Skip to content

Check for DTensor due to FSDP2 in OptimizerMonitor #3921

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Check for DTensor due to FSDP2 in OptimizerMonitor #3921

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
10 changes: 10 additions & 0 deletions composer/callbacks/optimizer_monitor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,11 @@
from composer.loggers import Logger
from composer.utils import dist

try:
from torch.distributed.tensor import DTensor
except ImportError:
DTensor = None

__all__ = ['OptimizerMonitor']


Expand Down Expand Up @@ -86,6 +91,11 @@ def batch_end(self, state: State, logger: Logger):
param_grad_norm = torch.linalg.vector_norm(p.grad)
optimizer_metrics[f'l2_norm/grad/{name}'] = param_grad_norm

# Added for compatibility with FSDP2, since all-reduce on random DTensor can have issues
for name, metric_val in optimizer_metrics.items():
if isinstance(metric_val, DTensor):
optimizer_metrics[name] = metric_val.to_local()

if state.fsdp_enabled and dist.get_world_size() > 0 and self.log_optimizer_metrics:
# If FSDP is enabled, the optimizer state lives on different ranks and must be reduced
# and combined before we can compute metrics.
Expand Down