ncnn

Why we included this project

ncnn is a C++ inference engine built for running neural networks where resources are tight, like phones and embedded boards. It has no third-party runtime dependencies, so it fits into Android, iOS, Linux, and WebAssembly builds without pulling in a heavy framework. The pnnx companion tool converts PyTorch and ONNX models into ncnn's compact param/bin format, keeping the trip from training to deployment short. CPU and Vulkan GPU backends let you trade portability against acceleration as your target demands, and the framework has been running inside Tencent apps such as WeChat and QQ for years, so the code has seen real production traffic. If you are new to mobile inference, the build guides and Python bindings lower the entry bar considerably.