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

Jump to bottom

Move loading best adapter to the trainer class #487

Merged
merged 3 commits into from
Feb 9, 2023
Merged

Move loading best adapter to the trainer class #487

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
afd2692
Move loading best adapter to the trainer class
MaBeHen Jan 28, 2023
c27bc62
format
MaBeHen Jan 28, 2023
1142109
Merge branch 'adapter-hub:master' into master
MaBeHen Feb 9, 2023
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
43 changes: 22 additions & 21 deletions src/transformers/adapters/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,6 +217,28 @@ def _load_heads(self, resume_from_checkpoint):
):
self.model.load_head(os.path.join(resume_from_checkpoint, file_name))

def _load_best_model(self):
model = self.model_wrapped if is_sagemaker_mp_enabled() else self.model
logger.info(
f"Loading best adapter(s) from {self.state.best_model_checkpoint} (score: {self.state.best_metric})."
)
# attempt to re-load all adapters from checkpoint
for adapter in model.config.adapters.adapters:
adapter_dir = os.path.join(self.state.best_model_checkpoint, adapter)
if os.path.exists(adapter_dir):
model.load_adapter(adapter_dir)
if self.train_adapter_fusion:
logger.info(
f"Loading best adapter fusion(s) from {self.state.best_model_checkpoint} (score:"
f" {self.state.best_metric})."
)
# attempt to re-load all adapter fusions from checkpoint
for fusion in model.config.adapters.fusions:
fusion_dir = os.path.join(self.state.best_model_checkpoint, fusion)
if os.path.exists(fusion_dir):
model.load_adapter_fusion(fusion_dir)
model.to(self.args.device)


class AdapterTrainerCallback(TrainerCallback):
def __init__(self, trainer):
Expand All @@ -232,27 +254,6 @@ def on_train_begin(self, args: TrainingArguments, state: TrainerState, control:
" method"
)

def on_train_end(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
model = kwargs.pop("model")
if args.load_best_model_at_end and state.best_model_checkpoint is not None:

logger.info(f"Loading best adapter(s) from {state.best_model_checkpoint} (score: {state.best_metric}).")
# attempt to re-load all adapters from checkpoint
for adapter in model.config.adapters.adapters:
adapter_dir = os.path.join(state.best_model_checkpoint, adapter)
if os.path.exists(adapter_dir):
model.load_adapter(adapter_dir)
if self.trainer.train_adapter_fusion:
logger.info(
f"Loading best adapter fusion(s) from {state.best_model_checkpoint} (score: {state.best_metric})."
)
# attempt to re-load all adapter fusions from checkpoint
for fusion in model.config.adapters.fusions:
fusion_dir = os.path.join(state.best_model_checkpoint, fusion)
if os.path.exists(fusion_dir):
model.load_adapter_fusion(fusion_dir)
model.to(args.device)

def on_step_end(self, args: TrainingArguments, state: TrainerState, control: TrainerControl, **kwargs):
# apply adapter fusion weight regularization on the value matrix
model = kwargs.pop("model")
Expand Down