LiteRT Support in Latest Tensorflow version 2.20.0

[Apoorv-1009]

Hey,

I had a quick question regarding LiteRT support. The release notes under the latest TensorFlow 2.20.0 release mention that:
tf.lite will be deprecated, in favor of the new repo https://github.com/google-ai-edge/LiteRT.

Does this mean that TensorFlow 2.20.0 already includes support for LiteRT internally, or is LiteRT integration planned for a future TensorFlow release?
If LiteRT is not yet supported under tensorflow/lite, does the LiteRT repo still include all the previous TensorFlow Lite headers? I’m updating some code that used tensorflow/lite, but I can’t find a few headers in the LiteRT repo, that existed in TensorFlow Lite under tensorflow/lite.

Thank you

[gaikwadrahul8]

Hi, @Apoorv-1009
I apologize for the delayed response, tf.lite is being replaced by LiteRT. The tf.lite module will be deprecated with development for on-device inference moving to a new, independent repository: LiteRT. The new APIs are available in Kotlin and C++. This code base will decouple from the TensorFlow repository and tf.lite will be removed from future TensorFlow Python packages, so we encourage migration of projects to LiteRT to receive the latest updates. please refer this blog

The TensorFlow 2.20.0 release acts as the definitive point of architectural decoupling. The release notes confirm that the legacy tf.lite module is officially deprecated and its duplicated source code has been removed from the TensorFlow repository.

LiteRT is now a separate, external dependency and this separation is mandatory moving forward. You must install LiteRT as an independent package. For Python this involves using the ai-edge-litert PIP package and for C/C++ obtain the LiteRT AAR file from MavenCentral and configure your build system (CMake/Gradle) to point to the extracted headers and native libraries within the AAR. Please refer this Tools for building with C and C++ section to build your apps.

Please refer this Migrate to LiteRT from TensorFlow Lite documentation. Please let us know if you have any further questions. Thank you for your cooperation and patience.

[github-actions]

Marking this issue as stale since it has been open for 7 days with no activity. This issue will be closed if no further activity occurs.

[AIWintermuteAI]

@gaikwadrahul8 , thanks for the response!
Complementary to the original question, could you clarify how can one match tensorflow version to lite-rt version for example in Python (but also other langauges).
Before it as rather simple, as tflite_runtime and tensorflow packages had the same versioning. E.g. tensorflow 2.11 would correspond to tflite_runtime 2.11.