Deep Convolutional Generative Adversarial Networks training question

[kli-casia]

In https://gluon.mxnet.io/chapter14_generative-adversarial-networks/dcgan.html#

with autograd.record():
            # train with real image
            output = netD(data).reshape((-1, 1))
            errD_real = loss(output, real_label)
            metric.update([real_label,], [output,])

            # train with fake image
            fake = netG(latent_z)
            output = netD(fake.detach()).reshape((-1, 1))
            errD_fake = loss(output, fake_label)
            errD = errD_real + errD_fake
            errD.backward()
            metric.update([fake_label,], [output,])

I am confused by this line of code

fake = netG(latent_z)

In my opinion, this line is used to generate fake data, and netG should in predict mode, am I right?
If I am right, since fake = netG(latent_z) is in the context of with autograd.record():, which imply that netG is in train mode . (https://gluon.mxnet.io/chapter03_deep-neural-networks/mlp-dropout-gluon.html#Integration-with-autograd.record)