Can a network that uses depthwise separable convolution be converted into an INN and trained using the code？

[Iayerlen]

I encountered this error while trying to convert a structure like MobilenetV3 to INN and train it.
"RuntimeError: Given groups=64, weight of size [64, 64, 3, 3], expected input [64, 64, 160, 160] to have 4096 channels, but got 64 channels installed".
May I ask if your current code supports depthwise separable convolution?Or, in principle, does INN support depthwise separable convolutions?

Thank you very much if you could answer my question

[b1n0]

Unfortunately, the current implementation does not support such convolutions. However it looks like it's easy to add.
We will add this soon in the next update. Thank you for your comment.