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.

Sign up for GitHub

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

Save and load a module trained with DataParallel or DistributedDataParallel #55

Merged
merged 3 commits into from
Oct 21, 2020
Merged

Save and load a module trained with DataParallel or DistributedDataParallel #55

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3a24a2e
add support dp and ddp in agent.save and agent.load
shu65 Sep 3, 2020
a3033cf
fix flake8 errors
shu65 Sep 3, 2020
867c1bb
fix black errors
shu65 Oct 13, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions pfrl/agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,11 @@ def __save(self, dirname: str, ancestors: List[Any]):
), "Avoid an infinite loop"
attr_value.__save(os.path.join(dirname, attr), ancestors)
else:
if isinstance(
attr_value,
(torch.nn.parallel.DistributedDataParallel, torch.nn.DataParallel),
):
attr_value = attr_value.module
torch.save(
attr_value.state_dict(), os.path.join(dirname, "{}.pt".format(attr))
)
Expand All @@ -125,6 +130,11 @@ def __load(self, dirname: str, ancestors: List[Any]) -> None:
), "Avoid an infinite loop"
attr_value.load(os.path.join(dirname, attr))
else:
if isinstance(
attr_value,
(torch.nn.parallel.DistributedDataParallel, torch.nn.DataParallel),
):
attr_value = attr_value.module
attr_value.load_state_dict(
torch.load(
os.path.join(dirname, "{}.pt".format(attr)), map_location
Expand Down
31 changes: 31 additions & 0 deletions tests/test_agent.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,3 +91,34 @@ def test_loop2(self):
# Otherwise it seems to raise OSError: [Errno 63] File name too long
with self.assertRaises(AssertionError):
parent1.save(dirname)

def test_with_data_parallel(self):
parent = Parent()
parent.link.param.detach().numpy()[:] = 1
parent.child.link.param.detach().numpy()[:] = 2
parent.link = torch.nn.DataParallel(parent.link)

# Save
dirname = tempfile.mkdtemp()
parent.save(dirname)
self.assertTrue(os.path.isdir(dirname))
self.assertTrue(os.path.isfile(os.path.join(dirname, "link.pt")))
self.assertTrue(os.path.isdir(os.path.join(dirname, "child")))
self.assertTrue(os.path.isfile(os.path.join(dirname, "child", "link.pt")))

# Load Parent without data parallel
parent = Parent()
self.assertEqual(int(parent.link.param.detach().numpy()), 0)
self.assertEqual(int(parent.child.link.param.detach().numpy()), 0)
parent.load(dirname)
self.assertEqual(int(parent.link.param.detach().numpy()), 1)
self.assertEqual(int(parent.child.link.param.detach().numpy()), 2)

# Load Parent with data parallel
parent = Parent()
parent.link = torch.nn.DataParallel(parent.link)
self.assertEqual(int(parent.link.module.param.detach().numpy()), 0)
self.assertEqual(int(parent.child.link.param.detach().numpy()), 0)
parent.load(dirname)
self.assertEqual(int(parent.link.module.param.detach().numpy()), 1)
self.assertEqual(int(parent.child.link.param.detach().numpy()), 2)