-
Notifications
You must be signed in to change notification settings - Fork 3.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
PoC: Accelerator refactor [wip] [skip ci] #5616
PoC: Accelerator refactor [wip] [skip ci] #5616
Conversation
@justusschock took me almost two hours to rebase and resolve conflicts on the way, but still, can you check that all is fine? |
@Borda Thank you. |
Happy to help =) |
What does this PR do?
This PR is a revamp of #5385 since I messed up with git.
Closes #5385
Fixes #4510
This PR separates the Accelerator (Hardware Part) from the actual different training routines.
Workflow actions:
Remaining TODOs:
block_backward_sync
andprepare_for_backward
) from old DDPPlugin to new to avoid performance hits (Adrian/Justus)So far this PR was co-authored with @awaelchli !
cc @awaelchli @tchaton @SeanNaren who will likely work on this!
Slides for motivation and high-level overview
List of PRs to look out for (when rebasing, code we need to manually copy over to new files):
#5221, #5195, #5300, #5388