🐛 Bug

When I add validation loss in progress bar training, tensor object is printed whereas only loss value is expected.

For example :

Epoch 1: 100%|██████████| 5/5 [00:04<00:00, 1.21it/s, loss=82.423, v_num=52, val_loss=tensor(76.4331, dtype=torch.float32)]

Validation loss is added with the following command : self.log('val_loss', loss, prog_bar=True)

I tried self.log('val_loss', loss.item(), prog_bar=True) with no effect.

To Reproduce

The bug is reproductible with the minimal code example (https://github.com/PyTorchLightning/pytorch-lightning/blob/master/pl_examples/bug_report_model.py). See code sample below with validation_step overridden :

Code sample

class TestModel(BoringModel):

    def validation_step(self, batch, batch_idx):
        output = self.layer(batch)
        loss = self.loss(batch, output)
        self.log('val_loss', loss, prog_bar=True)
        return {"x": loss}

Expected behavior

It is expected to only obtain value of validation loss in progress bar and not tensor object.

Environment

• CUDA:
  • GPU:
  • available: False
  • version: None
• Packages:
  • numpy: 1.19.1
  • pyTorch_debug: False
  • pyTorch_version: 1.6.0
  • pytorch-lightning: 0.10.0
  • tqdm: 4.50.1
• System:
  • OS: Darwin
  • architecture:
    • 64bit
  • processor: i386
  • python: 3.8.6
  • version: Darwin Kernel Version 18.7.0: Thu Jun 18 20:50:10 PDT 2020; root:xnu-4903.278.43~1/RELEASE_X86_64