v2.21.0
Breaking Changes📦 tensorflowView on GitHub →
⚠ 2 breaking✨ 8 features🐛 1 fixes🔧 7 symbols
Summary
Release 2.21.0 removes support for Python 3.9 and the TensorBoard dependency, while introducing significant quantization and data type support enhancements in tf.lite and JPEG XL support in tf.image.
⚠️ Breaking Changes
- Support for Python 3.9 has been removed.
- The TensorBoard (TB) dependency has been removed.
Migration Steps
- If you rely on Python 3.9, you must upgrade to Python 3.10 or newer.
- Remove any direct dependencies or usage related to the TensorBoard package if it was previously managed by TensorFlow.
✨ New Features
- tf.lite: Adds int8 and int16x8 support for SQRT operator.
- tf.lite: Adds int16x8 support for EQUAL and NOT_EQUAL operators.
- tf.lite: Adds support for int2 type.
- tf.lite: Adds support for int2/int4 in tfl.cast.
- tf.lite: Adds support for SRQ int2 in tfl.fully_connected.
- tf.lite: Adds support for int4 in tfl.slice.
- tf.lite: Adds support for uint4 type.
- tf.image: Adds JPEG XL support in decode_image.
🐛 Bug Fixes
- tf.data: Adds `NoneTensorSpec` to the public API so that `None`s in `element_spec` can be identified via `isinstance(..., tf.NoneTensorSpec)`.