Basic ML

Base images with cuda support for varioud ML-related python workloads:

• Inference APIs
• Jupyter Environments
• Cluster Workers

---

Structure

All images are divided into two categories:

1. Base images Provide basic image with python, tini, gpg and CUDA-drivers if applicable

   

   

2. Target images Provide task-specific software and configuration

   2.1 Jupyter environments Image with pre-installed jupyter as well as multiple useful utilities. Jupyter Server is protected with password by default. You could either use default password or set your own via docker run -e PASSWORD="..." .... Exposed port is 8888.

   

   2.2 SSH-accessible environemnts Image with pre-installed and configured openssh-server and a script, that populates the system with either authentication option: password, single public key or whole ~/.ssh/authroized_keys file

   

---

Quick Start:

To run those images you will need Docker or any other container platform. In case of gpu system - docker must be configured with nvidia-docker2. Example script to run container is below:

docker run \
    --gpus=all \
    --ipc=host \
    -p 8888:8888 \
    -v `pwd`/notebooks:/workspace/notebooks \
    -v `pwd`/data:/workspace/data \
    -d \
    --name basic-ml \
        rexhaif/basic-ml:jupyter-cuda11.6-cudnn8-devel

To run SSH-accessible container use this script:

docker run \
    --gpus=all \
    --ipc=host \
    -p 2222:22 \
    -e AUTHORIZED_KEYS="`cat ~/.ssh/authorized_keys`" \
    -e PUBLIC_KEY="ssh-ec25519 ..." \
    -e PASSWORD="the-password" \
    -v `pwd`/projects:/workspace/code \
    -v `pwd`/data:/workspace/data \
    -d \
    --name basic-ml \
        rexhaif/basic-ml:ssh-cuda11.6-cudnn8-devel

You can choose any suitable option to configure authentication, either with one of AUTHORIZED_KEYS / PUBLIC_KEY / PASSWORD, or with all of them simultaneously. Port 2222 from -p 2222:22 can be changed to any port of your choice.