This repository regroups multiple Singularity containers recipes for building and compiling high performance Python libraries used in various deep learning usage such as OpenCV, PyTorch and TensorFlow. These recipes are currently deployed in École de technologie supérieure in Software Engineering computer laboratories and research servers.

This repository aims at keeping various versions of these libraries compiled in Singularity containers with all necessary libraries for a common usage of them.

• Authors
• References
• Project architecture
  • Why Singularity ?
  • Folder structure
  • Continuous Integration
• Contributing
• Branch naming
• Commits syntax
• Acknowledgments

• Patrice Dion
• Pierre-Luc Delisle - pldelisle

https://docs.nvidia.com/cuda/archive/9.2/cuda-installation-guide-linux/index.html
https://docs.nvidia.com/cuda/archive/10.0/cuda-installation-guide-linux/index.html
https://docs.nvidia.com/cuda/archive/10.1/cuda-installation-guide-linux/index.html
https://devblogs.nvidia.com/gpu-containers-runtime/
https://www.nvidia.com/object/unix.html
https://developer.nvidia.com/nccl
https://developer.nvidia.com/cudnn

sylabs/singularity#1292
https://singularity.lbl.gov/docs-recipes#best-practices-for-build-recipes

See respective README in containers' folders.

We decided to choose Singularity over Docker because Singularity allows to run containers in user space, without the need to increase user privilege (e.g. adding user to a certain group with more privileges). This allows running containers on any Linux machine without any security risk.

├── .travis                 - Contains Travis-CI files to automatically build containers on Travis' infrastructure.
|
├── icons                   - Contains project's artwork.
|
├── opencv                  - OpenCV containers.
│   ├── opencv-3.4.6-cuda-9.2-Ubuntu-18.04-amd64.def.def  - CUDA enabled OpenCV container.
│   └── opencv-3.4.6-Ubuntu-18.04-amd64.def.def           - x86-64 optimized OpenCV container.
|
├── pytorch                 - PyTorch containers.
|   └── .def
|
├── tensorflow              - Tensorflow containers.
|   └── .def

As of now, Travis-CI automatic building works. Unfortunately, the free package doesn't allow more than 50 minutes for a build. After that time, builds will fail. This repository is ready for a paid Travis-CI package where the option for unlimited minutes for building are included.

Bamboo server is configured for building containers on a private server : pldelisle.no-ip.info:8086. This Bamboo server runs in a Docker container where Golang and Singularity are both installed.

To run this container :

docker run -v /home/<USER>/Documents/bamboo:/var/atlassian/application-data/bamboo --name="bamboo-server" --privileged -d -p 54663:54663 -p 8086:8085 --restart always pldelisle/bamboo-server-singularity

Image is stored on Dockerhub

If you find a bug or have an idea for an improvement, please follow this procedure:

• Create a branch by feature and/or bug fix
• Get the code
• Commit and push
• Create a pull request

+ Added [Short Description] [Issue Number]

- Deleted [Short Description] [Issue Number]

* Changed [Short Description] [Issue Number]

Y Merged [Short Description]

Thanks to École de technologie supérieure and Patrice Dion for providing recipes for these containers.

Icons made by Freepik from www.flaticon.com is licensed by CC 3.0 BY