Skip to content

fix(batch_norm): allow string cross_replica_axis for vmap support #859

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
aditya161205 wants to merge 2 commits into
base: main
Choose a base branch
from
Open

fix(batch_norm): allow string cross_replica_axis for vmap support #859

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b6ad578
fix(batch_norm): allow string cross_replica_axis for vmap support
aditya161205 Jan 8, 2026
403b6a0
trigger cla check
aditya161205 Jan 8, 2026
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
21 changes: 13 additions & 8 deletions haiku/_src/batch_norm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -170,14 +170,19 @@ def __call__(
mean = jnp.mean(inputs, axis, keepdims=True)
mean_of_squares = jnp.mean(jnp.square(inputs), axis, keepdims=True)
if self.cross_replica_axis:
mean = jax.lax.pmean(
mean,
axis_name=self.cross_replica_axis,
axis_index_groups=self.cross_replica_axis_index_groups)
mean_of_squares = jax.lax.pmean(
mean_of_squares,
axis_name=self.cross_replica_axis,
axis_index_groups=self.cross_replica_axis_index_groups)
try:
mean = jax.lax.pmean(
mean,
axis_name=self.cross_replica_axis,
axis_index_groups=self.cross_replica_axis_index_groups)
mean_of_squares = jax.lax.pmean(
mean_of_squares,
axis_name=self.cross_replica_axis,
axis_index_groups=self.cross_replica_axis_index_groups)
except NameError:
# If the axis is not bound (e.g. during init or non-mapped execution),
# we skip the sync and fall back to local statistics.
pass
var = mean_of_squares - jnp.square(mean)
else:
mean = self.mean_ema.average.astype(inputs.dtype)
Expand Down