TensorRT

Why we included this project

TensorRT is the engine most production teams reach for when a trained model has to run fast on NVIDIA GPUs, and its job is narrow: take the model, compile it into an optimized engine, and let the GPU do the rest. The open-source portion of this repo matters if you deploy on your own hardware or in containers, because it ships the plugins, ONNX parser, and sample applications you need to build and benchmark custom engines instead of relying only on the closed distribution. Models can come in through ONNX, Torch-TensorRT, HuggingFace/Optimum, or the network definition API, and the supported list covers LLMs, vision, audio, diffusion, and multimodal work, so it is a serious option when latency is tight and you control the serving stack.