PLEASE NOTE, THE DEEP LEARNING FRAMEWORK WAR IS OVER, THIS PROJECT IS NO LONGER BEING MAINTAINED.

Deepo is an open framework to assemble specialized docker images for deep learning research without pain. It provides a “lego set” of dozens of standard components for preparing deep learning tools and a framework for assembling them into custom docker images.

At the core of Deepo is a Dockerfile generator that

• allows you to customize your deep learning environment with Lego-like modules
  • define your environment in a single command line,
  • then deepo will generate Dockerfiles with best practices
  • and do all the configuration for you
• automatically resolves the dependencies for you
  • deepo knows which combos (CUDA/cuDNN/Python/PyTorch/Tensorflow, ..., tons of dependancies) are compatible
  • and will pick the right versions for you
  • and arrange sequence of installation procedures using topological sorting

We also prepare a series of pre-built docker images that

• allows you to instantly set up common deep learning research environment
• supports almost all commonly used deep learning frameworks
• supports GPU acceleration (CUDA and cuDNN included), also works in CPU-only mode
• works on Linux (CPU version/GPU version), Windows (CPU version) and OS X (CPU version)

• Quick Start
  • GPU Version
    • Installation
    • Usage
  • CPU Version
    • Installation
    • Usage
• Customization
  • Unhappy with all-in-one solution?
  • Jupyter support
  • Build your own customized image
• Comparison to Alternatives
• Tags
  • Available Tags
  • Deprecated Tags
• Citation
• Contributing
• Licensing

docker pull ufoym/deepo

For users in China who may suffer from slow speeds when pulling the image from the public Docker registry, you can pull deepo images from the China registry mirror by specifying the full path, including the registry, in your docker pull command, for example:

docker pull registry.docker-cn.com/ufoym/deepo

Now you can try this command:

docker run --gpus all --rm ufoym/deepo nvidia-smi

This should work and enables Deepo to use the GPU from inside a docker container. If this does not work, search the issues section on the nvidia-docker GitHub -- many solutions are already documented. To get an interactive shell to a container that will not be automatically deleted after you exit do

docker run --gpus all -it ufoym/deepo bash

If you want to share your data and configurations between the host (your machine or VM) and the container in which you are using Deepo, use the -v option, e.g.

docker run --gpus all -it -v /host/data:/data -v /host/config:/config ufoym/deepo bash

This will make /host/data from the host visible as /data in the container, and /host/config as /config. Such isolation reduces the chances of your containerized experiments overwriting or using wrong data.

Please note that some frameworks (e.g. PyTorch) use shared memory to share data between processes, so if multiprocessing is used the default shared memory segment size that container runs with is not enough, and you should increase shared memory size either with --ipc=host or --shm-size command line options to docker run.

docker run --gpus all -it --ipc=host ufoym/deepo bash

docker pull ufoym/deepo:cpu

Now you can try this command:

docker run -it ufoym/deepo:cpu bash

If you want to share your data and configurations between the host (your machine or VM) and the container in which you are using Deepo, use the -v option, e.g.

docker run -it -v /host/data:/data -v /host/config:/config ufoym/deepo:cpu bash

This will make /host/data from the host visible as /data in the container, and /host/config as /config. Such isolation reduces the chances of your containerized experiments overwriting or using wrong data.

Please note that some frameworks (e.g. PyTorch) use shared memory to share data between processes, so if multiprocessing is used the default shared memory segment size that container runs with is not enough, and you should increase shared memory size either with --ipc=host or --shm-size command line options to docker run.

docker run -it --ipc=host ufoym/deepo:cpu bash

You are now ready to begin your journey.

$ python

>>> import tensorflow
>>> import sonnet
>>> import torch
>>> import keras
>>> import mxnet
>>> import cntk
>>> import chainer
>>> import theano
>>> import lasagne
>>> import caffe
>>> import paddle

$ caffe --version

caffe version 1.0.0

$ darknet

usage: darknet <function>

Note that docker pull ufoym/deepo mentioned in Quick Start will give you a standard image containing all available deep learning frameworks. You can customize your own environment as well.

If you prefer a specific framework rather than an all-in-one image, just append a tag with the name of the framework. Take tensorflow for example:

docker pull ufoym/deepo:tensorflow

docker pull ufoym/deepo

docker run --gpus all -it -p 8888:8888 -v /home/u:/root --ipc=host ufoym/deepo jupyter lab --no-browser --ip=0.0.0.0 --allow-root --LabApp.allow_origin='*' --LabApp.root_dir='/root'

git clone https://github.com/ufoym/deepo.git
cd deepo/generator

For example, if you like pytorch and lasagne, then

python generate.py Dockerfile pytorch lasagne

or with CUDA 11.1 and CUDNN 8

python generate.py Dockerfile pytorch lasagne --cuda-ver 11.1 --cudnn-ver 8

This should generate a Dockerfile that contains everything for building pytorch and lasagne. Note that the generator can handle automatic dependency processing and topologically sort the lists. So you don't need to worry about missing dependencies and the list order.

You can also specify the version of Python:

python generate.py Dockerfile pytorch lasagne python==3.6

docker build -t my/deepo .

This may take several minutes as it compiles a few libraries from scratch.

@misc{ming2017deepo,
    author = {Ming Yang},
    title = {Deepo: set up deep learning environment in a single command line.},
    year = {2017},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/ufoym/deepo}}
}

We appreciate all contributions. If you are planning to contribute back bug-fixes, please do so without any further discussion. If you plan to contribute new features, utility functions or extensions, please first open an issue and discuss the feature with us.

Deepo is MIT licensed.