Skip to content

perf: disable find_unused_parameters for faster DDP training #68

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
Luodian wants to merge 1 commit into
base: main
Choose a base branch
from
Open

perf: disable find_unused_parameters for faster DDP training #68

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
14 changes: 12 additions & 2 deletions training/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,13 +370,23 @@ def _expand(name, v):
global_step = 0

def wrap_ddp(model):
# Performance optimization: disable find_unused_parameters
#
# find_unused_parameters=True adds overhead from extra autograd graph traversal.
# Analysis shows this training loop does NOT have unused parameters:
# - All backbone outputs flow to loss via pfc() for all data strategies
# - Branching is about input shape/indexing, not skipping modules
# - Only "optional" module is pooling head (gated by use_head), always used
#
# If you encounter "Expected to have finished reduction" errors, re-enable
# with find_unused_parameters=True and debug with TORCH_DISTRIBUTED_DEBUG=INFO
return torch.nn.parallel.DistributedDataParallel(
module=model,
broadcast_buffers=False,
device_ids=[local_rank],
bucket_cap_mb=32,
find_unused_parameters=True,
static_graph=True,
find_unused_parameters=False, # Disabled for performance
static_graph=True, # Enables further optimizations when graph is fixed
)

backbone_ddp = wrap_ddp(backbone)
Expand Down