Google launches LiteRT Tensor API for on-device machine learning
Google launched LiteRT Tensor API, a lightweight C++ API for building, customizing, and optimizing on-device machine-learning models. [[fact:fact-xzicngjd5]] Developers can build computation graphs operator by operator in C++ and run the same graph through LiteRT on CPU, GPU, and NPU hardware without rewriting it for each target. [[fact:fact-d6lkjryym]] The API supports loading weights directly from checkpoints, including Hugging Face .safetensors files. [[fact:fact-739unxanr]] Google says a WebGPU test at 512×512 resolution raised frame rate from below 10fps to above 30fps, with the iterative optimization completed by a coding agent. [[fact:fact-rvkjgv5m7]] Google also outlined planned improvements, including native ML Drift backend support, more NPU-friendly operations, automatic whole-graph quantization, and expanded dynamic-shape support. [[fact:fact-ifjzddg2j]]