Allow DDP to wrap multi-GPU modules #19271
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
mrshenli
changed the title
mrshenli
force-pushed
the
export-D14822375
branch
from
17f4652
to
b891c96
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
b891c96
to
68d676c
Compare
mrshenli
changed the title
mrshenli
force-pushed
the
export-D14822375
branch
from
68d676c
to
19acac3
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
19acac3
to
b2c2c32
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
b2c2c32
to
29baf9e
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
29baf9e
to
9670ad8
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
9670ad8
to
0da1e6f
Compare
pietern
approved these changes
Apr 17, 2019
| model = QuadraGpuNet(gpus) | ||
|
|
||
| ddp_model = DistributedDataParallel( | ||
| copy.deepcopy(model), |
There was a problem hiding this comment.
No need for deepcopy here?
There was a problem hiding this comment.
Don't we need to make sure that model and ddp_model operate on independent params so that we can compare?
There was a problem hiding this comment.
Never mind, it's needed because of the numerical equivalence testing.
test/test_c10d.py
Outdated
| self.assertEqual(len(gpus), 4, "expecting 4 gpus per process") | ||
| gpus = gpus[:4] | ||
| gpu_strs = list(map(lambda i: torch.device('cuda:' + str(i)), gpus)) | ||
| self._test_gloo_backend(gpus, gpu_strs, True) |
There was a problem hiding this comment.
These two can be factored into another helper that calls _test_gloo_backend. At the top level it's good to have them be separate tests so that we see the ones that get skipped.
test/test_c10d.py
Outdated
| self.assertEqual(len(gpus), 4, "expecting 4 gpus per process") | ||
| gpus = gpus[:4] | ||
| gpu_strs = list(map(lambda i: torch.device('cuda:' + str(i)), gpus)) | ||
| self._test_nccl_backend(gpus, gpu_strs, True) |
There was a problem hiding this comment.
These two can be factored into another helper that calls _test_nccl_backend.
torch/nn/parallel/distributed.py
Outdated
| @@ -153,11 +153,22 @@ class DistributedDataParallel(Module): | |||
|
|
|||
| Args: | |||
| module (Module): module to be parallelized | |||
| device_ids (list of int or torch.device): CUDA devices (default: all devices) | |||
| output_device (int or torch.device): device location of output (default: device_ids[0]) | |||
| device_ids (list of int or torch.device): CUDA devices. This should | |||
There was a problem hiding this comment.
Extra whitespace -- is this intentional?
There was a problem hiding this comment.
no, not intentional. I will edit, thanks!
pietern
added
oncall: distributed
mrshenli
force-pushed
the
export-D14822375
branch
from
0da1e6f
to
c71acdc
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
c71acdc
to
f326925
Compare
mrshenli
force-pushed
the
export-D14822375
branch
from
f326925
to
c7b2c4d
Compare
Summary: Pull Request resolved: pytorch#19271 allow DDP to take multi-gpu models Reviewed By: pietern Differential Revision: D14822375 fbshipit-source-id: 8c8bcd4526643be5fa44134620d58fcf2c197238
mrshenli
force-pushed
the
export-D14822375
branch
from
c7b2c4d
to
ea8bdc9
Compare
|
This pull request has been merged in 6732358. |
This was referenced Apr 18, 2019
zhangguanheng66
pushed a commit
to zhangguanheng66/pytorch
that referenced
this pull request
May 6, 2019
Summary: Pull Request resolved: pytorch#19271 allow DDP to take multi-gpu models Reviewed By: pietern Differential Revision: D14822375 fbshipit-source-id: 1eebfaa33371766d3129f0ac6f63a573332b2f1c
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.
Summary: allow DDP to take multi-gpu models
Differential Revision: D14822375