Engine: kwargs #1156
Merged
Engine: kwargs #1156
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
Scitator
reviewed
Apr 5, 2021
catalyst/engines/apex.py
Outdated
| model = ApexDistributedDataParallel(model, delay_allreduce=self.delay_all_reduce) | ||
| model, optimizer = amp.initialize(model, optimizer, **self.apex_kwargs) | ||
| # TODO: kwargs for Apex DDP ? | ||
| model = ApexDistributedDataParallel(model) # , delay_allreduce=self.delay_all_reduce) |
There was a problem hiding this comment.
could we also add ddp_kwargs and pass them here?
There was a problem hiding this comment.
in this case, we have to remove ** from the init and make apex_krargs and ddp_kwargs dict storages
Scitator
reviewed
Apr 8, 2021
There was a problem hiding this comment.
the PR looks amazing, nevertheless could we please make a few extra changes:
- rename
ddp_kwargstodist_kwargsorprocess_kwargs, as far as they are used fortorch.distributed.init_process_group - add truly
ddp_kwargsand use them forApexDistributedDataParallelandDistributedDataParallelwrappers - add an extra expectation for such cases - I mean, we should raise an error if we could not wrap the model correctly
Huge thanks in advance!
Scitator
reviewed
Apr 8, 2021
Scitator
reviewed
Apr 9, 2021
catalyst/engines/torch.py
Outdated
Comment on lines
334
to
335
| if "device_ids" not in self.ddp_kwargs: | ||
| self.ddp_kwargs["device_ids"] = [self.device] |
There was a problem hiding this comment.
we should move this under def setup_process(self, rank: int = -1, world_size: int = 1): in the end
cause self.device = None
| os.environ["MASTER_ADDR"] = str(self.address) | ||
| os.environ["MASTER_PORT"] = str(self.port) | ||
| dist.init_process_group(self.backend, rank=self.rank, world_size=self.world_size) | ||
| dist.init_process_group(**self.process_group_kwargs) | ||
| torch.cuda.set_device(int(self._rank)) | ||
| self.device = f"cuda:{int(self._rank)}" |
There was a problem hiding this comment.
Suggested change
| self.device = f"cuda:{int(self._rank)}" | |
| self.device = f"cuda:{int(self._rank)}" | |
| if "device_ids" not in self.ddp_kwargs: | |
| self.ddp_kwargs["device_ids"] = [self.device] |
catalyst/engines/torch.py
Outdated
Comment on lines
333
to
335
| self.ddp_kwargs = copy.deepcopy(ddp_kwargs) | ||
| if "device_ids" not in self.ddp_kwargs: | ||
| self.ddp_kwargs["device_ids"] = [self.device] |
There was a problem hiding this comment.
Suggested change
| self.ddp_kwargs = copy.deepcopy(ddp_kwargs) | |
| if "device_ids" not in self.ddp_kwargs: | |
| self.ddp_kwargs["device_ids"] = [self.device] | |
| self.ddp_kwargs = copy.deepcopy(ddp_kwargs) |
Scitator
approved these changes
Apr 9, 2021
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.
Before submitting (checklist)
catalyst-make-codestyle && catalyst-check-codestyle(pip install -U catalyst-codestyle).make check-docs?pytest .?latestandminimalrequirements?Description
Related Issue
Type of Change
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
PS