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

This post introduces LiteRT.js, a JavaScript binding for LiteRT that enables high-performance AI inference directly within web browsers. It highlights the ability to run ML models client-side using WebAssembly, leveraging hardware acceleration (XNNPACK for CPU, ML Drift for GPU, and WebNN for NPUs). The post details features like PyTorch conversion, tailored quantization, and provides performance benchmarks demonstrating significant speedups over existing web AI solutions. It also showcases real-world integrations with YOLO, depth estimation, and image upscaling models.