●Change8

v2.19.0

Breaking Changes
📦 tensorflowView on GitHub →
⚠ 2 breaking✨ 1 features🐛 1 fixes⚡ 1 deprecations🔧 6 symbols

Summary

TensorFlow 2.19 introduces breaking changes to LiteRT constants and deprecates `tf.lite.Interpreter`, adds bfloat16 support to `tfl.Cast`, and stops publishing `libtensorflow` packages.

⚠️ Breaking Changes

  • In the C++ LiteRT API, the public constants `tflite::Interpreter:kTensorsReservedCapacity` and `tflite::Interpreter:kTensorsCapacityHeadroom` have been changed from `constexpr` compile‑time constants to const references, breaking code that expects compile‑time values.
  • The Python class `tf.lite.Interpreter` now emits a deprecation warning and will be removed in TensorFlow 2.20; it has been moved to `ai_edge_litert.interpreter.Interpreter`.

Migration Steps

  1. Replace imports of `tf.lite.Interpreter` with `from ai_edge_litert.interpreter import Interpreter`.
  2. If your C++ code relied on the constexpr values of `tflite::Interpreter:kTensorsReservedCapacity` or `kTensorsCapacityHeadroom`, update it to handle these symbols as const references rather than compile‑time constants.

✨ New Features

  • The `tfl.Cast` operation now supports the `bfloat16` datatype in its runtime kernel.

🐛 Bug Fixes

  • Publishing of `libtensorflow` packages has been discontinued; the library can still be extracted from the TensorFlow PyPI package.

🔧 Affected Symbols

tflite::Interpreter:kTensorsReservedCapacitytflite::Interpreter:kTensorsCapacityHeadroomtf.lite.Interpreterai_edge_litert.interpreter.Interpretertfl.Castlibtensorflow

⚡ Deprecations

  • The `tf.lite.Interpreter` class is deprecated and scheduled for removal in TensorFlow 2.20; use `ai_edge_litert.interpreter.Interpreter` instead.