Skip to content

support optim cases #21

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
Chao1Han wants to merge 2 commits into
base: master
Choose a base branch
from
Open

support optim cases #21

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e992eaa
support optim cases
Chao1Han Jul 21, 2025
62eb2ca
blacklist gloo
Chao1Han Jul 29, 2025
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
16 changes: 6 additions & 10 deletions torch/distributed/algorithms/ddp_comm_hooks/ddp_zero_hook.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -232,13 +232,11 @@ def hook_with_zero_step(
)
ddp_ref = weakref.ref(ddp)

# NOTE: Gloo may hang with this overlapping approach, so we require
# NCCL/HCCL backend for now; see https://github.com/pytorch/pytorch/issues/62300
# NOTE: Gloo may hang with this overlapping approach; see https://github.com/pytorch/pytorch/issues/62300
pg = dist.get_backend(ddp_ref().process_group) # type: ignore[union-attr]
if (pg != dist.Backend.NCCL) and (pg != "hccl"):
if pg == dist.Backend.GLOO:
raise RuntimeError(
"Overlapping DDP with ZeRO using this approach currently requires "
"NCCL/HCCL backend to avoid hangs"
"Gloo backend using Overlapping DDP with ZeRO may meet hangs"
)

if shard_buckets:
Expand Down Expand Up @@ -394,13 +392,11 @@ def hook_with_zero_step_interleaved(
)
ddp_ref = weakref.ref(ddp)

# NOTE: Gloo may hang with this overlapping approach, so we require
# NCCL/HCCL backend for now; see https://github.com/pytorch/pytorch/issues/62300
# NOTE: Gloo may hang with this overlapping approach; see https://github.com/pytorch/pytorch/issues/62300
pg = dist.get_backend(ddp_ref().process_group) # type: ignore[union-attr]
if (pg != dist.Backend.NCCL) and (pg != "hccl"):
if pg == dist.Backend.GLOO:
raise RuntimeError(
"Overlapping DDP with ZeRO using this approach currently requires "
"NCCL/HCCL backend to avoid hangs"
"Gloo backend using Overlapping DDP with ZeRO may meet hangs"
)

if shard_buckets:
Expand Down