Batch size finder is not working if batch_size is specified in LightningDataModule

[maxjeblick]

🐛 Bug

The batch size finder won't work if the batch size is specified in the LightningDataModule, only (it is natural to define it there).

An instance of a LightningModule always has the attribute hparams; the batch_size finder raises a MisconfigurationException if batch_size isn't found there.

Please reproduce using the BoringModel and post here

https://colab.research.google.com/drive/1gruW3UwitVijzkhcUYIzHlIpoIB1shzt?usp=sharing

Expected behavior

Batch size finder works.

Environment

• CUDA:
  • GPU:
    • Tesla T4
  • available: True
  • version: 10.1
• Packages:
  • numpy: 1.18.5
  • pyTorch_debug: False
  • pyTorch_version: 1.6.0+cu101
  • pytorch-lightning: 0.10.0
  • tqdm: 4.41.1
• System:
  • OS: Linux
  • architecture:
    • 64bit
  • processor: x86_64
  • python: 3.6.9
  • version: Proposal for help #1 SMP Thu Jul 23 08:00:38 PDT 2020

[SkafteNicki]

That is strange, we have a test for that specific case:
https://github.com/PyTorchLightning/pytorch-lightning/blob/7c4f80a1afe3d7b0f1e9ee834aacaf8439195cdf/tests/trainer/test_trainer_tricks.py#L233-L270
It must be the logic in lightning_hasattr, that is broken, because it should explicit check if dataloader has a field called batch_size:
https://github.com/PyTorchLightning/pytorch-lightning/blob/7c4f80a1afe3d7b0f1e9ee834aacaf8439195cdf/pytorch_lightning/utilities/parsing.py#L177-L197

[maxjeblick]

In the test mentioned above, the batch_size is also specified inside the LightningModule. Both attributes batch_size (inside the module and inside the datamodule) are updated correctly.
The batch size finder fails if batch_size is specified in the datamodule, only.

[edenlightning]

@maxjeblick would you like to try and submit a PR to fix?

[maxjeblick]

@edenlightning Yes I can have a look.