dl-env

Docker files for my deep learning env, with CUDA-supported-PyTorch, fastai, Tensorflow, Keras etc.

开箱即用的深度学习环境(含有 jupyter notebook, 支持CUDA的PyTorch, fastai, Tensorflow, Keras, etc. )

Features

• common packages included(via base Docker image):
  • Python3.6/Anaconda
  • Tensorflow latest
• extra packages
  • PyTorch latest
  • fastai
  • Keras
  • numpy/pandas
  • ...
• time zone default to Asia/Shanghai
• a default password
• GUI-based nbextension_configurator
• auto-build at https://hub.docker.com/r/koyo922/dl-env/
• NB: add something like -e "TZ=Asia/Shanghai" and rebuild the image, if you need to use another time zone setting

Usage

mkdir -p ~/workspace/home/me && cd ~/workspace  # NB: IT IS NECESSARY

docker run --gpus all -it -p 8888:8888 -p 8000-8100:8000-8100 \
	-u $(id -u ${USER}):$(id -g ${USER}) \
	-v $PWD:/workspace -v $PWD/home/me:/home/me -w /workspace \
	--name dl \
	koyo922/dl-env:gpu

# for older version `docker`, use `nvidia-docker run` instead of `docker run --gpus all`
# for CPU machine, please modify the above `nvidia-docker` into `docker`, and `koyo922/dl-env:gpu` into `koyo922/dl-env:cpu`

Note:

• How to access the jupyter server
  1. read the docker container output
  2. look for something like http://(4cd7fac856a5 or 127.0.0.1):8888/?token=4c5bf6171097f89ced23ed52a84aced0f6f7e3a24b777554
  3. copy it to you web browser address bar
  4. modify the hostname part (4cd7fac856a5 or 127.0.0.1) with its Internet IP
  5. hit enter
• To install new packages in notebooks, you need pip install --user ...
• nvidia-driver 参考官网正常安装即可。阿里云上除vgn5i以外的GPU机器都支持自动安装驱动和CUDA，启动实例时勾上就好
• nvidia-docker 请参考这里安装 https://zhuanlan.zhihu.com/p/76464450

Reference

• https://zhuanlan.zhihu.com/p/59617018
• https://github.com/jxcodetw/docker-jupyter-pytorch/blob/master/Dockerfile
• https://github.com/jupyter/docker-stacks/tree/master/tensorflow-notebook