AI智能摘要
想在JetsonOrin平台上用上最新Python3.12，却发现官方TensorRTwhl包还没适配？别担心，这篇手把手教程将带你从零开始，在Orin设备上成功编译出专属的TensorRTPythonwhl包。我们不仅会详细列出所需的环境参数和构建步骤，更会帮你避开依赖路径、架构兼容性等常见“坑”，确保你能顺利安装并完成核心功能验证。
— AI 生成的文章内容摘要
基础信息
| 类别 | 关键参数 | 记录值 | 备注 |
| 基础工具 | CMake / GCC | 4.2.1 / 11.4.0 | 建议 CMake > 3.20 |
| Python | Version / Path | 3.12.12 / .venv | 确保在虚拟环境下编译 |
| GPU 环境 | CUDA / cuDNN | 12.6 / 9.3.0 | |
| GPU 架构 | CUDA_ARCH | sm_87 (Orin) | 决定硬件兼容性 |
| TensorRT | Version | 10.3.0.26 | 构建产物 |
构建步骤
拉取项目代码
mkdir -p ~/makenv && cd ~/makenv
git clone --recursive --branch release/10.3 https://github.com/NVIDIA/TensorRT.git
cd ~/makenv/TensorRT/
创建虚拟环境
uv venv .venv --python 3.12
source .venv/bin/activate
uv pip install setuptools
创建 pybind
export PYTHON_INCLUDE_DIR=$(python3 -c "from sysconfig import get_paths; print(get_paths()['include'])")
export EXT_PATH=~/makenv/external
git clone https://github.com/pybind/pybind11.git $EXT_PATH/pybind11
mkdir -p $EXT_PATH/python3.12/include
ln -s $PYTHON_INCLUDE_DIR/* $EXT_PATH/python3.12/include/
构建
export TENSORRT_LIBPATH=/usr/lib/aarch64-linux-gnu
export TRT_OSSPATH=~/makenv/TensorRT
cd $TRT_OSSPATH/python
TENSORRT_MODULE=tensorrt PYTHON_MAJOR_VERSION=3 PYTHON_MINOR_VERSION=12 TARGET_ARCHITECTURE=aarch64 ./build.sh
whl输出目录
~/makenv/TensorRT/python/build/bindings_wheel/dist/tensorrt-10.3.0-cp312-none-linux_aarch64.whl
安装
uv pip install ~/makenv/TensorRT/python/build/bindings_wheel/dist/tensorrt-10.3.0-cp312-none-linux_aarch64.whl测试验证
最小可用性测试
python -c "
import tensorrt as trt
print('TensorRT version:', trt.__version__)
logger = trt.Logger(trt.Logger.INFO)
builder = trt.Builder(logger)
network = builder.create_network(
1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
)
config = builder.create_builder_config()
config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, 1 << 20)
print('Builder OK')
print('Network OK')
print('Config OK')
"
Comments NOTHING