Skip to content

[FSDP] Don't move full tensor to device in deferred_init #4819

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
Aug 22, 2023
Merged

[FSDP] Don't move full tensor to device in deferred_init #4819

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 1 addition & 3 deletions torch_xla/distributed/fsdp/_init_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,6 @@ def _materialize_module(
is_torchdistX_deferred_init = (
not is_meta_module and _TORCHDISTX_AVAIL and
any(fake.is_fake(param) for param in managed_params))
materialization_device = xm.xla_device()
if (is_meta_module or
is_torchdistX_deferred_init) and param_init_fn is not None:
if not callable(param_init_fn):
Expand All @@ -65,7 +64,7 @@ def _materialize_module(
return True
elif is_meta_module:
# Run default meta device initialization
module.to_empty(device=materialization_device)
module.to_empty(device="cpu")
try:
with torch.no_grad():
module.reset_parameters() # type: ignore[operator]
Expand All @@ -78,7 +77,6 @@ def _materialize_module(
elif is_torchdistX_deferred_init:
# Run default torchdistX initialization
deferred_init.materialize_module(module, check_fn=deferred_init_check_fn)
module.to(materialization_device)
return True
return False

Expand Down