Unable to import pytorch_lightning on google colab

Question

I have done the following:

!pip install pytorch_lightning -qqq
import pytorch_lightning

But get the following error:

ImportError                               Traceback (most recent call last)
<ipython-input-7-d883b15aac58> in <module>()
----> 1 import pytorch_lightning

----------------------------------9 frames------------------------------------------------
/usr/local/lib/python3.7/dist-packages/pytorch_lightning/utilities/apply_func.py in <module>()
     26 
     27 if _TORCHTEXT_AVAILABLE:
---> 28     from torchtext.data import Batch
     29 else:
     30     Batch = type(None)

ImportError: cannot import name 'Batch' from 'torchtext.data' (/usr/local/lib/python3.7/dist-packages/torchtext/data/__init__.py)

What could the issue be?

Solution 1

As said in Issue #6415 on Github, try installing from the GitHub.
It worked for me.

!pip install git+https://github.com/PyTorchLightning/pytorch-lightning
import pytorch_lightning as pl
print(pl.__version__)

Output:

1.3.0dev

It seems that the error is coming from Issue #6210 and they say it was fixed. I guess it wasn't uploaded to PyPi.

Solution 2

Example working env: https://colab.research.google.com/drive/1GSCd3Gz3EOQIln3v065VKWKbB3_F8xqK?usp=sharing

Can you try after restarting your env.

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

import pytorch_lightning as pl
print(pl.__version__)
...

There appears to be a bug that has not hit pip yet with pytorch lightning not referencing the newest torchtext.

Solution 3

Late, but if you are having similar difficulties with PyTorch Lightning Bolts you can solve it at the time of posting by making use of the same strategy:

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

Comments

[PythonSnek]

fails, with an OSError

[Avi Thaker]

Could you send the error? I am unable to replicate. You can also try with !pip install pytorch=1.7.1 in addition to the other commands.

[PythonSnek]

/usr/lib/python3.7/ctypes/__init__.py in __init__(self, name, mode, handle, use_errno, use_last_error) 362 363 if handle is None: --> 364 self._handle = _dlopen(self._name, mode) 365 else: 366 self._handle = handle

[PythonSnek]

Also 1.7.1 doesn't exist

[Avi Thaker]

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

[Avi Thaker]

Then restart your env

[Avi Thaker]

Let us continue this discussion in chat.

[Kyle_397]

For Colab users, then you can solve this by reinstalling (or upgrading) pytorch_lightning version 1.3.0dev without any dependencies except fsspec. !pip install git+https://github.com/PyTorchLightning/pytorch-lightning fsspec --no-deps --target=$nb_path

[PythonSnek]

Thanks, I didn't know that was possible.