paddlepaddle/paddle-lite is a lightweight deep learning inference framework that runs PaddlePaddle models on mobile, embedded and edge devices. The latest release listed in the FACTS is v2.14-rc. However, the provided material does not say what changed in that release, so the changes cannot be identified. The repository description introduces it as an engine that supports a wide range of hardware platforms.
The project provides both the conversion and the execution paths needed to deploy a model on a device. The opt tool optimizes models through quantization, subgraph fusion and kernel selection. Quantization changes how numbers are represented in the model, and subgraph fusion is an optimization that groups several operations together. The PaddlePaddle model format is supported directly, and models from other frameworks can be converted to the PaddlePaddle format with X2Paddle. It exposes C++, Java and Python APIs as runtime interfaces and also includes usage examples for several platforms. The repository's language is C++ and its license is Apache-2.0.
The repository was created on 2017-09-20 and has 7289 stars. The pipeline's momentum score is 1.1/10, and no new-technology, technical or quality scores were provided. The supplied facts contain no performance figures and no comparison with other inference engines. The existence of the optimization tools is therefore confirmed, but the actual performance gains or the specific changes in v2.14-rc cannot be judged from this material alone.