[Feature] Support for Pytorch-lightning 1.5.0

[jamesbsilva]

Search before asking

• I had searched in the issues and found no similar feature requirement.

Description

As of Pytorch-lightning 1.5.0

https://github.com/PyTorchLightning/pytorch-lightning/blob/49d09aa28b4bdde7a272c3e77dc3f13613802672/CHANGELOG.md#150---2021-11-02

The property running_sanity_check in the Trainer class has been deprecated in favor of sanity_checking as of 1.5.0 as linked above. This property is used in the following lines which I have been patching to support 1.5.0

ray/python/ray/tune/integration/pytorch_lightning.py

Line 177 in 3b77840

if trainer.running_sanity_check:

ray/python/ray/tune/integration/pytorch_lightning.py

Line 231 in 3b77840

if trainer.running_sanity_check:

Is there any plan to support 1.5.0+? I am not sure what the preferences are as far as dealingl with backcompatibility otherwise could submit a PR.

I tried searching about this although from the changelog above the complete deprecation of the property was just released in packages of just a few weeks ago.

Use case

Support for using the newer versions of pytorch-lightning .

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[richardliaw]

I believe a similar PR to achieve this is being implemented by the great @kaushikb11?

#20562

[jamesbsilva]

It looks like that will indeed make this a subset of that functionality. Nice, I cant wait until this is released.

[amogkam]

Will keep this open until #20562 is merged

[dynamicwebpaige]

Would it make sense to close this issue, given that #20562 is now merged?

We might also want to reach out to the prob-ml/bliss team to reintroduce their Ray dependency, as it looks like they might have been blocked on Lightning support.

[amogkam]

Yes this can be closed- Ray Tune supports PTL 1.5 as of #20562!