Google's new LiteRT.js speeds up browser AI by 3 times

Google just launched LiteRT.js, a new library to make AI and machine learning run faster in web browsers. It uses WebAssembly plus WebGPU and WebNN for hardware acceleration, replacing the old JavaScript core in TensorFlow.js.

WebAssembly is a binary format that lets web pages run high-performance computing nearly as fast as native code. TensorFlow.js is Google's library for building, training, and deploying ML models in JavaScript.

On a 2024 MacBook Pro with M4 chip, tests show LiteRT.js is 3 times faster than the old approach. Google notes actual speed may vary on older hardware or different browser engines.