Vultr DocsLatest Docs

How to Build a PyTorch Container Image

Vultr Container RegistryContainer ImageDockerKubernetes
HM
Updated on April 24, 2024

Header Image

Introduction

PyTorch is a flexible machine learning library for building deep learning models that can perform a wide range of tasks such as image recognition and language processing.

This article explains how to build a PyTorch container image using the Vultr Container Registry.

Prerequisites

  • Deploy an instance using Vultr's GPU Marketplace App

  • Access the server using SSH.

  • Start the Docker service.

    console 
    $ sudo systemctl start docker

  • Add the non-root user to the Docker group. For example, linuxuser.

    console 
    $ sudo usermod -aG docker linuxuser

  • Switch to the user:

    console 
    $ su - linuxuser

Set Up the Server

  1. Create a new directory to store your PyTorch project files.

    console 
    $ mkdir llama-project

  2. Switch to the directory.

    console 
    $ cd llama-project

  3. Clone the Pytorch project repository using Git.

    console 
    $ git clone https://github.com/pytorch/pytorch

  4. List all files and verify that a new pytorch directory is available.

    console 
    $ ls

  5. Switch to the pytorch project directory.

    console 
    $ cd pytorch

  6. List all directory files and verify that the available Dockerfile resources.

    console 
    $ ls -a

    The PyTorch project directory includes the following Dockerfile resources:

    • Dockerfile: Contains the PyTorch build context for both CPU systems.
    • docker.makefile: Contains advanced GPU build context and image automation directives with support for custom dependency packages.

    Use the above resources in the next sections to build a CPU or GPU system container image.

Build a PyTorch Container Image for CPU Systems

Follow the steps below to build a new PyTorch container image for CPU systems using the Dockerfile resource on your server.

  1. Build a new PyTorch container image using the Dockerfile resource. Replace pytorch-image with your desired image name.

    console 
    $ docker build -f Dockerfile -t pytorch-image .

  2. View all Docker images and verify that your new Pytoch image is available.

    console 
    $ docker images

    Output:

    REPOSITORY     TAG                                                   IMAGE ID       CREATED       SIZE
pytorch-image  latest                                                d3674f1db4d5   2 hours ago   11.9GB

Build a PyTorch Container Image for GPU Systems

Follow the steps below to build a new PyTorch container image for GPU systems using docker.Makefile that contains the CUDA dependency build context compatible with NVIDIA GPU servers.

  1. Build a new PyTorch container image for GPU using the docker.Makefile.

    console 
    $ docker build -f Dockerfile -t pytorch-image .

    Wait at least 2 hours for the build process to complete. When successful, your output should look like the one below:

    #29 29224.1 copying torch/include/ATen/ops/_sparse_coo_tensor_with_dims_ops.h -> build/lib.linux-x86_64-cpython-310/torch/include/ATen/ops
#29 29224.1 copying torch/include/ATen/ops/absolute_compositeimplicitautograd_dispatch.h -> build/lib.linux-x86_64-cpython-310/torch/include/ATen/ops
#29 29224.1 copying torch/include/ATen/ops/gradient_ops.h -> build/lib.linux-x86_64-cpython-310/torch/include/ATen/ops
#29 29224.1 copying torch/include/ATen/ops/crow_indices_compositeexplicitautograd_dispatch.h -> build/lib.linux-x86_64-cpython-310/torch/include/ATen/ops
#29 29224.1 copying torch/include/ATen/ops/isinf.h -> build/lib.linux-x86_64-cpython-310/torch/include/ATen/ops
#29 29224.1 copying torch/include/ATen/ops/_fore
#29 29224.1 [output clipped, log limit 2MiB reached]
#29 29235.3 warning: no previously-included files matching '*.o' found anywhere in distribution
#29 29235.4 warning: no previously-included files matching '*.dylib' found anywhere in distribution
#29 29235.7 warning: no previously-included files matching '*.swp' found anywhere in distribution
#29 DONE 29239.6s

#30 [dev 1/1] COPY --from=build /opt/conda /opt/conda
#30 DONE 10.8s

#31 exporting to image
#31 exporting layers
#31 exporting layers 41.0s done
#31 writing image sha256:d3674f1db4d5e9585a5e5b64a0cdfcb0c7d312358e267000e23b6fa9d8e7e917 done
#31 naming to docker.io/root/pytorch:v1.8.0-rc1-38416-g4dc160864b3-cuda12.1-cudnn8-devel done
#31 DONE 41.0s

  2. View all Docker images on the server and verify that a new root/pytorch image is available.

    console 
    $ docker images

    Output:

    REPOSITORY     TAG                                                   IMAGE ID       CREATED       SIZE
root/pytorch   v1.8.0-rc1-38416-g4dc160864b3-cuda12.1-cudnn8-devel   d3674f1db4d5   3 hours ago   20.9GB

  3. Verify the PyTorch GPU image target CUDA version and verify that your system uses the driver version or higher.

    console 
    $ cat docker.Makefile | grep CUDA_VERSION

    Output:

    CUDA_VERSION_SHORT       ?= 12.1
CUDA_VERSION             ?= 12.1.1
BASE_DEVEL                = nvidia/cuda:$(CUDA_VERSION)-cudnn$(CUDNN_VERSION)-devel-ubuntu22.04

    Based on the above output, NVIDIA GPU Servers with CUDA version 12.1 or higher can run your PyTorch GPU container image.

Upload the PyTorch Container Image to the Vultr Container Registry

  1. Open the Vultr Customer Portal.

  2. Click Products and navigate to Container Registry on the main navigation bar.

    Manage a Vultr Container Registry

  3. Click your target Vultr Container Registry to open the management panel and view the registry access credentials.

  4. Copy the Registry URL value, Username, and API Key to use when accessing the registry.

    Open the Vultr Container Registry

  5. Switch to your server terminal session and log in to your Vultr Container Registry. Replace exampleregistry, exampleuser, registry-password with your actual registry details.

    console 
    $ docker login https://sjc.vultrcr.com/exampleregistry -u exampleuser -p registry-password

  6. Tag your PyTorch container image with your desired Vultr Container Registry tag. For example, tag your PyTorch GPU image ID with your registry repository tag such as sjc.vultrcr.com/exampleregistry/pytorch-gpu-image.

    console 
    $ docker tag d3674f1db4d5 sjc.vultrcr.com/exampleregistry/pytorch-gpu-image

    The above command tags the Docker image with container ID d3674f1db4d5 with your Vultr Container Registry tag.

  7. View all Docker images on the server and verify that the new tagged image is available.

    console 
    $ docker images

    Output:

    REPOSITORY           TAG                                                   IMAGE ID       CREATED       SIZE
root/pytorch         v1.8.0-rc1-38416-g4dc160864b3-cuda12.1-cudnn8-devel   d3674f1db4d5   4 hours ago   20.9GB
sjc.vultrcr.com/exampleregistry/pytorch-gpu-image   latest                 d3674f1db4d5   4 hours ago   20.9GB

  8. Push the tagged image to your Vultr Container Registry repository.

    console 
    $ docker push sjc.vultrcr.com/exampleregistry/pytorch-gpu-image

  9. Open your Vultr Container Registry management panel and click Repositories on the top navigation bar to verify that the new repository is available.

    View Vultr Container Registry Repositories