Google introduces LiteRT.js for high-speed local web AI inference

Google has released LiteRT.js, a JavaScript binding that brings its high-performance LiteRT AI inference engine directly to the web browser using WebAssembly.

The new library supports local execution of .tflite models with hardware acceleration for CPU, GPU, and experimental NPU backends.

LiteRT.js also allows PyTorch models to run in the browser with single-step conversion and tailored quantization.

Benchmarks show speedups of up to 60x in real-time applications using GPU or NPU acceleration.