search

ray-project / ray_lightning

Pytorch Lightning Distributed Accelerators using Ray
Apache License 2.0
211 stars 34 forks source link

Update ray-lightning with the latest PL Strategy API changes #129

Closed kaushikb11 closed 2 years ago

kaushikb11 commented 2 years ago

Hey @amogkam @richardliaw!

With Lightning v1.5, we have introduced a new Trainer Strategy API. You could read more about it in this blog.

I have implemented the required updates in this PR. Do check it out and let me know if you have any questions! Thanks.

richardliaw commented 2 years ago

Hey @kaushikb11, seems like some of the examples still do Trainer(plugin=RayStrategy) -- should it be Trainer(strategy=RayStrategy)?

amogkam commented 2 years ago

Hey @kaushikb11 thanks a bunch for making this PR!

Although the strategy argument was added in 1.5, it seems like the strategies package and DDPStrategy, etc. classes are only going to be introduced in 1.6.

Would it make sense to hold off on this PR until the PTL 1.6 release?

MarkusSpanring commented 2 years ago

@kaushikb11 @amogkam really looking forward to seeing this PR in the next ray_lightning release! Since PTL 1.6 is now out when will be the earliest to expect it?

rokrokss commented 2 years ago

@amogkam could you revisit this?

amogkam commented 2 years ago

Just re-triggered CI- let's see if they pass!

amogkam commented 2 years ago

@kaushikb11 looks like the tests are still failing with PTL 1.6. Do you mind taking a look?

JiahaoYao commented 2 years ago

Hi all, the pr is already merged in another pr https://github.com/ray-project/ray_lightning/pull/163. this pr is closed.