Skip to content

Set default padding_multiple to 32 in DPTrainer #291

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

Open
copybara-service wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Set default padding_multiple to 32 in DPTrainer #291

Show file tree
Hide file tree
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
5 changes: 3 additions & 2 deletions jax_privacy/experimental/training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,7 +135,8 @@ class DPTrainer:
optimizer: An ``AugmentedGradientTransformation`` or a plain
``optax.GradientTransformation``.
padding_multiple: If set, batch sizes are padded to a multiple of this
value.
value. Padding reduces JIT recompilations from variable Poisson batch
sizes, since each unique batch shape triggers a separate XLA compilation.
"""

plan: execution_plan.DPExecutionPlan
Expand All @@ -144,7 +145,7 @@ class DPTrainer:
aug_optimizers.AugmentedGradientTransformation
| optax.GradientTransformation
)
padding_multiple: int = 1
padding_multiple: int = 32

def train_step(
self,
Expand Down
10 changes: 10 additions & 0 deletions tests/experimental/training_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -126,6 +126,16 @@ def test_padding_multiple(self):

self.assertEqual(int(state.step), 2)

def test_default_padding_multiple(self):
"""Default padding_multiple should be 32 to prevent JIT recompilation."""
plan = _make_plan(iterations=1)
trainer = training.DPTrainer(
plan=plan,
loss_fn=_quadratic_loss,
optimizer=optax.sgd(0.01),
)
self.assertEqual(trainer.padding_multiple, 32)

def test_zero_iterations_config_raises(self):
"""BandMFConfig requires iterations >= 1."""
with self.assertRaises(Exception):
Expand Down
Loading