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.

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

init #2322

Merged
merged 3 commits into from
Jul 25, 2024
Merged

init #2322

Changes from 1 commit
Commits
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
Next Next commit
init
  • Loading branch information
vmoens committed Jul 25, 2024
commit 6268443634f7d9ff16e2ff653b6580c18a178a6b
17 changes: 16 additions & 1 deletion torchrl/modules/models/multiagent.py
Original file line number Diff line number Diff line change
Expand Up @@ -70,8 +70,23 @@ def __init__(
break
self.initialized = initialized
self._make_params(agent_networks)
# We make sure all params and buffers are on 'meta' device
# To do this, we set the device keyword arg to 'meta', we also temporarily change
# the default device. Finally, we convert all params to 'meta' tensors that are not params.
kwargs["device"] = "meta"
self.__dict__["_empty_net"] = self._build_single_net(**kwargs)
with torch.device("meta"):
try:
self._empty_net = self._build_single_net(**kwargs)
except NotImplementedError as err:
if "Cannot copy out of meta tensor" in str(err):
raise RuntimeError(
"The network was built using `factory().to(device), build the network directly "
"on device using `factory(device=device)` instead."
)
# Remove all parameters
TensorDict.from_module(self._empty_net).data.to("meta").to_module(
self._empty_net
)

@property
def vmap_randomness(self):
Expand Down
Loading