TensorRT
NVIDIATensorRT is NVIDIA's SDK for compiling trained networks into optimized inference engines for NVIDIA GPUs, applying layer fusion and explicit quantization. This repository holds the open-source components, the ONNX parser, the plugin library, the Python bindings and samples. The optimizing builder itself ships as a prebuilt binary release that the repository links against.
Openness
2 high confidence- license
- Apache-2.0(OSI)
- source
- partial
TensorRT's Apache-2.0 license covers only the parsers, plugins, bindings and samples in this repository; the optimizing builder that is the actual product ships as a prebuilt binary the build links against rather than source in the tree. That makes the source only partial, regardless of how permissive the covering license is.
- https://raw.githubusercontent.com/NVIDIA/TensorRT/main/LICENSE recorded 2026-09-30
LICENSE file is the verbatim Apache-2.0 text, opening with "Apache License" and "Version 2.0, January 2004" and "TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION".
- https://raw.githubusercontent.com/NVIDIA/TensorRT/main/README.md recorded 2026-09-30
README build step tells the reader to "download and extract the TensorRT GA build from [NVIDIA Developer Zone](https://developer.nvidia.com)" as a tar.zst release, then "export TRT_LIBPATH=`pwd`/TensorRT-11.3.0.99/lib", so the optimizing builder is a prebuilt binary. Searched the body for license key, proprietary and closed source: none; the tarballs are named "TensorRT-Enterprise-11.3.0.99" and enterprise support is offered via "NVIDIA AI Enterprise".
Adoption
3 high confidenceAdoption is measured from PyPI downloads of the `tensorrt` package, NVIDIA's Python bindings for the SDK.
- https://pypistats.org/api/packages/tensorrt/recent recorded 2026-08-18
last_month downloads = 189,754 for tensorrt
Capability
4 medium confidenceTensorRT applies graph-level fusion and explicit quantization, with ONNX ingest and a plugin interface for custom layers, compiling to NVIDIA GPU engines. That toolchain is narrowed to one vendor's hardware, one step below Apache TVM's multi-vendor reach.
- https://docs.nvidia.com/deeplearning/tensorrt/latest/inference-library/quantized-types-explicit-quantization.html recorded 2026-09-30
Explicit quantization page says "When TensorRT performs these fusions, it replaces the quantizable layers with quantized layers".
- https://raw.githubusercontent.com/NVIDIA/TensorRT/main/documents/import_workflows.md recorded 2026-09-30
Import guide says "The ONNX path is the most portable way to bring models from PyTorch, TensorFlow, JAX, or any framework with an ONNX exporter." and "A supported NVIDIA GPU (see the [Support Matrix]".
- https://raw.githubusercontent.com/NVIDIA/TensorRT/main/documents/supported_models.md recorded 2026-09-30
Supported models doc says "TensorRT is a general-purpose neural-network graph execution engine, not a model zoo." and that the "Custom Plugin" section "covers the escape hatch for ops TensorRT does not yet implement natively."
- https://raw.githubusercontent.com/NVIDIA/TensorRT/main/README.md recorded 2026-09-30
README says "Implicit quantization and related APIs have been removed, replaced by [Explicit Quantization]" and "IPluginV2 and related APIs have been removed, replaced by [IPluginV3]" in TensorRT 11.X, and the OSS components include "the sources for TensorRT plugins and ONNX parser".
Verified 2026-09-24