Skip to content
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

Avoid duplicated dist checkpoint save #7555

Merged
merged 1 commit into from
Sep 28, 2023
Merged

Avoid duplicated dist checkpoint save #7555

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
35 changes: 22 additions & 13 deletions nemo/collections/nlp/parts/nlp_overrides.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -291,6 +291,10 @@ def save_checkpoint(
checkpoint_dir = ckpt_to_dir(filepath)

fs = get_filesystem(checkpoint_dir)
if fs.isdir(checkpoint_dir) and dist_checkpointing.check_is_distributed_checkpoint(checkpoint_dir):
logging.info(f'Distributed checkpoint at path {checkpoint_dir} already exists, skipping saving')
return

if is_global_rank_zero():
fs.makedirs(checkpoint_dir, exist_ok=True)

Expand Down Expand Up @@ -465,19 +469,24 @@ def save_to(self, model, save_path: str):
# model weights is a directory
dist_ckpt_dir = ckpt_to_dir(os.path.join(dir_name, self.model_weights_ckpt))
fs = get_filesystem(dist_ckpt_dir)
if is_global_rank_zero():
fs.makedirs(dist_ckpt_dir, exist_ok=True)
sharded_state_dict = model.sharded_state_dict()
# dist checkpoint needs torch.distributed to save the checkpoint
if parallel_state.is_unitialized():

def dummy():
return

if model.trainer.strategy.launcher is not None:
model.trainer.strategy.launcher.launch(dummy, trainer=model.trainer)
model.trainer.strategy.setup_environment()
dist_checkpointing.save(sharded_state_dict=sharded_state_dict, checkpoint_dir=dist_ckpt_dir)

if fs.isdir(dist_ckpt_dir) and dist_checkpointing.check_is_distributed_checkpoint(dist_ckpt_dir):
logging.info(f'Distributed checkpoint at path {dist_ckpt_dir} already exists, skipping saving')
else:
if is_global_rank_zero():
fs.makedirs(dist_ckpt_dir, exist_ok=True)

sharded_state_dict = model.sharded_state_dict()
# dist checkpoint needs torch.distributed to save the checkpoint
if parallel_state.is_unitialized():

def dummy():
return

if model.trainer.strategy.launcher is not None:
model.trainer.strategy.launcher.launch(dummy, trainer=model.trainer)
model.trainer.strategy.setup_environment()
dist_checkpointing.save(sharded_state_dict=sharded_state_dict, checkpoint_dir=dist_ckpt_dir)

else:

Expand Down