Shortcuts

使用 Docker 镜像

本文简述如何使用Docker安装mmdeploy

获取镜像

为了方便用户,mmdeploy在Docker Hub上提供了多个版本的镜像,例如对于mmdeploy==1.2.0, 其镜像标签为openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy1.2.0,而最新的镜像标签为openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy。 镜像相关规格信息如下表所示:

Item Version
OS Ubuntu20.04
CUDA 11.8
CUDNN 8.9
Python 3.8.10
Torch 2.0.0
TorchVision 0.15.0
TorchScript 2.0.0
TensorRT 8.6.1.6
ONNXRuntime 1.15.1
OpenVINO 2022.3.0
ncnn 20230816
openppl 0.8.1

用户可选择一个镜像并运行docker pull拉取镜像到本地:

export TAG=openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy
docker pull $TAG

构建镜像(可选)

如果已提供的镜像无法满足要求,用户可修改docker/Release/Dockerfile并在本地构建镜像。其中,构建参数MMDEPLOY_VERSION可以是mmdeploy项目的一个标签或者分支。

export MMDEPLOY_VERSION=main
export TAG=mmdeploy-${MMDEPLOY_VERSION}
docker build docker/Release/ -t ${TAG} --build-arg MMDEPLOY_VERSION=${MMDEPLOY_VERSION}

运行 docker 容器

当拉取或构建 docker 镜像后,用户可使用 docker run 启动 docker 服务:

export TAG=openmmlab/mmdeploy:ubuntu20.04-cuda11.8-mmdeploy
docker run --gpus=all -it --rm $TAG

常见问答

  1. CUDA error: the provided PTX was compiled with an unsupported toolchain:

    这里所说,更新 GPU 的驱动到您的GPU能使用的最新版本。

  2. docker: Error response from daemon: could not select device driver “” with capabilities: [gpu].

    # Add the package repositories
    distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
    curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
    curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list
    
    sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
    sudo systemctl restart docker
    
Read the Docs v: latest
Versions
latest
stable
v1.3.0
v1.2.0
v1.1.0
v1.0.0
0.x
v0.14.0
Downloads
pdf
html
epub
On Read the Docs
Project Home
Builds

Free document hosting provided by Read the Docs.