Skip to content

Latest commit

 

History

History
47 lines (35 loc) · 2.15 KB

install_with_docker.md

File metadata and controls

47 lines (35 loc) · 2.15 KB
Raw

Install BladeDISC With Docker

Docker is a light container system, it helps users to package software and isolate BladeDISC runtime environment from the rest of the system. BladeDISC CI system released BladeDISC with different tag on Docker Hub repository.

Nvidia Container Toolkit is an easy way to use GPU on Linux, please make sure you have installed it on your host.

Download a BladeDISC Docker Image

BladeDISC released TensorFlow and PyTorch frontend packages in separate Docker images on bladedisc/bladedisc. The released Image tag is as the following table:

Docker tag Description
latest-runtime-torch1.7.1 latest release of BladeDISC, includes PyTorch 1.7.1 and CUDA 11.0
latest-runtime-tensorflow1.15 latest release of BladeDISC, includes TensorFlow 1.15 and CUDA 11.0
latest-runtime-tensorflow2.4 latest release of BladeDISC, includes TensorFlow 1.15 and CUDA 11.0
latest-devel-cuda10.0 latest build of development environment, includes CUDA 11.0 and required development toolkit
latest-devel-cuda11.0 latest build of development environment, includes CUDA 11.0 and required development toolkit

Note: Users located in China can use registry.cn-shanghai.aliyuncs.com/bladedisc/bladedisc to get higher download speed.

Start a Docker Container

To launch a BladeDISC Docker container with GPU support, you can use the following command:

docker run --rm -it --gpus all -v [host-src/container-desc] bladedisc/bladedisc:[tag] [command]
  • --rm automatically remove it after the container stops.
  • -it runs the container with interactive mode.
  • -v [host-src/container-dest] mount a volume from host to container.

An example to execute the entry.py PyTorch script with BladeDISC Docker:

nvidia-docker run --rm -it -v $PWD:/work bladedisc/bladedisc:latest-runtime-torch1.7.1  python /work/entry.py