Skip to content

chore: add assert for dtensor v2 cpu offload #1817

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
terrykong merged 1 commit into from
Jan 30, 2026
Merged

chore: add assert for dtensor v2 cpu offload #1817

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
8 changes: 7 additions & 1 deletion nemo_rl/models/automodel/setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,8 @@ def validate_and_prepare_config(
if "generation" in config and config["generation"] is not None:
is_generation_colocated = config["generation"]["colocated"]["enabled"]

# Set NCCL environment variable
# Explicitly set NCCL_CUMEM_ENABLE to 1 to avoid the P2P initialization error for PyNCCLCommunicator.
# See https://github.com/NVIDIA-NeMo/RL/issues/564 for more details.
if not is_generation_colocated:
os.environ["NCCL_CUMEM_ENABLE"] = "1"

Expand Down Expand Up @@ -282,6 +283,11 @@ def setup_distributed(

# Force setup distributed for world size 1 as FSDP2Manager skips it
if world_size == 1:
if cpu_offload:
raise NotImplementedError(
"CPUOffload doesn't work on single GPU for AutoModel. "
"If you need this feature, please file an issue on https://github.com/NVIDIA-NeMo/Automodel."
)
manager._setup_distributed()

return manager
Expand Down
Loading