Skip to content

Cleanup, use mean reduction #472

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
josephdviviano merged 1 commit into from
Feb 16, 2026
Merged

Cleanup, use mean reduction #472

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
6 changes: 1 addition & 5 deletions tutorials/examples/train_hypergrid.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -928,13 +928,9 @@ def _model_builder() -> Tuple[GFlowNet, torch.optim.Optimizer]:
env,
training_objects, # type: ignore
recalculate_all_logprobs=(not is_on_policy_iter),
reduction="sum" if args.distributed or args.loss == "SubTB" else "mean", # type: ignore
reduction="sum" if args.loss == "SubTB" else "mean", # type: ignore
Copy link
Collaborator

@josephdviviano josephdviviano Feb 16, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry - what was the bug here? This can produce unintended behaviour as you scale up with DDP, but maybe it actually produces incorrect scaling for distributed async training. We should discuss on the call tomorrow.

)

# Normalize the loss by the local batch size if distributed.
if args.distributed:
loss = loss / (per_node_batch_size)

# Barrier.
with Timer(
timing, "barrier 0", enabled=(args.timing and args.distributed)
Expand Down