Skip to content

Fix: prevent loading best model when PEFT adapters are active (#3056) #3470

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 2 commits into from
Jul 29, 2025
Merged

Fix: prevent loading best model when PEFT adapters are active (#3056) #3470

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8c4b06d
Fix: prevent loading best model when PEFT adapters are active (#3056)
sahibpreetsingh12 Jul 28, 2025
f6173cb
Reuse _load_from_checkpoint in _load_best_model
tomaarsen Jul 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
17 changes: 5 additions & 12 deletions sentence_transformers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -594,26 +594,19 @@ def evaluation_loop(
def _load_best_model(self) -> None:
# Attempt to load the model from self.state.best_model_checkpoint
logger.info(f"Loading best model from {self.state.best_model_checkpoint} (score: {self.state.best_metric}).")
try:
dummy_model = self.model.__class__(
self.state.best_model_checkpoint,
trust_remote_code=self.model.trust_remote_code,
)
except Exception as exc:
logger.error(f"Could not load the best model from {self.state.best_model_checkpoint}. Error: {str(exc)}")
return

# Store the best model checkpoint in the model card
try:
if checkpoint := self.state.best_model_checkpoint:
step = checkpoint.rsplit("-", 1)[-1]
self.model.model_card_data.set_best_model_step(int(step))
except Exception:
pass

# Ideally, the only changes between self.model and the dummy model are the weights
# so we should be able to just copy the state dict
self.model.load_state_dict(dummy_model.state_dict())
try:
self._load_from_checkpoint(self.state.best_model_checkpoint)
except Exception as exc:
logger.error(f"Could not load the best model from {self.state.best_model_checkpoint}. Error: {str(exc)}")
return

def validate_column_names(self, dataset: Dataset, dataset_name: str | None = None) -> None:
if isinstance(dataset, dict):
Expand Down
Loading