Skip to content

Distillation optimizer fix #3446

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
copybara-service merged 1 commit into from
Mar 18, 2026
Merged

Distillation optimizer fix #3446

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
6 changes: 5 additions & 1 deletion src/maxtext/trainers/post_train/distillation/train_distill.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -199,7 +199,11 @@ class MaxTextDistillationTrainer(peft_trainer.PeftTrainer):
"""

def __init__(self, model, strategy, optimizer, training_config, **kwargs):
super().__init__(model=model, optimizer=optimizer, training_config=training_config, **kwargs)
# We pass a dummy optimizer to the base PeftTrainer temporarily to prevent PeftTrainer from eagerly
# allocating massive optimizer states for the entire ModelBundle (including the frozen teacher) before
# redefining the trainer optimizer here.
dummy_optimizer = optax.set_to_zero()
super().__init__(model=model, optimizer=dummy_optimizer, training_config=training_config, **kwargs)

self.strategy = strategy

Expand Down
Loading