aychang95 / docker-deep-learning

GPU-accelerated Python Docker Images for Deep Learning

Image Types

The images are based on nvidia/cuda, and are intended to be drop-in replacements for the corresponding CUDA images in order to make it easy to add FastAI libraries while maintaining support for existing CUDA applications.

Images come in two types.

The aychang/deep-learning repo contains the following:

• runtime - extends the base image by adding a notebook server and example notebooks.
  • TIP: Use this image if you want to explore the image environment through notebooks and examples.
• devel - contains the full Python source tree including CUDA development libraries.
  • TIP: Use this image to develop Python and CUDA-based libraries from source.

Image Tag Naming Scheme

The tag naming scheme for th images incorporates key platform details into the tag as shown below:

cuda10.1-runtime-ubuntu18.04-py3.7
     ^    ^        ^         ^
     |    type     |         python version
     |             |
     cuda version  |
                   |
                   linux version

Prerequisites

• NVIDIA Pascal™ GPU architecture or better
• CUDA 10.1/10.2/11.0 with a compatible NVIDIA driver
• Ubuntu 16.04/18.04 or CentOS 7
• Docker CE v18+
• nvidia-docker v2+

Usage

Start Container and Notebook Server

Preferred - Docker CE v19+ and nvidia-container-toolkit

$ docker pull aychang/deep-learning:cuda10.2-runtime-ubuntu18.04-py3.7
$ docker run --gpus all --rm -it -p 8888:8888 \
         aychang/deep-learning:cuda10.2-runtime-ubuntu18.04-py3.7

Legacy - Docker CE v18 and nvidia-docker2

$ docker pull aychang/deep-learning:0.0.1-cuda10.2-runtime-ubuntu18.04-py3.7
$ docker run --runtime=nvidia --rm -it -p 8888:8888 \
         aychang/deep-learning:cuda10.2-runtime-ubuntu18.04-py3.7

Container Ports

The following ports are used by the runtime containers only (not base containers):

• 8888 - exposes a JupyterLab notebook server

Acknowledgement

Docker image setup based off of FastNN's docker component

Docker readme heavily influenced by RAPIDS docker readme here