CSGHub Omnibus Deployment Guide

Tips:

This method is currently in the testing phase and is not suitable for production deployment.

Now supports amd64/arm64 architecture (supports Docker Desktop Rosetta).

中文文档

Overview

Omnibus CSGHub is a way for OpenCSG to quickly deploy CSGHub using Docker, mainly for rapid functional experience and testing. The Docker deployment method allows users to deploy CSGHub on local computers at a lower cost. This deployment method is very suitable for proof of concept and testing, allowing users to immediately access the core functions of CSGHub (including models, dataset management, space application creation, and model inference and fine-tuning (GPU required)).

Advantages

Quick configuration: Supports one-click deployment and quick start.
Unified management: Supports integrated model, dataset, space application management, and built-in multi-source synchronization function.
Simple operation: Supports model inference, fine-tune instance quick start.

Deployment

Prerequisites

The host used for deployment has Docker Desktop or Docker Engine installed.
The operating system is Linux, macOS, Windows, and the configuration is not less than 4c8g.
Fine-tune instances requires GPU (currently only supports NVIDIA)

Installation Steps

Tips:

HTTPS access configuration is not supported yet, you can adjust the Nginx configuration in the container yourself.

IfSERVER_DOMAINandSERVER_PORTare modified, it is recommended to delete the persistent data directory and recreate it.

Cloud serverSERVER_DOMAIN = <external public ip>

Note:

Please make sure that your local IP address segment and the docker default address segment (172.17.0.0) do not overlap. If they overlap, please try changing the local network connection (for example, changing the Ethernet network).

Quick Installation (Space and model inference & fine-tuning functions cannot be used)

Linux

Tips:

Please replace <your ip address> with your host IPv4 address.

To view the IPv4 address, enter the following command in the terminal command line:

ip -4 -o addr show $(ip route show default | awk '/default/ {print $5}')

Quick start without data persistence volumes

export SERVER_DOMAIN=$(ip addr show $(ip route show default | awk '/default/ {print $5}') | grep 'inet ' | awk '{print $2}' | cut -d/ -f1)
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 8000:8000 \
    -p 9000:9000 \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

Normal startup with persistent data volumes

export SERVER_DOMAIN=$(ip addr show $(ip route show default | awk '/default/ {print $5}') | grep 'inet ' | awk '{print $2}' | cut -d/ -f1)
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 8000:8000 \
    -p 9000:9000 \
    -v /srv/csghub/data:/var/opt \
    -v /srv/csghub/log:/var/log \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

macOS

Tips:

To deploy with Docker Desktop, please enable Rosetta. The method to enable it is as follows:

Settings > General > Use Rosetta for x86_64/amd64 emulation on Apple Silicon

Please replace <your ip address> with your host IPv4 address.

To view the IPv4 address, enter the following command in the terminal command line:

ipconfig getifaddr $(route get default | grep interface | awk '{print $2}')

Note: Rosetta runs Slightly slower. Before version v1.2.0, container running in Rosetta mode will prompt WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested. Just ignore it.

Manually pull the image

docker pull opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

Quick start without data persistence volumes

export SERVER_DOMAIN=$(ipconfig getifaddr $(route get default | grep interface | awk '{print $2}'))
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 8000:8000 \
    -p 9000:9000 \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

Normal startup with persistent data volumes

export SERVER_DOMAIN=$(ipconfig getifaddr $(route get default | grep interface | awk '{print $2}'))
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 8000:8000 \
    -p 9000:9000 \
    -v ~/Documents/csghub/data:/var/opt \
    -v ~/Documents/csghub/log:/var/log \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

Windows

Tips:

Please replace <your ip address> with your host IPv4 address.

To view the IPv4 address:

Use the key combination Win + R, enter cmd, and after the window opens, enter ipconfig to obtain the IPv4 address.

Powershell

$env:SERVER_DOMAIN = ((Get-NetAdapter -Physical | Get-NetIPAddress -AddressFamily IPv4)[0].IPAddress) 
$env:SERVER_PORT = "80"
docker run -it -d `
    --name omnibus-csghub `
    --hostname omnibus-csghub `
    -p ${env:SERVER_PORT}:80 `
    -p 2222:2222 `
    -p 8000:8000 `
    -p 9000:9000 `
    -e SERVER_DOMAIN=$env:SERVER_DOMAIN `
    -e SERVER_PORT=$env:SERVER_PORT `
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

CMD

set SERVER_DOMAIN=<your ip address>
set SERVER_PORT=80
docker run -it -d ^
    --name omnibus-csghub ^
    --hostname omnibus-csghub ^
    -p %SERVER_PORT%:80 ^
    -p 2222:2222 ^
    -p 8000:8000 ^
    -p 9000:9000 ^
    -e SERVER_DOMAIN=%SERVER_DOMAIN% ^
    -e SERVER_PORT=%SERVER_PORT% ^
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

WSL

Please refer to the Linux deployment method.

