Some docker containers for experimenting with deep learning.

Nvidia provides three types of images for each os / cuda version.

• base: Includes the CUDA runtime (cudart)
• runtime: Builds on the base and includes the CUDA math libraries, and NCCL. A runtime image that also includes cuDNN is available.
• devel: Builds on the runtime and includes headers, development tools for building CUDA images. These images are particularly useful for multi-stage builds.

See: https://hub.docker.com/r/nvidia/cuda

The idea of these images is to rebase the jupyter stack docker iamges: See: https://github.com/jupyter/docker-stacks with a Nvidia cuda base for the purposes of experimenting with deeplearning and other GPU based ML things.

• deeplearn base:

  • Mirrors jupyter-stack base-notebook with cuda base image

• deeplearn minimal:

  • Mirrors jupyter-stack minimal-notebook with cuda base image

• deeplearn_opencv:

  • Opencv can be hard to install and configure properly especially to enable CUDA support as well.
  • Builds a base image

# build all the images listed in ALL_IMAGES
make build-all

# build individual images:
# check makefile for individual build strings
make build-xxx  

This is the jupyter docker base build but on top of nvidia docker instead to enable GPU functionality. Also with added nbdev and debugger components

see: https://github.com/jupyter/docker-stacks/tree/master/base-notebook https://hub.docker.com/repository/docker/datadrone/deeplearn_base

This is the jupyter docker minimal build on top of the base

see: https://github.com/jupyter/docker-stacks/tree/master/minimal-notebook https://hub.docker.com/repository/docker/datadrone/deeplearn_minimal

This is Minimal + OpenCV https://hub.docker.com/repository/docker/datadrone/deeplearn_opencv

Current compiled with opencv 4.5.1 with Compute Model 7.5

This is docker OpenCV plus pytorch 1.7 Latest is currently on CUDA 11.0 with Torch 1.7.1

This is docker OpenCV plus TensorFlow 2.x Latest is currently on CUDA 11.2 with TF 2.4.1

This is docker OpenCV plus MXNet 1.9.0 nightly There were issues with compiling and there is no 1.8.x stable release on cuda 11.x at the moment.

This adds in the Nvidia RAPIDS.ai libraries 0.18 with dask dashboard for jupyter 3.0

Make the base image an arg? Make build vs pip/conda installed versions of Pytorch/MXNet/TF? Make a parameter in the Makefile for CUDA versions?