Skip to content

Fix async checkpoint timing in DCP recipe #3688

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

Fix async checkpoint timing in DCP recipe #3688

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
8 changes: 5 additions & 3 deletions recipes_source/distributed_async_checkpoint_recipe.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -257,12 +257,14 @@ checkpoint requests users can take advantage of direct memory access to speed up
for step in range(10):
optimizer.zero_grad()
model(torch.rand(8, 16, device="cuda")).sum().backward()
optimizer.step()

state_dict = { "app": AppState(model, optimizer) }
# Wait for the previous checkpoint to finish before optimizer.step() modifies weights in-place
if checkpoint_future is not None:
# waits for checkpointing to finish, avoiding queuing more then one checkpoint request at a time
checkpoint_future.result()

optimizer.step()

state_dict = { "app": AppState(model, optimizer) }
checkpoint_future = dcp.async_save(state_dict, storage_writer=writer, checkpoint_id=f"{CHECKPOINT_DIR}_step{step}")

cleanup()
Expand Down