Skip to content

Fix local batch size #467

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
younik merged 2 commits into from
Feb 5, 2026
Merged

Fix local batch size #467

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7cbea56
fix minibatch size
younik Feb 4, 2026
45e2166
second fix
younik Feb 5, 2026
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
12 changes: 2 additions & 10 deletions tutorials/examples/train_hypergrid.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -807,7 +807,7 @@ def _model_builder() -> Tuple[GFlowNet, torch.optim.Optimizer]:
gflownet = gflownet.to(device)

n_iterations = ceil(args.n_trajectories / args.batch_size)
per_node_batch_size = args.batch_size // distributed_context.world_size
per_node_batch_size = args.batch_size // distributed_context.num_training_ranks
Copy link
Collaborator

@josephdviviano josephdviviano Feb 5, 2026

Choose a reason for hiding this comment

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

good catch!

modes_found = set()
# n_pixels_per_mode = round(env.height / 10) ** env.ndim
# Note: on/off-policy depends on the current strategy; recomputed inside the loop.
Expand All @@ -828,14 +828,6 @@ def _model_builder() -> Tuple[GFlowNet, torch.optim.Optimizer]:
)
prof.start()

if args.distributed:
# Create and start error handler.
def cleanup():
logger.info("Process %d: Cleaning up...", rank)

rank = torch.distributed.get_rank()
torch.distributed.get_world_size()

# Initialize some variables before the training loop.
timing = {}
time_start = time.time()
Expand Down Expand Up @@ -897,7 +889,7 @@ def cleanup():
)
trajectories = gflownet.sample_trajectories(
env,
n=args.batch_size,
n=per_node_batch_size,
save_logprobs=is_on_policy_iter, # Reuse on-policy log-probs.
save_estimator_outputs=not is_on_policy_iter, # Off-policy caches estimator outputs.
epsilon=float(getattr(args, "agent_epsilon", 0.0)),
Expand Down
Loading