Correctly convert namedtuples in DDP #44220
Closed
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
Closes #44009 Currently if a dataloader returns objects created with a collections.namedtuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/) [ghstack-poisoned]
rohan-varma
requested review from
apaszke,
mrshenli,
pritamdamania87 and
zhaojuanmao
as code owners
Closes #44009 Currently if a dataloader returns objects created with a `collections.namedtuple` or `typing.NamedTuple`, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/) [ghstack-poisoned]
rohan-varma
added a commit
that referenced
this pull request
Sep 4, 2020
Pull Request resolved: #44220 Closes #44009 Currently if a dataloader returns objects created with a collections.namedtuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 ghstack-source-id: 111478085 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/)
💊 CI failures summary and remediationsAs of commit 939378b (more details on the Dr. CI page): Commit 939378b was recently pushed. Waiting for builds... This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 15 times. |
|
Not sure who the best reviewer for this change is since |
test/distributed/test_distributed.py
Outdated
Comment on lines
3134
to
3135
| @require_backend({"nccl", "gloo"}) | ||
| @require_n_gpus_for_nccl_backend(int(os.environ["WORLD_SIZE"]), os.environ["BACKEND"]) |
There was a problem hiding this comment.
Can we also add some unit test that directly test the scatter API with a variety of test cases? For example the case for namedtuple(a=1, b=1) you mentioned in the docs is a good test.
There was a problem hiding this comment.
I added this test in test/test_cuda.py since it doesn't look like there were tests for this API anywhere else.
Closes #44009 Currently if a dataloader returns objects created with a `collections.namedtuple` or `typing.NamedTuple`, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/) [ghstack-poisoned]
rohan-varma
added a commit
that referenced
this pull request
Sep 15, 2020
Pull Request resolved: #44220 Closes #44009 Currently if a dataloader returns objects created with a collections.namedtuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 ghstack-source-id: 112050439 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/)
Closes #44009 Currently if a dataloader returns objects created with a `collections.namedtuple` or `typing.NamedTuple`, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/) [ghstack-poisoned]
rohan-varma
added a commit
that referenced
this pull request
Sep 16, 2020
Pull Request resolved: #44220 Closes #44009 Currently if a dataloader returns objects created with a collections.namedtuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 ghstack-source-id: 112198252 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/)
Codecov Report
@@ Coverage Diff @@
## gh/rohan-varma/164/base #44220 +/- ##
===========================================================
+ Coverage 67.85% 68.50% +0.65%
===========================================================
Files 384 409 +25
Lines 49919 52587 +2668
===========================================================
+ Hits 33873 36026 +2153
- Misses 16046 16561 +515
Continue to review full report at Codecov.
|
|
@pritamdamania87 Added the tests for scatter, any more thoughts on this diff? Thanks! |
pritamdamania87
approved these changes
Oct 2, 2020
|
|
||
| scatter_out = scatter_gather.scatter(inp, target_gpus) | ||
| for x in scatter_out: | ||
| self.assertTrue(isinstance(x, type(inp))) |
There was a problem hiding this comment.
Can we also assert that the _fields attribute is preserved correctly and also the appropriate values?
There was a problem hiding this comment.
Thanks, adding the tests for valued actually uncovered a bug in the implementation that I just fixed.
Closes #44009 Currently if a dataloader returns objects created with a `collections.namedtuple` or `typing.NamedTuple`, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/) [ghstack-poisoned]
rohan-varma
added a commit
that referenced
this pull request
Oct 2, 2020
Pull Request resolved: #44220 Closes #44009 Currently if a dataloader returns objects created with a collections.namedtuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple. Fix this in `scatter_gather.py` to resolve the issue reported in #44009 ghstack-source-id: 113423287 Differential Revision: [D23536752](https://our.internmc.facebook.com/intern/diff/D23536752/)
|
This pull request has been merged in f8c1ca5. |
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.
Stack from ghstack:
Closes #44009
Currently if a dataloader returns objects created with a
collections.namedtupleortyping.NamedTuple, this will incorrectly be cast to a tuple. As a result, if we have data of these types, there can be runtime errors during the forward pass if the module is expecting a named tuple.Fix this in
scatter_gather.pyto resolve the issue reported in#44009
Differential Revision: D23536752