v2.20.1

What's Changed

• Fix save_weights when saving weights in TF format with a remote path. by @copybara-service[bot] in #837
• Automated Code Change by @copybara-service[bot] in #839
• This change proactively relaxes floating point comparisons in unit tests to avoid breaking them when updating the Eigen library in cl/747497592. by @copybara-service[bot] in #841
• Internal change in a BUILD file. by @copybara-service[bot] in #840
• Update API goldens for Python 3.12. by @copybara-service[bot] in #842
• Adds support for Call-Context Arguments by @copybara-service[bot] in #843
• Disallow pickle loading in npz files. by @copybara-service[bot] in #849
• Initial implementation of PR-gain AUC metrics based on http://people.cs.bris.ac.uk/~flach/PRGcurves/PRcurves.pdf by @copybara-service[bot] in #845
• Reformat code that doesn't follow the standard. by @copybara-service[bot] in #859
• Add sample weight support; pass training=True in SAM's train_step to support layers like BatchNorm and Dropout. by @copybara-service[bot] in #858
• Add hour as a suffix to the wheel name. by @copybara-service[bot] in #860
• Add rng_type to BaseRandomLayer.get_config(). by @copybara-service[bot] in #852
• Patch version 0.20.1 by @sachinprasadhs in #861

New Contributors

• @sachinprasadhs made their first contribution in #861

Full Changelog: v2.19.0...v2.20.1

v2.19.0

What's Changed

• Update tf_keras Nightly Version to 2.19.0 by @copybara-service in #802
• Fix pip build by @jonas-eschle in #779
• Preserve tuples in TypeSpecs. by @copybara-service in #809
• Fix serialization / deserialization. by @copybara-service in #808
• Make UnitNormalization layer stateless. by @copybara-service in #810
• A small fix for Python 3.12 compatibility by @copybara-service in #813
• Explictly use Py310 in github workflows by @sampathweb in #818
• Allow passing a custom cache_dir to tf.keras.datasets.load_data. This is helpful when the default location ~/.keras in home directory has limited disk space. by @copybara-service in #817
• Explictly use Py310 in github workflows by @sampathweb in #820
• PR #820: Explictly use Py310 in github workflows by @copybara-service in #821
• remove deprecated assertDictContainsSubset by @copybara-service in #819
• Fixes for Sequential model with multiple inputs. by @copybara-service in #823
• Fixes "'float' object cannot be interpreted as an integer" in Python 3.12 by @copybara-service in #824
• Remove srcs_version and python_version attributes, as they already default to "PY3" by @copybara-service in #825

New Contributors

• @jonas-eschle made their first contribution in #779

Full Changelog: v2.18.0...v2.19.0

v2.18.0

What's Changed

• Bump version 2.18 by @sampathweb in #781
• Prevent load statements from being combined by using `# buildifier:… by @hertschuh in #783
• Manual sync of github changes. by @copybara-service in #785
• Update version to 2.18 in setup.py. by @copybara-service in #786
• Fix object_registration_test. by @copybara-service in #787
• Update example code in timeseries_dataset.py by @SuryanarayanaY in #14
• [numpy] Fix users of NumPy APIs that are removed in NumPy 2.0. by @copybara-service in #788
• Add support for autocast keyword argument in Layer.add_weight. by @copybara-service in #792
• Update numpy version in requirements.txt to match TF's version. by @copybara-service in #801

New Contributors

• @hertschuh made their first contribution in #783
• @SuryanarayanaY made their first contribution in #14

Full Changelog: v2.17.0...v2.18.0

TF-Keras v2.17.0

This release matches tensorflow==2.17.0.

TF-Keras v2.16.0

This release matches tensorflow==2.16.1.

TF-Keras v2.15.0

This release matches tensorflow==2.15.0.