Skip to content

fix: workaround for optimizer offload #2239

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

Draft
yuki-97 wants to merge 1 commit into
base: super-v3
Choose a base branch
from
+1 −2
Draft

fix: workaround for optimizer offload #2239

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
3 changes: 1 addition & 2 deletions nemo_rl/models/policy/workers/megatron_policy_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1133,12 +1133,11 @@ def prepare_for_training(self, *args, **kwargs):
self.model.train()

# Move optimizer state to CUDA if it exists
# colocated generation will always offload optimizer to cuda before refit
# Always move to CUDA to ensure the optimizer is on GPU before training; this is a no-op if it's already on GPU.
if (
hasattr(self, "optimizer")
and self.optimizer is not None
and not self.optimizer_cpu_offload
and (self.offload_optimizer_for_logprob or self.is_generation_colocated)
):
self.move_optimizer("cuda")

Expand Down
Loading