concat confusion

[Seth Neiman]

In some error calcs during validation, i'm working the labels and predictions from a network that has a final layer with 2 identical paths through a concat.

Batch size is 16, using cudnn.

The labels are in a tensor that is 16x200 - 1 set of 100 classification probabilities for each path through the concat. This was done to match my understanding of nn.Concat - that its simply concatenates the outputs of parallel path. The network trains without complaining.

Question #1 - is this correct?

The returned predictions are in a tensor that is 32x100. Obviously, its all there - but I need to understand how it is laid out - assuming my above understanding is valid.

Do the prediction values alternate between concat paths? or is it 16 from one, followed by 16 from the other.

thanks,

seth

[Seth Neiman]

Never mind - wrong dim in provided model.