apply_func.py: from torchtext.legacy.data import Batch

[dbonner]

🐛 Bug

In the latest torchtext (version 0.9.0a0+036df73) built from source:
The name Batch is no longer located under torchtext.data
Batch is now located under torchtext.legacy.data

--Error message--
File "/home/daniel/py38/lib/python3.8/site-packages/pytorch_lightning/utilities/apply_func.py", line 28, in
from torchtext.data import Batch
ImportError: cannot import name 'Batch' from 'torchtext.data' (/home/daniel/py38/lib/python3.8/site-packages/torchtext/data/init.py)

You can fix this by changing line line 28 of pytorch_lightning/utilities/apply_func.py to:
from torchtext.legacy.data import Batch

This does not reproduce with BoringModel because BoringModel does not import torchtext.data.Batch

Expected behavior

No error message.

Environment

• CUDA:
  - GPU:
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - A100-SXM4-40GB
  - available: True
  - version: 11.2
• Packages:
  - numpy: 1.19.5
  - pyTorch_debug: False
  - pyTorch_version: 1.9.0a0+92a4ee1
  - pytorch-lightning: 1.1.5
  - torchtext: 0.9.0a0+036df73 <--------------------------
  - tqdm: 4.57.0
• System:
  - OS: Linux
  - architecture:
  - 64bit
  - ELF
  - processor: x86_64
  - python: 3.8.8
  - version: self-balancing architecture #50-Ubuntu SMP Tue Feb 9 06:29:41 UTC 2021

Additional context

[dbonner]

This should fix it:
#6211
I was able to successfully create the option to use torchtext.legacy.data or torchtext.data depending on the torchtext version after my 3rd commit on this pull request.

[dbonner]

My fourth commit in the pull request (#6211 ) uses distutils.version.LooseVersion

[dbonner]

Thanks everyone :)

[Vlad-Galeev]

It still fails when Run on Google Colab, as seen in Unable to import pytorch_lightning on google colab and yes it has been tested.

[monkiravn]

It still fails when Run on Google Colab, as seen in Unable to import pytorch_lightning on google colab and yes it has been tested.
I have the same problem as you. Have you fixed it?
I

[Vlad-Galeev]

Kind of. If you look at the Stack page that I linked, you will see two solutions. Both work.
One installs the pytorch_lightning straight from here(GitHub).
The second does some string of installs and it works, but if you do the installs separately, it fails.

!pip install torchtext==0.8.0 torch==1.7.1 pytorch-lightning==1.2.2

That works in this Colab notebook: https://colab.research.google.com/drive/1GSCd3Gz3EOQIln3v065VKWKbB3_F8xqK?usp=sharing
My solution installs the newest version of pytorch_lightning, including development versions.

!pip install git+https://github.com/PyTorchLightning/pytorch-lightning

[Dheeraj-Arremsetty]

I have followed the above instructions, and when I have installed it. I can see this error message. Is this ok to ignore this message?

[Vlad-Galeev]

You won't be able to use the torchvision package/module, if you use that solution. If you use the other one, and use the comment that has been added, you can use everything (I think).