Deep Learning Reference Stack with Pytorch and Intel® oneAPI Deep Neural Network Library (oneDNN)

Building Locally

We have created a set of Dockerfiles that allow you to build DLRS Pytorch with different configurations. You will be able to pick the OS (Ubuntu or Centos), then pick the flavour, either core, which is a leaner image with just the framework, and full which includes the framework and extra features. Please refer to the table below to see all features and configurations.

Framework Features
Core X
Full X X

Framework:

  • PyTorch* 1.8

  • Torchvision

Features:

  • Transformers

  • Flair*

  • Horovod

  • Seldon core

  • Pytorch lightning

NOTE: We recommend you enable Docker Buildkit to have concurrent dependency resolution and automatic garbage collection. Docker Buildkit has been integrated in Docker since 18.06, if you have an older version, please ignore this note.

Enable Docker Buildkit (see note above)

export DOCKER_BUILDKIT=1

Building Ubuntu based DLRS

Core:

docker build -t dlrs-pytorch-ubuntu:v0.9.1 -f Dockerfile.ubuntu --target ubuntu-core .

Full:

docker build -t dlrs-pytorch-ubuntu:v0.9.1 -f Dockerfile.ubuntu --target ubuntu-full .

Building Centos based DLRS

Core:

docker build -t dlrs-pytorch-centos:v0.9.1 -f Dockerfile.centos --target core-core .

Full:

docker build -t dlrs-pytorch-centos:v0.9.1 -f Dockerfile.centos --target centos-full .