Skip to content

Conversation

@SeanNaren
Copy link
Contributor

@SeanNaren SeanNaren commented Mar 1, 2021

What does this PR do?

I've had a few reports that the default for DeepSpeed is incredibly slow and is unusable. After investigation, this is because CPU offload by default is set to True, and this can be prohibitive when running DeepSpeed. Instead we should turn off CPU offload, and keep ZeRO on allowing the user to use CPU Offload if necessary.

This can also been seen from the benchmarks here: https://github.com/SeanNaren/minGPT#ddp-vs-deepspeed

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?

Make sure you had fun coding 🙃

@SeanNaren SeanNaren added bug Something isn't working distributed Generic distributed-related topic 3rd party Related to a 3rd-party labels Mar 1, 2021
@SeanNaren SeanNaren added this to the 1.2.x milestone Mar 1, 2021
@SeanNaren SeanNaren self-assigned this Mar 1, 2021
@codecov
Copy link

codecov bot commented Mar 1, 2021

Codecov Report

Merging #6262 (66522bb) into master (3371d32) will decrease coverage by 0%.
The diff coverage is n/a.

@@          Coverage Diff           @@
##           master   #6262   +/-   ##
======================================
- Coverage      93%     93%   -0%     
======================================
  Files         159     159           
  Lines       11365   11365           
======================================
- Hits        10608   10579   -29     
- Misses        757     786   +29     

@SeanNaren SeanNaren mentioned this pull request Mar 1, 2021
@Borda Borda enabled auto-merge (squash) March 1, 2021 17:44
@Borda Borda added the ready PRs ready to be merged label Mar 1, 2021
@mergify mergify bot added the has conflicts label Mar 1, 2021
@mergify mergify bot removed the has conflicts label Mar 2, 2021
@Borda Borda merged commit efda48f into master Mar 2, 2021
@Borda Borda deleted the fix/default_deepspeed branch March 2, 2021 07:53
kaushikb11 pushed a commit to kaushikb11/pytorch-lightning that referenced this pull request Mar 2, 2021
* Change default for CPU offload to false for best throughput/memory efficiency

* Add changelog

* default

Co-authored-by: Jirka Borovec <[email protected]>
kaushikb11 pushed a commit to kaushikb11/pytorch-lightning that referenced this pull request Mar 2, 2021
* Change default for CPU offload to false for best throughput/memory efficiency

* Add changelog

* default

Co-authored-by: Jirka Borovec <[email protected]>
lexierule pushed a commit that referenced this pull request Mar 5, 2021
* Change default for CPU offload to false for best throughput/memory efficiency

* Add changelog

* default

Co-authored-by: Jirka Borovec <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

3rd party Related to a 3rd-party bug Something isn't working distributed Generic distributed-related topic ready PRs ready to be merged

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants