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

Making gradient clipping optional & max gradient norm variable #3240

Merged
merged 3 commits into from
Aug 8, 2023
Merged

Making gradient clipping optional & max gradient norm variable #3240

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
00d9d49
Made gradient clipping optional & max gradient norm variable
plonerma May 16, 2023
0de88ce
Merge branch 'master' into variable_grad_norm_clipping
alanakbik Aug 8, 2023
21e6ade
Ruff
alanakbik Aug 8, 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
6 changes: 5 additions & 1 deletion flair/trainers/trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -298,6 +298,7 @@ def train_custom(
optimizer: Type[torch.optim.Optimizer] = SGD,
train_with_dev: bool = False,
train_with_test: bool = False,
max_grad_norm: Optional[float] = 5.0,
# evaluation and monitoring
main_evaluation_metric: Tuple[str, str] = ("micro avg", "f1-score"),
monitor_test: bool = False,
Expand Down Expand Up @@ -345,6 +346,8 @@ def train_custom(
monitor_train_sample: Set this to evaluate on a sample of the train data at the end of each epoch.
If you set an int, it will sample this many sentences to evaluate on. If you set a float, it will sample
a percentage of data points from train.
max_grad_norm (Optional[float]): If not None, gradients are clipped to this value before an optimizer.step is
called.
use_final_model_for_eval (bool): If True, the final model is used for the final evaluation. If False, the
model from the best epoch as determined by main_evaluation_metric is used for the final evaluation.
gold_label_dictionary_for_eval: Set to force evaluation to use a particular label dictionary
Expand Down Expand Up @@ -594,7 +597,8 @@ def train_custom(

# do the optimizer step
scaler.unscale_(self.optimizer)
torch.nn.utils.clip_grad_norm_(self.model.parameters(), 5.0)
if max_grad_norm is not None:
torch.nn.utils.clip_grad_norm_(self.model.parameters(), max_grad_norm)
scale_before = scaler.get_scale()
scaler.step(self.optimizer)
scaler.update()
Expand Down