/ml-docker

Container for Deep Learning with built-in Jupyter/Tensorboard and latest DL Frameworks

Primary LanguageDockerfile

NVIDIA DL Docker

Image for Deep Learning with built-in Jupyter/Tensorboard and latest DL Frameworks

Requirements

  • Ubuntu 18.04

Environment

  • CUDA Toolkit 10.0
  • CuDNN 7.x
  • NCCL 2
  • Docker
  • NVIDIA-Docker 2

Packages

  • Python 3.7
  • Tensorflow 1.14.0
  • PyTorch 1.1.0
  • Keras
  • Tensorboard
  • Jupyter
  • ...other useful packages

Quickstart

  • Clone this repository
    git clone https://github.com/lucidyan/ml-docker

  • Install CUDA-10


- Old instruction: https://gist.github.com/bogdan-kulynych/f64eb148eeef9696c70d485a76e42c3a - New instruction: https://gist.github.com/Mahedi-61/2a2f1579d4271717d421065168ce6a73 - Best guide: https://www.tensorflow.org/install/gpu

  • Install NVIDIA-Docker
    cd ml-docker; sudo chmod a+x nvidia_docker_install.sh; sudo ./nvidia_docker_install.sh

  • Reboot system after Docker installation (necessary for running Docker without sudo rights)

  • Build the image
    docker build --build-arg USER_ID=$(id -u) --build-arg GROUP_ID=$(id -g) --tag "lucidyan/ml-docker:20.02.1" ." .

  • Run it with command
    python3 run_docker_jupyter.py -pj 8888 -pt 6006
    where 8888 and 6006 your local unoccupied ports for Jupyter and Tensorboard respectively