Skip to content

[minor] fix: use init_empty_weights instead of torch.device("meta") #1587

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
vermouth1992 merged 1 commit into from
May 21, 2025
Merged

[minor] fix: use init_empty_weights instead of torch.device("meta") #1587

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
3 changes: 2 additions & 1 deletion verl/utils/checkpoint/fsdp_checkpoint_manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@

import torch
import torch.distributed
from accelerate import init_empty_weights
from torch.distributed.fsdp import FullStateDictConfig, ShardedOptimStateDictConfig, ShardedStateDictConfig, StateDictType
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from transformers import GenerationConfig, PreTrainedTokenizer, ProcessorMixin
Expand Down Expand Up @@ -197,7 +198,7 @@ def save_checkpoint(self, local_path: str, hdfs_path: str = None, global_step: i
else:
raise NotImplementedError(f"Unknown architecture {model_config['architectures']}")

with torch.device("meta"):
with init_empty_weights():
save_model = auto_model_cls.from_config(model_config, torch_dtype=torch.bfloat16)
save_model.to_empty(device="cpu")

Expand Down
Loading