, 5, 1) torch.Size([32, 1, 5, 5]) cuda:0 torch.float32
 None
 <class 'tuple'>
     1
     1
 <class 'tuple'>
     4
     4
 <class 'tuple'>
     2
     2
 32

despite conv2d being said to support channels_last property.

To Reproduce

Steps to reproduce the behavior:

1. Follow the memory_format tutorial on a convolutional model containing ???
2. ???

Expected behavior

The conv2d should keep the channels_last property

Environment

PyTorch version: 1.5.0+cu101
Is debug build: No
CUDA used to build PyTorch: 10.1

OS: CentOS Linux 7 (Core)
GCC version: (GCC) 8.3.0
CMake version: Could not collect

Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: 10.1.243
GPU models and configuration: GPU 0: Tesla V100-SXM2-32GB
Nvidia driver version: 418.67
cuDNN version: Could not collect

Versions of relevant libraries:
[pip3] numpy==1.18.2
[pip3] pytorch-lightning==0.7.4
[pip3] torch==1.5.0+cu101
[pip3] torchvision==0.6.0+cu101

Additional context

cc @ngimel @csarofeen @ptrblck @VitalyFedyunin @jamesr66a