Skip to content

Fix: Use DDP for PPO traning will cause AttributeError: 'DistributedDataParallel' object has no attribute 'config' error #5822

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
merged 3 commits into from
Sep 19, 2025
+16 −5
Merged

Fix: Use DDP for PPO traning will cause AttributeError: 'DistributedDataParallel' object has no attribute 'config' error #5822

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5016a0d
fix: unwrap model for ppo trainer in _save_checkpoint
kiritoxkiriko Sep 15, 2025
62bd950
Update swift/trainers/rlhf_trainer/ppo_trainer.py
kiritoxkiriko Sep 15, 2025
6aa4c30
lint: reformat ppo trainer
kiritoxkiriko Sep 19, 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
21 changes: 16 additions & 5 deletions swift/trainers/rlhf_trainer/ppo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,6 @@
from contextlib import contextmanager
from typing import Optional

import transformers
from packaging import version
from torch.utils.data import DataLoader
from transformers import PreTrainedModel, Trainer
from trl import PPOTrainer as HFPPOTrainer
Expand Down Expand Up @@ -38,8 +36,14 @@ def __init__(self, model: PreTrainedModel, ref_model: PreTrainedModel, *_args, *
with self._patch_dataloader(kwargs['data_collator']):
new_kwargs = {
k: v
for k, v in kwargs.items()
if k in ['train_dataset', 'data_collator', 'reward_model', 'value_model', 'eval_dataset', 'callbacks']
for k, v in kwargs.items() if k in [
'train_dataset',
'data_collator',
'reward_model',
'value_model',
'eval_dataset',
'callbacks',
]
}
parameters = inspect.signature(ppo_trainer_init).parameters
if 'config' in parameters:
Expand All @@ -63,7 +67,14 @@ def train(self, *args, **kwargs):

def _save_checkpoint(self, *args, **kwargs):
kwargs.pop('metrics', None)
return super()._save_checkpoint(*args, **kwargs)

backup_model = self.model
try:
# Unwrap model if needed
self.model = self.accelerator.unwrap_model(self.model)
return super()._save_checkpoint(*args, **kwargs)
finally:
self.model = backup_model

def save_model(self, output_dir: Optional[str] = None, _internal_call: bool = False):
# https://github.com/huggingface/trl/issues/2122
Expand Down
Loading