Sharded Plugin 3/n: Expose step input to DDP plugin #4686
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
SeanNaren
requested review from
ananyahjha93,
awaelchli,
Borda,
justusschock,
nateraw,
tchaton,
teddykoker and
williamFalcon
as code owners
| @@ -61,21 +61,23 @@ def train(self): | |||
| return self.ddp_train(process_idx=self.task_idx, mp_queue=None, model=model) | |||
|
|
|||
| def training_step(self, args): | |||
| return self._step(args) | |||
There was a problem hiding this comment.
nice. this is better haha.
Codecov Report
@@ Coverage Diff @@
## master #4686 +/- ##
======================================
Coverage 93% 93%
======================================
Files 117 117
Lines 8939 8941 +2
======================================
+ Hits 8306 8308 +2
Misses 633 633 |
justusschock
approved these changes
Nov 18, 2020
Borda
reviewed
Nov 18, 2020
Borda
reviewed
Nov 18, 2020
Co-authored-by: Jirka Borovec <[email protected]>
|
Hello @SeanNaren! Thanks for updating this PR. There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2020-11-18 14:16:08 UTC |
Borda
approved these changes
Nov 18, 2020
| return self._step(args) | ||
|
|
||
| def _step(self, args): | ||
| args = self.ddp_plugin.on_before_forward(self.trainer.get_model(), *args) |
williamFalcon
approved these changes
Nov 18, 2020
rohitgr7
pushed a commit
that referenced
this pull request
Nov 21, 2020
* Allow ddp plugin to move the input to a different device if needed * Swapped name to on_before_forward to align with hooks in the future * Update pytorch_lightning/plugins/ddp_plugin.py Co-authored-by: Jirka Borovec <[email protected]> * Pass variable arg type to hook, add example * Remove blank space (pep check) * Added blank line Co-authored-by: William Falcon <[email protected]> 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
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Relates to #4178
ShardedDDP needs to move the device to the GPU before passing through to the model. This is not the case for DDP which handles this internally. Allows other custom DDP parallel plugins an opportunity to modify the input before passing into the wrapper!
cc @williamFalcon @tchaton @awaelchli @Borda
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 in short, see following bullet-list:
Did you have fun?
Make sure you had fun coding 🙃