LiteRT: Google’s Transformation of TensorFlow Lite Allows Versatile Model Execution Across Multiple Platforms

Google has announced a name change for TensorFlow Lite (TF Lite), the runtime for running AI models on portable devices, to a new name LiteRT (RT = Runtime). The reason for the change is that TensorFlow Lite has evolved far beyond TensorFlow itself, now supporting models developed with other frameworks such as PyTorch, JAX, and Keras. It supports various types of models including ML, LLM, and diffusion, making the inclusion of the word TensorFlow in the name confusing as to which models it can be used with.

The new name LiteRT carries on the legacy of the original Lite name and aims to show that it is a lightweight runtime designed to run on-device AI tasks on a variety of portable devices. In addition to the project name and package name, LiteRT will continue to be developed in the same direction, moving from being a sub-project under TensorFlow to being under Google AI Edge instead.

TLDR: Google has renamed TensorFlow Lite to LiteRT for running AI models on portable devices due to its compatibility with various frameworks and model types, emphasizing its lightweight nature and future development under Google AI Edge.