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

Efficient gradient accumulation in LightningLite #14966

Merged
merged 33 commits into from
Oct 19, 2022
Merged

Conversation

awaelchli
Copy link
Contributor

@awaelchli awaelchli commented Sep 30, 2022

What does this PR do?

Fixes #14713
Also sets the stage for #9234

# Accumulate gradient 8 batches at a time
is_accumulating = batch_idx % 8 != 0

with self.skip_backward_sync(model, enabled=is_accumulating):
    output = model(input)
    loss = ...
    self.backward(loss)
    ...

if not is_accumulating:
    optimizer.step()
    optimizer.zero_grad()

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

None.

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 update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

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

  • 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?

I made sure I had fun coding 🙃

cc @Borda @carmocca @justusschock @awaelchli

@awaelchli awaelchli added feature Is an improvement or enhancement fabric lightning.fabric.Fabric labels Sep 30, 2022
@awaelchli awaelchli added this to the pl:future milestone Sep 30, 2022
@awaelchli awaelchli self-assigned this Oct 1, 2022
@github-actions github-actions bot added the pl Generic label for PyTorch Lightning package label Oct 16, 2022
@awaelchli awaelchli removed this from the future milestone Oct 16, 2022
Copy link
Member

@justusschock justusschock left a comment

Choose a reason for hiding this comment

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

ltgm in general few comments regarding naming and errors

src/lightning_lite/lite.py Outdated Show resolved Hide resolved
src/lightning_lite/lite.py Outdated Show resolved Hide resolved
src/lightning_lite/strategies/ddp.py Outdated Show resolved Hide resolved
src/lightning_lite/lite.py Outdated Show resolved Hide resolved
@mergify mergify bot added the ready PRs ready to be merged label Oct 18, 2022
@codecov
Copy link

codecov bot commented Oct 19, 2022

Codecov Report

Merging #14966 (d6a095a) into master (4acb10f) will increase coverage by 2%.
The diff coverage is 89%.

Additional details and impacted files
@@            Coverage Diff            @@
##           master   #14966     +/-   ##
=========================================
+ Coverage      82%      84%     +2%     
=========================================
  Files         408      288    -120     
  Lines       29940    21959   -7981     
=========================================
- Hits        24686    18499   -6187     
+ Misses       5254     3460   -1794     

@mergify mergify bot added has conflicts and removed ready PRs ready to be merged labels Oct 19, 2022
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 !

docs/source-pytorch/starter/lightning_lite.rst Outdated Show resolved Hide resolved
@mergify mergify bot added ready PRs ready to be merged and removed has conflicts ready PRs ready to be merged labels Oct 19, 2022
@awaelchli awaelchli enabled auto-merge (squash) October 19, 2022 19:14
@awaelchli awaelchli merged commit 045c2f5 into master Oct 19, 2022
@awaelchli awaelchli deleted the lite/gradient-sync branch October 19, 2022 19:55
@carmocca carmocca modified the milestones: v1.9, v1.8 Oct 20, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
fabric lightning.fabric.Fabric feature Is an improvement or enhancement pl Generic label for PyTorch Lightning package ready PRs ready to be merged
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Provide a method to skip gradient synchronization in LightningLite
4 participants