pytorch_docker
Build pytorch docker images. It's works on linux.
Build images
-
cd pytorch_docker
-
sh init_torch.sh [
pytorch version
] [pytorch git commit if needed
] [torchvision git commit if needed
] -
make
Image Labels
docker inspect -f '{{json .Config.Labels}}' switchwang/pytorch:v1.7.0-cuda11.1-cudnn8-ubuntu18.04-devel
{
"com.nvidia.cudnn.version": "8.0.4.30",
"com.nvidia.volumes.needed": "nvidia_driver",
"maintainer": "NVIDIA CORPORATION <cudatools@nvidia.com>"
}
Label Name | Description |
---|---|
maintainer |
Maintainer of the image |
Load images
-
pytorch: v1.6.0, torchvision: v0.7.0, cuda: 10.2, cudnn: 7, devel images
docker pull switchwang/pytorch:v1.6.0-cuda10.2-cudnn7-ubuntu18.04-devel
-
pytorch: v1.6.0, torchvision: v0.7.0, cuda: 10.2, cudnn: 7, runtime images
docker pull switchwang/pytorch:v1.6.0-cuda10.2-cudnn7-ubuntu18.04-runtime