How to use Captum on a network with matrix output

I am trying to use Integrated Gradients on a network whose output is of dimension (batchsize, 12, 2).

ig = IntergratedGradients(model)
attr, delta = ig.attribute(input)
then it says that I should specify the target. What parameter should I use for the target? I could not find any reasonable explanation on this.

Assuming you are doing classification, the target is the index of the target class.