-
Notifications
You must be signed in to change notification settings - Fork 1.2k
Description
Like many other issues here, I was running into many environment-related issues trying to use PyTorch-VAE. I was using Python 3.9. I found this comment in a closed issue that helped me:
The versions mentioned in the requirements.txt are compatible with >=Python3.5 and <=Python3.7. So if you use python3.8 then the corresponding versions of the libraries will not be compatible. If you use one of the python versions mentioned, you should not face this issue.
Originally posted by @AsthaGarg16 in #33
Switching to Python 3.6 and running pip install -r requirements.txt worked without issues. I haven't tested Python versions 3.5 or 3.7, but I think there likely is a cap on Python versions in which this is compatible, and simply listing "Python >= 3.5" is misleading.
The only caveat I noticed with the older version is that newer GPUs are incompatible with older CUDA binaries required to run. I was successful with a TITAN Xp, which is from 2017.
I hope this helps someone, and thank you to the commenter above who made the original observation!