Announcing TensorFlow Lite

In November 2017, the TensorFlow team announced the developer preview of TensorFlow Lite, a streamlined version of TensorFlow designed for mobile and embedded devices. This new tool allows for low-latency inference of machine learning models on various platforms, including Android and iOS, and is optimized for fast initialization and small binary size. TensorFlow Lite supports hardware acceleration through the Android Neural Networks API, enabling efficient processing of machine learning workloads on custom hardware, while also providing a fallback to optimized CPU execution. The architecture includes components such as the TensorFlow Lite Converter and Interpreter, which reduce the model size significantly compared to TensorFlow Mobile. It initially supports models like MobileNet and Inception v3, with plans to expand functionality based on user needs. TensorFlow Lite is considered the evolution of TensorFlow Mobile, and though still in development, it aims to simplify the developer experience and broaden model deployment across devices.