f93beff338925cc1bf1b3ff1b32a2c440a4c9427,tests/_tests_scripts/z_mvp_mnist_gan.py,CustomRunner,_handle_batch,#CustomRunner#Any#,18

Before Change

        Docs.
        
        images, _ = batch
        images = images.view(images.size(0), -1)
        bs = images.shape[0]
        z = torch.randn(bs, 128).to(self.device)
        generated_images = self.model["generator"](z)

        // generator step
        // predictions & labels
        generated_labels = torch.ones(bs, 1).to(self.device)
        generated_pred = self.model["discriminator"](generated_images)

        // loss
        loss_generator = F.binary_cross_entropy(
            generated_pred, generated_labels
        )
        self.state.batch_metrics["loss_generator"] = loss_generator

        // discriminator step
        // real
        images_labels = torch.ones(bs, 1).to(self.device)

After Change

        )

        // Decode them to fake images
        generated_images = self.model["generator"](
            random_latent_vectors
        ).detach()
        // Combine them with real images
        combined_images = torch.cat([generated_images, real_images])

        // Assemble labels discriminating real from fake images
        labels = torch.cat(
            [torch.ones((batch_size, 1)), torch.zeros((batch_size, 1))]

Instances