General installation (Space, model inference & fine-tuning features can be used (requires NVIDIA GPU))

Linux

Prerequisites:

A Kubernetes cluster with Knative Serving deployed is required.

For other precautions, see the Quick Installation section.

Quickly configure the k8s cluster

curl -sfL https://raw.githubusercontent.com/OpenCSGs/csghub-installer/refs/heads/main/docker/scripts/k3s-install.sh | bash -s

# If enable NVIDIA GPU
curl -sfL https://raw.githubusercontent.com/OpenCSGs/csghub-installer/refs/heads/main/docker/scripts/k3s-install.sh | ENABLE_NVIDIA_GPU=true bash -s

Configure Docker

# Add insecure registry
cat <<EOF > /etc/docker/daemon.json
{
  "insecure-registries": ["<your ip address>:5000"]
}
EOF

# Restart docker
systemctl restart docker

Install CSGHub

export SERVER_DOMAIN=$(ip addr show $(ip route show default | awk '/default/ {print $5}') | grep 'inet ' | awk '{print $2}' | cut -d/ -f1)
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 5000:5000 \
    -p 8000:8000 \
    -p 9000:9000 \
    -v /srv/csghub/data:/var/opt \
    -v /srv/csghub/log:/var/log \
    -v ~/.kube:/etc/.kube \
    -v /var/run/docker.sock:/var/run/docker.sock \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

macOS/Windows

Please configure Kubernetes cluster yourself and make sure the file ~/.kube/config exists. Then use a command similar to the following to install it:

macOS

export SERVER_DOMAIN=$(ipconfig getifaddr $(route get default | grep interface | awk '{print $2}'))
export SERVER_PORT=80
docker run -it -d \
    --name omnibus-csghub \
    --hostname omnibus-csghub \
    -p ${SERVER_PORT}:80 \
    -p 2222:2222 \
    -p 5000:5000 \
    -p 8000:8000 \
    -p 9000:9000 \
    -v ~/Documents/csghub/data:/var/opt \
    -v ~/Documents/csghub/log:/var/log \
    -v ~/.kube:/etc/.kube \
    -v /var/run/docker.sock:/var/run/docker.sock \
    -e SERVER_DOMAIN=${SERVER_DOMAIN} \
    -e SERVER_PORT=${SERVER_PORT} \
    opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsg_public/omnibus-csghub:latest

Windows

Not supported yet.

Visit CSGHub

After the above deployment is successful, use the following method to access:

Access address: http://<SERVER_DOMAIN>:<SERVER_PORT>, for example http://192.168.1.12

Default administrator: root

Default password: Root@1234

Command Line Tools

Currently, a simple command line tool is provided for restarting services and viewing logs separately.

Service Managemnet

# View all services status
csghub-ctl status 

# View all jobs status 
# EXITED is expected status
csghub-ctl jobs

# Restart a service
csghub-ctl restart nginx

Log

# View all service logs in real time
csghub-ctl tail 

# View a service log in real time
csghub-ctl tail nginx

Others

All other command options are inherited from supervisorctl.

Function Exploration

CSGHub provides several key functions:

Model hosting:
- Currently supports model hosting, easy upload and management of models.
- Supports creating inference and fine-tuning instances (NVIDIA GPU required).
- Multi-source synchronization is enabled by default, and multi-source synchronization will automatically start after startup (synchronization takes a while to complete).
Dataset hosting:
- Simplified tools for processing datasets, ideal for rapid testing and verification.
Application hosting:
- Quickly create large model applications through custom programs and model combinations.

Destroy the container

If you are not using or need to rebuild the container, you can do the following:

docker rm -f omnibus-csghub

If you also need to uninstall the k8s environment, you can do the following:

/usr/local/bin/k3s-uninstall.sh

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

CSGHub Omnibus Deployment Guide

Overview

Advantages

Deployment

Prerequisites

Installation Steps

Quick Installation (Space and model inference & fine-tuning functions cannot be used)

General installation (Space, model inference & fine-tuning features can be used (requires NVIDIA GPU))

Visit CSGHub

Command Line Tools

Function Exploration

Destroy the container

Files

README.md

Latest commit

History

README.md

File metadata and controls

CSGHub Omnibus Deployment Guide

Overview

Advantages

Deployment

Prerequisites

Installation Steps

Quick Installation (Space and model inference & fine-tuning functions cannot be used)

General installation (Space, model inference & fine-tuning features can be used (requires NVIDIA GPU))

Visit CSGHub

Command Line Tools

Function Exploration

Destroy the container