LiteRT.js, Google's high performance Web AI Inference

Google has launched LiteRT.js, a JavaScript runtime that runs AI inference directly in the browser.

It supports existing .tflite models and uses WebAssembly with CPU, GPU, and emerging NPU acceleration.

The release includes an npm package, demos, PyTorch conversion tools, and quantization options.

Google says it can be up to 3x faster than other web AI runtimes in some benchmarks.