Skip to content
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.

Sign up for GitHub

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

Jump to bottom

Fix metric states storing and loading on checkpointing #12236

Closed
wants to merge 2 commits into from
Closed

Fix metric states storing and loading on checkpointing #12236

wants to merge 2 commits into from

Conversation

DuYicong515
Copy link
Contributor

@DuYicong515 DuYicong515 commented Mar 5, 2022 •
edited
Loading

What does this PR do?

Part of #12193

Fix a bug that metric states is synced before storing only when fault_tolerance_training is enabled, while at loading time we assumed it's always synced before stored as we load for rank_0 and reset the others.

Does your PR introduce any breaking changes? If yes, please list them.

N/A

Before submitting

  • Was this discussed/approved via a GitHub issue? (not for typos and docs)
  • Did you read the contributor guideline, Pull Request section?
  • Did you make sure your PR does only one thing, instead of bundling different changes together?
  • Did you make sure to update the documentation with your changes? (if necessary)
  • Did you write any new necessary tests? (not for typos and docs)
  • Did you verify new and existing tests pass locally with your changes?
  • Did you list all the breaking changes introduced by this pull request?
  • Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

  • [not yet, can't revert back to draft though] Is this pull request ready for review? (if not, please submit in draft mode)
  • Check that all items from Before submitting are resolved
  • Make sure the title is self-explanatory and the description concisely explains the PR
  • Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

@ananthsub ananthsub added the checkpointing Related to checkpointing label Mar 5, 2022
@DuYicong515 DuYicong515 force-pushed the bug/metric_checkpointing branch 2 times, most recently from 0e2b576 to 9698e24 Compare March 5, 2022 07:16
@DuYicong515 DuYicong515 force-pushed the bug/metric_checkpointing branch from 9698e24 to 5454d64 Compare March 7, 2022 23:03
@DuYicong515 DuYicong515 marked this pull request as ready for review March 7, 2022 23:21
@DuYicong515 DuYicong515 mentioned this pull request Mar 7, 2022
Closed
@DuYicong515 DuYicong515 changed the title Fix metric states storing and loading Fix metric states storing and loading on checkpointing Mar 7, 2022
@DuYicong515
Copy link
Contributor Author

DuYicong515 commented Mar 8, 2022 •
edited
Loading

test_deepspeed_multigpu_stage_3_checkpointing is failing if we make metric states a default part of checkpointing by removing the fault_tolerance_training check. The exception was complaining about tensors are on different devices though. I guess deep speed strategy checkpointing does not work with saving/loading metric states yet.

Though metric state syncing before saving should not happening on deepspeed stage 3 checkpointing anyways.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@tchaton tchaton Awaiting requested review from tchaton tchaton is a code owner

@SeanNaren SeanNaren Awaiting requested review from SeanNaren

@carmocca carmocca Awaiting requested review from carmocca

@Borda Borda Awaiting requested review from Borda Borda is a code owner

@williamFalcon williamFalcon Awaiting requested review from williamFalcon

@awaelchli awaelchli Awaiting requested review from awaelchli

@justusschock justusschock Awaiting requested review from justusschock justusschock is a code owner

@kaushikb11 kaushikb11 Awaiting requested review from kaushikb11

@rohitgr7 rohitgr7 Awaiting requested review from rohitgr7

Assignees
No one assigned
Labels
checkpointing Related to checkpointing has conflicts
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@DuYicong515 @ananthsub