savvihub/kernels repository overview

⁠images

SavviHub images for experiment and service

⁠Image Description

⁠Data science packages (DS)

All SavviHub Images have the most used data science python package installed. The full list of DS packages are as follows.

numpy, scipy, pandas, matplotlib, scikit-learn, opencv-python, seaborn, plotly, tqdm

⁠Image tag

All images are based on Ubuntu 18.04.

• py3x tags indicate python version 3.x
• -cuda1X.x tags are based on Nvidia CUDA version 1X.x. You need nvidia-docker to run them.
• full-cpu or full-gpu tags have three compatible machine learning frameworks installed. (PyTorch, TensorFlow, MXNet)
• jupyter tags include JupyterLab and some extensions. They start a Jupyter notebook server on boot.

⁠Disclaimer

One of the JupyterLab extensions, jupyter_tensorboard, is not installed on CUDA 11.0 related images because tensorflow conflict error occurs due to this issue⁠.

Dependency conflict:
jupyter_tensorboard requires tensorflow<2.2.0
CUDA 11.0           requires tensorflow>=2.4.0

⁠SavviHub Image

⁠Commands

# Use deploy script
python deploy.py --dockerhub {DOCKERHUB_REPO} -t {TAG} {DIR} --push

make push-base       # Build base images and push them to savvihub/kernels 
make push-experiment # Build experiment images and push them to savvihub/kernels
make push-service    # Build service images and push them to savvihub/kernels
make push-all        # Build all images and push them to savvihub/kernels

⁠Base

⁠Experiment

⁠Service (Jupyter)

⁠Public Image

⁠PyTorch

⁠Tensorflow