ahh yeah, I see that the current release has pinne...
# neural-forecast
m
ahh yeah, I see that the current release has pinned pytorch-lightning 1.6.5, could you create a new release, relaxing this constraint?
c
We recently updated the pytorch lightning version to >= 1.9.0. I will double check it is working properly
m
yes! that looks good! but when I install with
pip
, I get 1.4.0, and that still references 1.6.5
c
@fede can you also take a look at this?
m
thank you! we’d love to have neuralforecast supported form the start, and aiming for a release very soon, and our internal tests don’t pass right now when referencing the last published version.
c
you are right, the latest version in pip does not have the change to 1.9.0 yet. @fede will soon post a quick fix you can do before we change the pip version. We are still fixing some issues with the latest lightning version
f
hey @Mark Aron Szulyovszky! Sorry for the inconvenience, the problem arises due to the newest release of torch (2.0.0), so you could solve the issue by adding “torch<2.0.0” as a dependency. I tested locally with your repo and it worked as expected. I also opened a PR in your branch 🙂 Let me know if that works https://github.com/dream-faster/fold-models/pull/52
m
thank you @fede (nixtla) (they/them)! we really appreciate you looking into it!:)
f
my pleasure! :)
k
Are the changes released yet
I guess changing the lightning version changes the distributed training strategy to no more support for dp and ddp_notebook is new