No module named 'torch._six'

[cdeepali]

🐛 Bug

import pl_bolts fails with the following error:

  File "<mypath>/test_tmp/run_test.py", line 2, in <module>
    import pl_bolts
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/__init__.py", line 11, in <module>
    from pl_bolts import (  # noqa: E402
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/callbacks/__init__.py", line 6, in <module>
    from pl_bolts.callbacks.ssl_online import SSLOnlineEvaluator
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/callbacks/ssl_online.py", line 12, in <module>
    from pl_bolts.models.self_supervised.evaluator import SSLEvaluator
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/models/self_supervised/__init__.py", line 21, in <module>
    from pl_bolts.models.self_supervised.cpc.cpc_module import CPC_v2
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/models/self_supervised/cpc/__init__.py", line 1, in <module>
    from pl_bolts.models.self_supervised.cpc.cpc_module import CPC_v2
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/models/self_supervised/cpc/cpc_module.py", line 14, in <module>
    from pl_bolts.datamodules.stl10_datamodule import STL10DataModule
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/datamodules/__init__.py", line 1, in <module>
    from pl_bolts.datamodules.async_dataloader import AsynchronousLoader
  File "<myenv>/lib/python3.10/site-packages/pl_bolts/datamodules/async_dataloader.py", line 9, in <module>
    from torch._six import string_classes
ModuleNotFoundError: No module named 'torch._six'

See - pytorch/pytorch#94709

[rehammsalah]

I'm facing the same issue, did you find any solution?

[rbhernandez]

I did pip install --upgrade torch torchvision and it works

[devdastl]

Thanks @rbhernandez that solved while import from pl_bolts.models but while importing pl_bolts.datamodules I am again getting torch.six error

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
[<ipython-input-19-fbbe43d17848>](https://localhost:8080/#) in <cell line: 1>()
----> 1 from pl_bolts.datamodules import CIFAR10DataModule
      2 
      3 datamodule = CIFAR10DataModule('.')

7 frames
[/usr/local/lib/python3.9/dist-packages/pl_bolts/datamodules/async_dataloader.py](https://localhost:8080/#) in <module>
      7 import torch
      8 from torch import Tensor
----> 9 from torch._six import string_classes
     10 from torch.utils.data import DataLoader, Dataset
     11 

ModuleNotFoundError: No module named 'torch._six'

---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.
---------------------------------------------------------------------------

[rbhernandez]

@devdastl you can try upgrade lighting_bolts. Torch2.0 is now stable, and it must be most users choice now.
If you want compatible with older torch, just make a judgement on version to import.

Originally posted by @lucasjinreal in #992 (comment)

[devdastl]

@rbhernandez thanks for the input. Do you know which version of pl_bolts work for 2.0.0+cu118?
Thanks again for the help!

[Wheest]

I'm also experiencing this issue with PyTorch 2.0.0+cu118, with pl_bolts version 0.6.0.post1. I had to downgrade to torch==1.13.0.

[devdastl]

@Wheest
Since I was using colab, I found out it was more difficult to downgrade So I just changed that line in source code with string_classes = str in /usr/local/lib/python3.9/dist-packages/pl_bolts/datamodules/async_dataloader.py

@Borda is this the correct solution? Should we replace from torch._six import string_classes with string_classes = str

[lucasjinreal]

@rbhernandez Please upgrade lighting_bolts rather than ask user to downgrading torch,

1. torch2.0 is now stable, and it must be most users choice now, please upgrade to it;
2. if you want compatible with older torch, just make a judgement on version to import