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.

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

Keep global step update in the loop #8856

Merged
merged 19 commits into from
Sep 14, 2021
Merged

Conversation

carmocca
Copy link
Contributor

@carmocca carmocca commented Aug 11, 2021

What does this PR do?

The gradient accumulation logic check and global step update should be contained within the loops.
This PR changes the TTP API so that we only use it to know whether the optimization is handled internally in the case of the IPU and DeepSpeed plugins.

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

Beta TrainingTypePlugin and Accelerator API:

  • Replaced update_global_step with the accumulate_grad_batches property.

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)
  • [n/a] 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

  • 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

@carmocca carmocca added refactor design Includes a design discussion breaking change Includes a breaking change labels Aug 11, 2021
@carmocca carmocca added this to the v1.5 milestone Aug 11, 2021
@carmocca carmocca self-assigned this Aug 11, 2021
@codecov
Copy link

codecov bot commented Aug 11, 2021

Codecov Report

Merging #8856 (6c509f2) into master (b294c57) will decrease coverage by 4%.
The diff coverage is 57%.

@@           Coverage Diff           @@
##           master   #8856    +/-   ##
=======================================
- Coverage      93%     89%    -4%     
=======================================
  Files         180     180            
  Lines       15062   15051    -11     
=======================================
- Hits        13983   13369   -614     
- Misses       1079    1682   +603     

@carmocca carmocca marked this pull request as ready for review August 11, 2021 16:37
@carmocca carmocca force-pushed the refactor/ttp-accum-grad-batches branch from 41c4c39 to 10045b6 Compare August 11, 2021 19:59
@carmocca carmocca marked this pull request as draft August 11, 2021 20:21
@carmocca carmocca marked this pull request as ready for review September 11, 2021 17:50
@carmocca carmocca requested a review from tchaton September 11, 2021 17:50
Copy link
Contributor

@tchaton tchaton left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM ! Nice cleanup.

@mergify mergify bot added the ready PRs ready to be merged label Sep 13, 2021
@kaushikb11 kaushikb11 merged commit b1ed1db into master Sep 14, 2021
@kaushikb11 kaushikb11 deleted the refactor/ttp-accum-grad-batches branch September 14, 2021 13:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
breaking change Includes a breaking change design Includes a design discussion ready PRs ready to be merged refactor
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants