Skip to content

Multi-GPU training does not data-parallelize: the DDP wrapper is bypassed #49

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
TonyChen06 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Multi-GPU training does not data-parallelize: the DDP wrapper is bypassed #49

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
9 changes: 9 additions & 0 deletions curriculum_learning.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1115,6 +1115,15 @@ def _train_stage(
loss = self._get_model().compute_loss(batch)
loss.backward()

# _get_model() unwraps DDP, so its gradient sync is bypassed; average grads manually.
if dist.is_initialized():
for p in self._get_model().parameters():
if not p.requires_grad:
continue
if p.grad is None:
p.grad = torch.zeros_like(p)
dist.all_reduce(p.grad, op=dist.ReduceOp.AVG)

# Handle gradient clipping for distributed training
clip_grad_norm_(self._get_model().parameters(), GRAD_CLIP_NORM)

Expand Down