Skip to content

[BugFix] update_policy_weights_() with cudagraph #3003

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

Merged
merged 1 commit into from
Jun 16, 2025
+41 −37
Merged

[BugFix] update_policy_weights_() with cudagraph #3003

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
78 changes: 41 additions & 37 deletions test/test_collector.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1483,8 +1483,9 @@ def env_fn(seed):
assert_allclose_td(data10, data20)

@pytest.mark.parametrize("use_async", [False, True])
@pytest.mark.parametrize("cudagraph", [False, True])
@pytest.mark.skipif(not torch.cuda.is_available(), reason="no cuda device found")
def test_update_weights(self, use_async):
def test_update_weights(self, use_async, cudagraph):
def create_env():
return ContinuousActionVecMockEnv()

Expand All @@ -1504,48 +1505,51 @@ def create_env():
storing_device=[torch.device("cuda:0")] * 3,
frames_per_batch=20,
cat_results="stack",
cudagraph_policy=cudagraph,
)
# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
for worker in range(3):
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
torch.testing.assert_close(
state_dict[f"worker{worker}"]["policy_state_dict"][k],
policy_state_dict[k].cpu(),
)

# change policy weights
for p in policy.parameters():
p.data += torch.randn_like(p)

# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
# check they don't match
for worker in range(3):
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
with pytest.raises(AssertionError):
try:
# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
for worker in range(3):
assert "policy_state_dict" in state_dict[f"worker{worker}"], state_dict[f"worker{worker}"].keys()
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
torch.testing.assert_close(
state_dict[f"worker{worker}"]["policy_state_dict"][k],
policy_state_dict[k].cpu(),
)

# update weights
collector.update_policy_weights_()

# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
for worker in range(3):
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
torch.testing.assert_close(
state_dict[f"worker{worker}"]["policy_state_dict"][k],
policy_state_dict[k].cpu(),
)

collector.shutdown()
del collector
# change policy weights
for p in policy.parameters():
p.data += torch.randn_like(p)

# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
# check they don't match
for worker in range(3):
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
with pytest.raises(AssertionError):
torch.testing.assert_close(
state_dict[f"worker{worker}"]["policy_state_dict"][k],
policy_state_dict[k].cpu(),
)

# update weights
collector.update_policy_weights_()

# collect state_dict
state_dict = collector.state_dict()
policy_state_dict = policy.state_dict()
for worker in range(3):
for k in state_dict[f"worker{worker}"]["policy_state_dict"]:
torch.testing.assert_close(
state_dict[f"worker{worker}"]["policy_state_dict"][k],
policy_state_dict[k].cpu(),
)
finally:
collector.shutdown()
del collector


class TestCollectorDevices:
Expand Down
Loading