Sanity check LightningModule without create trainer

[gungui98]

🐛 Bug

Hi, I want to do a sanity check to a LightningModule, in the previous version 1.3.8 everything works fine with the following code, but with the latest version 1.4.5, the test fails because of trainer is NoneType, I do not want to create a whole trainer just to assert some properties of my model.

To Reproduce

import torch
from torch import nn
from torch.nn import functional as F
import pytorch_lightning as pl
import torchmetrics
from torchvision.models import resnet50


class MyResNet(nn.Module):
    def __init__(self, num_classes, pretrained=True):
        super(MyResNet, self).__init__()
        model = resnet50(pretrained)
        self.num_ftrs = model.fc.in_features  # 512

        self.shared = nn.Sequential(*list(model.children())[:-1])
        self.target = torch.nn.Sequential(nn.Linear(self.num_ftrs, num_classes))

    def forward(self, x):
        x = self.shared(x)
        x = torch.squeeze(x)  # dimensions of input of size 1 removed.
        return self.target(x)


class Classifier(pl.LightningModule):

    def __init__(self, num_classes):
        super().__init__()
        self.save_hyperparameters()
        self.backbone = MyResNet(num_classes)
        self.train_accuracy = torchmetrics.Accuracy()

    def forward(self, x):
        embedding = self.backbone(x)
        return embedding

    def training_step(self, batch, batch_idx):
        x, y, _ = batch
        y_hat = self.backbone(x.float())
        loss = F.cross_entropy(y_hat, y)
        self.log('train_loss', loss, on_epoch=True)
        return loss

    def validation_step(self, batch, batch_idx):
        x, y, _ = batch
        y_hat = self.backbone(x.float())
        loss = F.cross_entropy(y_hat, y)
        acc = self.train_accuracy(y_hat, y)
        self.log('valid_loss', loss, on_step=True)
        self.log('accuracy', acc, on_epoch=True, prog_bar=True)
        return acc

    def test_step(self, batch, batch_idx):
        x, y, _ = batch
        y_hat = self.backbone(x.float())
        loss = F.cross_entropy(y_hat, y)
        self.log('test_loss', loss)
        return loss


if __name__ == '__main__':
    model = Classifier(10, )
    batch = (torch.ones(16, 3, 30, 20), torch.zeros(16).long(), None)
    print(model.training_step(batch, 0))

Expected behavior

the code should return a scalar loss instead of

AttributeError: 'NoneType' object has no attribute '_results'

Environment

• CUDA:
  • GPU:
    • GeForce RTX 2070 with Max-Q Design
  • available: True
  • version: 10.2
• Packages:
  • numpy: 1.21.1
  • pyTorch_debug: False
  • pyTorch_version: 1.7.1
  • pytorch-lightning: 1.4.5
  • tqdm: 4.62.0
• System:
  • OS: Windows
  • architecture:
    • 64bit
    • WindowsPE
  • processor: Intel64 Family 6 Model 158 Stepping 10, GenuineIntel
  • python: 3.8.8
  • version: 10.0.19041

Additional context

[carmocca]

You didn't provide a stacktrace but I assume this is where it's failing:

https://github.com/PyTorchLightning/pytorch-lightning/blob/645eabe11055b210f0865aad142d1c26b7827012/pytorch_lightning/core/lightning.py#L407

The message for this error has improved in master:

https://github.com/PyTorchLightning/pytorch-lightning/blob/6892d533ea1c743f7e05171846a28e685db85f51/pytorch_lightning/core/lightning.py#L394-L398

This is not supported right now, there's a feature request for it in #8509. You'll have to avoid the log call until then.

Closing, feel free to comment your thoughts on the feature request.