Add Docker deployment capability

[jincdream]

Add Docker deployment capability to the project.

• Dockerfile: Create a Dockerfile to define the Docker image for the project using Python 3.12, copy project files, install dependencies, and set the entry point.
• docker-compose.yml: Create a docker-compose.yml to define the Docker services, including building the project service, exposing necessary ports, and setting environment variables.
• .circleci/config.yml: Add a job to build and push the Docker image to a registry, and update the workflow to include the new Docker jobs.
• README.md: Add instructions to build and run the Docker image, and use docker-compose to run the services.

---

For more details, open the Copilot Workspace session.

[AlexCheema]

Did you test this on Apple Silicon?

[macaws]

@jincdream if you need any help testing on Apply Silicon, let me know - would be glad to assist if you do not have access to it.

[jincdream]

@jincdream if you need any help testing on Apply Silicon, let me know - would be glad to assist if you do not have access to it。

That's great! I haven't tested it yet and I'm so glad you're willing to lend a hand. Looking forward to working with you on this. Thanks a bunch!

[bencurio]

Hello! I have also created a Docker version, but GPU acceleration is a bit more complicated.

However, this results in a large disk image, and it's important that the host machine's CUDA version matches the CUDA version inside the Docker container; otherwise, it won't work.

Currently, I am only using this for testing:

Dockerfile

FROM nvidia/cuda:12.4.0-devel-ubuntu22.04
ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get -yqq update \
    && apt-get -yqq dist-upgrade \
    && apt-get -yqq upgrade \
    && apt-get -y --no-install-recommends install \
        python3 \
        python3-pip \
        python3-venv \
        python3-clang \
        clang \
        libgl1 \
        libglib2.0-0 \
        git \
    && rm -rf /var/lib/apt/lists/*

RUN useradd -m -d /home/container container \
    && usermod -aG adm,audio,video container

COPY entrypoint.sh /entrypoint.sh
RUN chmod +x /entrypoint.sh

USER container

RUN git clone https://github.com/exo-explore/exo.git ~/exo  \
    && python3 -m venv ~/exo/.venv \
    && cd ~/exo \
    && . ~/exo/.venv/bin/activate \
    && pip3 install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124 \
    && pip3 install --no-cache-dir -e . \
    && cd ~ \
    && deactivate

EXPOSE 52415/tcp

ENTRYPOINT ["/entrypoint.sh"]

entrypoint.sh

#!/usr/bin/env bash

cd ~/exo
source ~/exo/.venv/bin/activate
exo ${EXO_ARGS}

docker-compose.yml

services:
  docker_exo:
    network_mode: bridge
    build: .
    container_name: docker_exo_container
    restart: unless-stopped
    ports:
      - "52415:52415"
    environment:
      - NVIDIA_DRIVER_CAPABILITIES=gpu,utility,video,compute
      - EXO_ARGS=--discovery-module=udp --data=/opt/exo-data --models-seed-dir=/opt/exo-seed
    volumes:
      - ./opt/exo-data:/opt/exo-data
      - ./opt/exo-seed:/opt/exo-seed
      #devices:
      #- "/dev/dri:/dev/dri"  # GPU
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu,utility,video,compute]

With this approach, exo recognizes the GPU.

[s3rj1k]

With similar Dockerfile I get image of around 10GiB, this is a bit insane :)

[s3rj1k]

related: https://github.com/exo-explore/exo/pull/770