auto_scale_batch_size fails with datamodule in pl==1.2.*, succeeds in pl==1.1.8

[colllin]

🐛 Bug

Running trainer = pl.Trainer(auto_scale_batch_size=True); trainer.tune(model, datamodule=dm) succeeds in pl==1.1.8, but fails in pl==1.2.* (tested both 1.2.0 and 1.2.1) with error:

Traceback (most recent call last):
  File "train.py", line 42, in <module>
    trainer.tune(model, datamodule=dm)
  File "/home/ubuntu/.local/share/virtualenvs/__project__/lib/python3.8/site-packages/pytorch_lightning/trainer/trainer.py", line 1062, in tune
    self.tuner.tune(model, train_dataloader, val_dataloaders, datamodule)
  File "/home/ubuntu/.local/share/virtualenvs/__project__/lib/python3.8/site-packages/pytorch_lightning/tuner/tuning.py", line 46, in tune
    self.scale_batch_size(
  File "/home/ubuntu/.local/share/virtualenvs/__project__/lib/python3.8/site-packages/pytorch_lightning/tuner/tuning.py", line 104, in scale_batch_size
    return scale_batch_size(
  File "/home/ubuntu/.local/share/virtualenvs/__project__/lib/python3.8/site-packages/pytorch_lightning/tuner/batch_size_scaling.py", line 79, in scale_batch_size
    raise MisconfigurationException(f'Field {batch_arg_name} not found in both `model` and `model.hparams`')
pytorch_lightning.utilities.exceptions.MisconfigurationException: Field batch_size not found in both `model` and `model.hparams` 

Please reproduce using the BoringModel

https://colab.research.google.com/drive/1vgPLCwLg7uACtb3fxVp-t-__NtZ3onsD?usp=sharing

Expected behavior

Prior to pl==1.2.0, it successfully detected and tuned the dm.batch_size property.

Environment

• CUDA:
  • GPU:
    • Tesla T4
  • available: True
  • version: 10.1
• Packages:
  • numpy: 1.19.5
  • pyTorch_debug: False
  • pyTorch_version: 1.7.1+cu101
  • pytorch-lightning: 1.2.1
  • tqdm: 4.41.1
• System:
  • OS: Linux
  • architecture:
    • 64bit
  • processor: x86_64
  • python: 3.7.10
  • version: Proposal for help #1 SMP Thu Jul 23 08:00:38 PDT 2020

[akihironitta]

@colllin Thank you for reporting the issue. I confirmed the bug in 1.2.* which didn't happen in <1.2. I think this issue should be addressed in #5968 by @awaelchli.

[rohitgr7]

closing this then. duplicate: #5967