package versions - reproductibility

[equalmc2]

I installed on a linux machine a conda virtual environment and then I run a pip install -r requirements.txt Unfortunately I always have errors on some package, like "from pytorch_lightning.utilities.seed import seed_everything" gives 'ImportError: cannot import name 'seed_everything' from 'pytorch_lightning.utilities.seed'" or "from pytorch_lightning.plugins import DDPPlugin
ImportError: cannot import name 'DDPPlugin' from 'pytorch_lightning.plugins'".

I tried different versions of packages, higher or lower, depending on what I found in Issues or on the internet. Is there any other method for testing this? A docker image, or the precise package numbers you used? Thank you

[zahrasalarian]

Not sure if this will still be helpful, but I faced the same issue and It was resolved by setting the following package versions:

torch==1.13
torchtext==0.6