LiteRT
On-device runtime for running ML and generative AI models on edge hardware
What it is
LiteRT is Google's runtime for running machine learning models directly on phones, desktops, browsers, and embedded devices, continuing the TensorFlow Lite line. It handles converted PyTorch, TensorFlow, and JAX models, picks CPU, GPU, or NPU acceleration automatically, and works alongside companion projects for on-device LLMs and in-browser inference via WebGPU and WASM. Mobile and embedded developers use it to ship models inside their apps without calling a cloud service.