torch2trt provides a Python API to convert PyTorch modules to TensorRT engines, enabling significant inference speedups on NVIDIA hardware. Users wrap a PyTorch model and example input data, and the converter produces a TRTModule that executes the optimized TensorRT inference graph. The tool supports custom layer converters via decorators, facilitating extension to unsupported operations. It was originally developed for the JetBot project to accelerate computer vision inference on Jetson devices.