In ConvnetBuilder class of fastai the init method has the below lines:
cut-=xtra_cut
layers = cut_model(f(pretrained), cut)
cut is intialized to 8(for resnet34) so cut eventually becomes 8-0 = 8(xtra_cut=0)
so layers variable contains the first 8 layers of resnet34.
Why is fast ai taking only the first 8 layers of resnet34.