How does AutoKeras select Models?

[takeofuture]

Question

I have tested Image Classifier with MNIST with 60000 Training Data + 10000 Validation Data
and I got one model.

Then I reduced MNIST training data 6000 and test data 1000, then I still get
the exactly the same model as 60,000 + 10,000 data

I tried fashion_mnist (60000 + 10000 , (28,28)
But I got still exactly the same model selected

How does Autokeras Image Classifier search and select model?
I expected to reduce the layer or so with the small number of data.
Does Autokears or Image Classfier automatically adjust the number of layer based on the number of data, image size etc?