SettledOpen source
Google releases LiteRT Tensor API for building on-device models op by op in C++
Google released the LiteRT Tensor API, a C++ layer for defining a model graph op by op and loading weights from .safetensors. Graphs run on CPU, GPU or NPU through LiteRT, or on CPU via XNNPACK. Google says a SAM 2 build ran about 1.5x faster than Apple MLX on a MacBook Pro (fp16). Python authoring is listed as future work.
- HEAT
- OUTLETS
- 0
- FIRST SEEN
- LAST SIGNAL
Primary source
Google for Developersdevelopers.googleblog.com/native-model-authoring-for-edge-with-litert-tensor-api/Coverage · 1 article, oldest first
Monday, October 19, 2026
Google for DevelopersPrimary
Native Model Authoring for Edge with LiteRT Tensor API