google colab: ModuleNotFoundError: No module named 'torchtext.legacy'

[hanp0]

The colab demo now has an module error when training and predicting, possibly due to to colab changes?

from torchtext.legacy.data import Batch ModuleNotFoundError: No module named 'torchtext.legacy'

[devriesewouter89]

I'm experiencing the same behaviour, though as the git repo hasn't changed a lot, I believe it could be a pytorch/torchtext compatibility issue.

[kiashann]

Did you solve this problem? @hanp0 @devriesewouter89 @windj007

[terwelja]

I solved this issue for myself @kiashann. You need to install an earlier version of torchtext and torchvision, I personally chose torchtext/torchvision 0.9.0, and wrote !pip install torchtext==0.9.0 --quiet and !pip install torchvision==0.9.0 --quiet after the fix opencv line. I think Google Colab installed a version of pytorch that was incompatible with LaMa.

[hanp0]

@terwelja when I try that I get

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
torchvision 0.12.0+cu113 requires torch==1.11.0, but you have torch 1.8.0 which is incompatible.
torchaudio 0.11.0+cu113 requires torch==1.11.0, but you have torch 1.8.0 which is incompatible.

when installing and then

RuntimeError: CUDA error: no kernel image is available for execution on the device

when trying to train?

[kiashann]

I solved this issue for myself @kiashann. You need to install an earlier version of torchtext and torchvision, I personally chose torchtext/torchvision 0.9.0, and wrote !pip install torchtext==0.9.0 --quiet and !pip install torchvision==0.9.0 --quiet after the fix opencv line. I think Google Colab installed a version of pytorch that was incompatible with LaMa.

Thanks for your response. How did you run train.py file? I used this command (!python bin/train.py -cn lama-fourier-celeba data.batch_size=10) but I got below error

Traceback (most recent call last): File "bin/train.py", line 21, in from saicinpainting.training.trainers import make_training_model ModuleNotFoundError: No module named 'saicinpainting'

[devriesewouter89]

I can confirm that the combination !pip install torchtext==0.9.0 --quiet and !pip install torchvision==0.9.0 --quiet works for the colab notebook. @kiashann, I guess you need to set your PYTHONPATH correctly (e.g. !PYTHON_PATH=/content/lama if you downloaded the lama git in a notebook)

Op wo 4 mei 2022 om 12:35 schreef kiashann @.***>:

I solved this issue for myself @kiashann https://github.com/kiashann. You need to install an earlier version of torchtext and torchvision, I personally chose torchtext/torchvision 0.9.0, and wrote !pip install torchtext==0.9.0 --quiet and !pip install torchvision==0.9.0 --quiet after the fix opencv line. I think Google Colab installed a version of pytorch that was incompatible with LaMa.

Thanks for your response. How did you run train.py file? I used this command (!python bin/train.py -cn lama-fourier-celeba data.batch_size=10) but I got below error

Traceback (most recent call last): File "bin/train.py", line 21, in from saicinpainting.training.trainers import make_training_model ModuleNotFoundError: No module named 'saicinpainting'

— Reply to this email directly, view it on GitHub https://github.com/saic-mdal/lama/issues/114#issuecomment-1117161906, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFMQIRHHGOMKVOIJS7SIVNTVIJHI3ANCNFSM5URVO52Q . You are receiving this because you were mentioned.Message ID: @.***>

-- Wouter Devriese +32 495 15 73 09

[hanp0]

@devriesewouter89 are you training in colab or only predicting?

[devriesewouter89]

@hannah I'm only inferring images (predict and replace background)

Op wo 4 mei 2022 om 17:41 schreef Hannah @.***>:

@devriesewouter89 https://github.com/devriesewouter89 are you training in colab or only predicting?

— Reply to this email directly, view it on GitHub https://github.com/saic-mdal/lama/issues/114#issuecomment-1117505991, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFMQIRCU6GMY7BEB5EGBM5LVIKLAZANCNFSM5URVO52Q . You are receiving this because you were mentioned.Message ID: @.***>

-- Wouter Devriese +32 495 15 73 09

[windj007]

Hi! It's great to see that you help each other! What questions remain unsolved?

[windj007]

We neither use nor install torchtext - that library comes by default in colab and they might have changed the default version - so reinstall another version (e.g. 0.9.0) is an appropriate solution

[windj007]

@kiashann to resolve No module named 'saicinpainting' issue, export PYTHONPATH=/content/lama environment variable

[hanp0]

Hi! It's great to see that you help each other! What questions remain unsolved?

I'm still encountering the following error when trying to train using the solution above

RuntimeError: CUDA error: no kernel image is available for execution on the device

[windj007]

@hanp0 You're encountering CUDA error: no kernel image is available for execution on the device because you've installed torchvision and torchaudio pre-built for cuda11.3 (see +cu113 in version) - but torch itself is built for another cuda. It seems that you've run pip install -U torchvision torchaudio - which just upgrades to the latest versions and does not check compatibility with the existing packages (that's how pip works).

[windj007]

I've checked now - colab now has torch 1.11.0+cu113 by default, which is too recent for the LaMa codebase. I'll try to install 1.8 and the corresponding libraries now

[windj007]

Fixed in 8ba8381