Skip to content

For sharded weights let's not delete explicitly #2431

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
merged 3 commits into from
Oct 10, 2025
Merged

For sharded weights let's not delete explicitly #2431

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
9f6f33b
For sharded weights let's not delete explicitly
amitsrivastava78 Oct 1, 2025
d95e9d1
removed some unnecessary conditions
amitsrivastava78 Oct 10, 2025
0dfb763
format corrected file
amitsrivastava78 Oct 10, 2025
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
7 changes: 6 additions & 1 deletion keras_hub/src/utils/preset_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -502,10 +502,15 @@ def jax_memory_cleanup(layer):
# For jax, delete all previous allocated memory to avoid temporarily
# duplicating variable allocations. torch and tensorflow have stateful
# variable types and do not need this fix.
# Skip deletion for sharded arrays to avoid breaking references in distributed setups.
if keras.config.backend() == "jax":
for weight in layer.weights:
if getattr(weight, "_value", None) is not None:
weight._value.delete()
# Do not delete sharded arrays, as they may be referenced in JAX's
# distributed computation graph and deletion can cause errors.
if not (hasattr(weight._value, 'sharding')
and weight._value.sharding is not None):
weight._value.delete()


def set_dtype_in_config(config, dtype=None):
Expand Down
Loading