TinyNeuralNetwork is an open-source deep learning model compression framework from Alibaba, published on GitHub under the MIT license. With about 880 stars and updates in 2026, it helps teams shrink neural networks so they can run efficiently on edge and IoT hardware.
The framework bundles the main techniques needed to take a trained PyTorch model and make it deployment-ready: neural architecture search, pruning, quantization, and model conversion. According to the project, it has been used in products such as Tmall Genie, Haier TVs, Youku video, and face-recognition check-in machines, equipping over 10 million IoT devices with AI capability. A graph tracer captures the connectivity of PyTorch operators, which automates pruning and quantization and can generate equivalent model description files.
The toolkit is organized around a few core modules. Its graph tracer captures how PyTorch operators connect, which is what allows pruning and quantization to be automated rather than hand-edited model by model. A dependency-resolving component handles the ripple effects when an operator is modified, preventing mismatches between connected subgraphs. The framework can also generate equivalent model description files from a PyTorch model, which helps when moving a compressed network toward deployment targets such as mobile and embedded runtimes.
TinyNeuralNetwork is free and open source under the MIT license. There is no paid tier, hosted service, or usage fee; you download the code and run it in your own environment.
It suits ML engineers and embedded developers who need to deploy deep learning models on resource-constrained devices, and teams looking for a single toolkit that covers pruning, quantization, and conversion instead of stitching together separate utilities.