Installation
Choose one path:
| Path | Current boundary |
|---|---|
| Release wheel | Linux aarch64, Python 3.10 or 3.12, glibc 2.39 or newer, and official TensorRT 11.1.0.106. |
| Source build | Linux x86_64 or aarch64 with Docker and NVIDIA Container Toolkit. |
Release wheels currently support aarch64 only. x86_64 wheels are planned; use Build from Source on x86_64 today.
Install a release wheel
This first-time path uses Python 3.12. Download the aarch64 +trt111 py312
.whl asset from the latest GitHub release.
If no matching wheel is published, use the source path instead.
python3.12 -m venv .venv-trtmc
. .venv-trtmc/bin/activate
WHEEL=/path/to/downloaded-wheel.whl
python -m pip install "$WHEEL"
Python 3.10 users can instead use the +trt111 py310 wheel and corresponding
python3.10 commands. Keep this virtual environment active and continue to
Quick Start, where the CLI is checked once.
Build and run from source
Continue in the source container prepared by
Build from Source. That page adds the source-built CLI to
PATH before sending you to Quick Start.
Do not mix a wheel, backend DSO, bundle, or TensorRT library from another architecture or TensorRT cohort.