Google releases LiteRT Tensor API for native on-device model authoring
Original titleNative Model Authoring for Edge with LiteRT Tensor API
AISummary
Google announces the LiteRT Tensor API, a C++ authoring layer for building on-device model graphs op by op. The same graph can run through LiteRT on CPU, GPU, or NPU, or directly on CPU through XNNPACK. In its SAM 2 demo, Google says the model runs about 1.5x faster than Apple MLX on a MacBook Pro for the image encoder plus mask decoder, and the code is open source.
AIWhy it matters
The post explains when to choose the Tensor API over model conversion, with a SAM 2 example and benchmarks showing how C++ graphs run on Metal and OpenCL.
Source: Google Developers Blog · developers.googleblog.comPublished · added here