When I use the official sample to carry out back propagation manually, I make mistakes. First, there is no optimizer, and second, there is no attribute in the image

[Hou-jing]

import torch
from torch import Tensor
from pytorch_lightning import LightningModule
class Generator:
    def __init__(self):
        pass
    def forward(self):
        pass

class Discriminator:
    def __init__(self):
        pass
    def forward(self):
        pass
class SimpleGAN(LightningModule):
    def __init__(self):
        super().__init__()
        self.G = Generator()
        self.D = Discriminator()

        # Important: This property activates manual optimization.
        self.automatic_optimization = False

    def sample_z(self, n) -> Tensor:
        sample = self._Z.sample((n,))
        return sample

    def sample_G(self, n) -> Tensor:
        z = self.sample_z(n)
        return self.G(z)

    def training_step(self, batch, batch_idx):
        # Implementation follows the PyTorch tutorial:
        # https://pytorch.org/tutorials/beginner/dcgan_faces_tutorial.html
        g_opt, d_opt = self.optimizers()

        X, _ = batch
        batch_size = X.shape[0]

        real_label = torch.ones((batch_size, 1), device=self.device)
        fake_label = torch.zeros((batch_size, 1), device=self.device)

        g_X = self.sample_G(batch_size)

        ##########################
        # Optimize Discriminator #
        ##########################
        d_x = self.D(X)
        errD_real = self.criterion(d_x, real_label)

        d_z = self.D(g_X.detach())
        errD_fake = self.criterion(d_z, fake_label)

        errD = errD_real + errD_fake

        d_opt.zero_grad()
        self.manual_backward(errD)
        d_opt.step()

        ######################
        # Optimize Generator #
        ######################
        d_z = self.D(g_X)
        errG = self.criterion(d_z, real_label)

        g_opt.zero_grad()
        self.manual_backward(errG)
        g_opt.step()

        self.log_dict({"g_loss": errG, "d_loss": errD}, prog_bar=True)

    def configure_optimizers(self):
        g_opt = torch.optim.Adam(self.G.parameters(), lr=1e-5)
        d_opt = torch.optim.Adam(self.D.parameters(), lr=1e-5)
        return g_opt, d_opt
batch=torch.randn(3,2)
batch_idx=torch.ones(3)
SimpleGAN().training_step(batch,batch_idx)

[akihironitta]

Hi @Hou-jing!

pl.LightningModule is supposed to be used with pl.Trainer. Would you mind trying it again with the trainer?

[akihironitta]

This example (adapted from the official examples) in my repo could be useful for you: https://github.com/akihironitta/gist/blob/master/pl_gan_manual_optimization/main.py

[Hou-jing]

When I upgraded the version to the latest version, I solved this problem。
However, my running speed has been greatly affected. I used to have an epoch every 8 minutes, but now it has been delayed for a long time, and the data can't be loaded. I don't know why
And this is the code
[https://colab.research.google.com/drive/1dCP7-1xK48-PohGc8-RKx3Ne2HWd4Jkq#scrollTo=frTD9xWvBEUT]

[akihironitta]

@Hou-jing Cannot access to your Colab notebook...

I'd suggest to profile your lightningmodule to identify the cause of the problem.
https://pytorch-lightning.readthedocs.io/en/latest/advanced/profiler.html

[akihironitta]

If you still have the issue, could you create another GitHub discussion? Each discussion/issue should be about only one topic.

[Hou-jing]

I think ，Because the loss is calculated twice, the time is delayed