Nvidia-AGX-spark部署手册——课堂质量诊断(jetpack:r36)

Kevin Wang7272026-08-01 16:49

Docker 镜像封装 Python 依赖和服务启动。重点是：AGX 不能随便用普通 nvidia/cuda 镜像，要匹配 L4T/JetPack 版本。

对于 NVIDIA AGX/Jetson 盒子这类设备，推荐 Dockerfile：

复制代码

FROM nvcr.io/nvidia/l4t-jetpack:r36.4.0

ENV DEBIAN_FRONTEND=noninteractive
ENV TZ=Asia/Shanghai
ENV LANG=en_US.UTF-8
ENV LANGUAGE=en_US:en
ENV LC_ALL=en_US.UTF-8

WORKDIR /opt/jyd01/wangruihua/jyd_speaker_agx

COPY . /opt/jyd01/wangruihua/jyd_speaker_agx

RUN apt-get update && apt-get install -y \
    python3-pip \
    python3-dev \
    python3-venv \
    wget \
    curl \
    git \
    vim \
    nano \
    sudo \
    lsof \
    ffmpeg \
    net-tools \
    iputils-ping \
    less \
    tzdata \
    locales \
    ncurses-bin \
    libgl1 \
    libglib2.0-0 \
    libopenblas-dev \
    libcairo2-dev \
    pkg-config \
    python3-dev \
    && ln -fs /usr/share/zoneinfo/Asia/Shanghai /etc/localtime \
    && echo Asia/Shanghai > /etc/timezone \
    && dpkg-reconfigure -f noninteractive tzdata \
    && locale-gen en_US.UTF-8 \
    && rm -rf /var/lib/apt/lists/*

RUN python3 -m pip install -U pip setuptools wheel packaging \
    -i https://pypi.tuna.tsinghua.edu.cn/simple

RUN python3 -m pip install -r requirements_agx.txt \
    -i https://pypi.tuna.tsinghua.edu.cn/simple

RUN chmod +x /opt/jyd01/wangruihua/jyd_speaker_agx/entrypoint.sh || true

EXPOSE 8003 8006

CMD ["bash", "/opt/jyd01/wangruihua/jyd_speaker_agx/entrypoint.sh"]

构建镜像：

复制代码

sudo docker build \
  --network=host \
  --progress=plain \
  -t jyd_speaker_agx:v1.1 .

创建容器：

复制代码

docker run -d \
  --name jyd_speaker_agx_container \
  --runtime nvidia \
  --restart always \
  --network host \
  -v /opt/jyd01/wangruihua/jyd_speaker_agx:/opt/jyd01/wangruihua/jyd_speaker_agx \
  jyd_speaker_agx:v1.0

5. 查看状态

复制代码

docker ps
docker logs -f jyd_speaker_agx_container