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

Deprecate Trainer.devices in favor of Trainer.num_devices and Trainer.device_ids #12151

Merged
merged 14 commits into from
Mar 18, 2022

Conversation

DuYicong515
Copy link
Contributor

@DuYicong515 DuYicong515 commented Mar 1, 2022

What does this PR do?

Part of #12171

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:

  • 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

cc @four4fish @awaelchli

@DuYicong515 DuYicong515 changed the title Deprecate Trainer.devices in favor of Trainer.num_devices and Trainer.device_ids WIP: Deprecate Trainer.devices in favor of Trainer.num_devices and Trainer.device_ids Mar 1, 2022
@DuYicong515 DuYicong515 changed the title WIP: Deprecate Trainer.devices in favor of Trainer.num_devices and Trainer.device_ids Deprecate Trainer.devices in favor of Trainer.num_devices and Trainer.device_ids Mar 1, 2022
@DuYicong515 DuYicong515 marked this pull request as ready for review March 1, 2022 20:32
pytorch_lightning/trainer/trainer.py Show resolved Hide resolved
pytorch_lightning/trainer/trainer.py Show resolved Hide resolved
tests/trainer/test_trainer.py Outdated Show resolved Hide resolved
@mergify mergify bot added the has conflicts label Mar 2, 2022
pytorch_lightning/trainer/trainer.py Outdated Show resolved Hide resolved
@mergify mergify bot added ready PRs ready to be merged and removed has conflicts labels Mar 2, 2022
@kaushikb11 kaushikb11 added this to the 1.6 milestone Mar 3, 2022
Copy link
Member

@ethanwharris ethanwharris left a comment

Choose a reason for hiding this comment

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

LGTM, small comment

pytorch_lightning/trainer/trainer.py Outdated Show resolved Hide resolved
@DuYicong515 DuYicong515 force-pushed the refactor/devices branch 2 times, most recently from f27d46b to 84ea0c5 Compare March 18, 2022 06:47
@mergify mergify bot removed the has conflicts label Mar 18, 2022
@ananthsub ananthsub merged commit ed2bcc5 into Lightning-AI:master Mar 18, 2022
Copy link
Contributor

@awaelchli awaelchli left a comment

Choose a reason for hiding this comment

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

Apologies for the delayed review.

@@ -2010,6 +2010,23 @@ def should_rank_save_checkpoint(self) -> bool:
def num_nodes(self) -> int:
return getattr(self.strategy, "num_nodes", 1)

@property
def device_ids(self) -> List[int]:
Copy link
Contributor

Choose a reason for hiding this comment

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

I would have preferred if this was implemented on the accelerator connector. It is IMO not the responsibility of the Trainer to compute the device_ids off the strategy. Note the the strategy is owned by the connector, not the trainer. As the trainer has already many responsibilities, we should delegate as much as possible to the concrete components.

Copy link
Contributor Author

@DuYicong515 DuYicong515 Mar 20, 2022

Choose a reason for hiding this comment

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

Thanks for reviewing @awaelchli!

My personal thoughts: the accelerator connector should stay internal, and accelerator connector should not expose those device related public properties.

Note the the strategy is owned by the connector, not the trainer.

Personally I feel the accelerator connector does a lot of checks/set up/initialization and configuration of the accelerator and strategy. After things are set up already, strategy/accelerator are already accessible on the Trainer, while the connector stays internal.

I think delegating computations using accelerator/strategy to the accelerator_connector will make us add many public properties to the internal-facing accelerator_connector. Deriving from strategy/accelerator directly are more straight forward, and they(strategy, accelerator) are very external facing as well.

I'm very new to this and my feelings might be wrong. Feel free to add more thoughts to issue #12171!

@DuYicong515 DuYicong515 deleted the refactor/devices branch March 20, 2022 23:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ready PRs ready to be merged refactor trainer
Projects
None yet
Development

Successfully merging this pull request may close these issues.

9 participants