Skip to content

zarr soft deprecation #2004

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
ko3n1g merged 5 commits into from
Oct 29, 2025
Merged

zarr soft deprecation #2004

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
412aa17
zarr soft deprecation
dimapihtar Oct 28, 2025
620635c
fix style
dimapihtar Oct 28, 2025
5ad362f
fix style
dimapihtar Oct 28, 2025
602b1cd
Merge branch 'main' into zarr_soft_deprecation
ko3n1g Oct 28, 2025
0161c26
Merge branch 'main' into zarr_soft_deprecation
ko3n1g Oct 28, 2025
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
15 changes: 11 additions & 4 deletions megatron/core/dist_checkpointing/strategies/zarr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,9 +77,9 @@ class ZarrSaveShardedStrategy(SaveShardedStrategy):

def __init__(self, backend: str, version: int):
super().__init__(backend, version)
logger.warning(
f"`zarr` distributed checkpoint backend is deprecated."
" Please switch to PyTorch Distributed format (`torch_dist`)."
raise CheckpointingException(
"`zarr` distributed checkpoint backend is no longer supported. "
"Please switch to PyTorch Distributed format (`torch_dist`)."
)

def save(self, sharded_state_dict: ShardedStateDict, checkpoint_dir: Union[str, Path]):
Expand Down Expand Up @@ -197,6 +197,13 @@ def _create_zarr_array(sharded_tensor: ShardedTensor, checkpoint_dir: Path):
class ZarrLoadShardedStrategy(LoadShardedStrategy):
"""Load strategy for the Zarr backend."""

def __init__(self, backend: str, version: int):
super().__init__(backend, version)
raise CheckpointingException(
"`zarr` distributed checkpoint backend is no longer supported. "
"Please switch to PyTorch Distributed format (`torch_dist`)."
)

def load(self, sharded_state_dict: ShardedStateDict, checkpoint_dir: Union[str, Path]):
if isinstance(checkpoint_dir, str):
checkpoint_dir = Path(checkpoint_dir)
Expand Down Expand Up @@ -310,7 +317,7 @@ def pad_to_expected_shape(x: torch.Tensor, expected_sharded_ten: ShardedTensor):
return torch.nn.functional.pad(x, pad_args)

# unsqueeze and squeeze to get shapes supported by cudnn
print(f"Replicating last row for {expected_sharded_ten.key}")
logger.info(f"Replicating last row for {expected_sharded_ten.key}")
if x.dtype == torch.bfloat16:
return (
torch.nn.functional.pad(x.float().unsqueeze(0), pad_args, mode="replicate")
Expand Down
Loading