Misleading InstanceNorm documentation?

[nikitati]

InstanceNorm documentation says:

Given an array with N dimensions, call the N-1th the channel dimension. (For
a batch of feature vectors this is just the data dimension, for WHCN images
it's the usual channel dimension.)

following with an example:

m = Chain(
  Dense(28^2, 64),
  InstanceNorm(64, relu),
  Dense(64, 10),
  InstanceNorm(10),
  softmax)

as far as I understand, the docs say that the layer can in principle be used on a batch of feature vectors (with shape DxN). However, the implementation actually does not support this behavior and throws an exception if it encounters 2D data:

 ndims(x) > 2 ||
    error("InstanceNorm requires at least 3 dimensions. With 2 dimensions an array of zeros would be returned")

This makes it impossible to reproduce the example. So if I interpret it corretly, the docs and the implementation disagree.