User documentation

Useful links

Homepage

A high level overview of the project.

Documentation

The main source of knowledge on how to use the project. Refer always to here in case of doubt.

Authentication

The authentication management for accessing the AI4OS stack.

Dashboard

Where users will typically search for modules developed by the community, and find the relevant pointers to use them. It allows authenticated users to deploy virtual machines on specific hardware (eg. gpus) to train a module.

NextCloud

The service that allows to store your data remotely and access them from inside your deployment.

Github

The code of all the modules and services behind the project is stored.

DockerHub

Where the Docker images of the modules are stored.

CI/CD pipeline

Continuous Integration and Continuous Development Jenkins instance to keep everything up-to-date with latest code changes.

Inference platform (OSCAR)

Scalable serverless inference of AI models.

Status of services

Check if a specific AI4OS service might be down for some reason.

Module template

Create new modules based on our project’s template.

New to the project? How about a quick dive?

• Quickstart
  • Run a module locally
  • Train a module on AI4OS dashboard
  • Develop and share your own module

Overview

A more in depth documentation, with detailed description on the architecture and components is provided in the following sections.

• AI4OS architecture
  • The Dashboard
  • The DEEPaaS API
  • The data storage resources
  • The Inference Platform (OSCAR)
• User roles and workflows
  • The basic user
  • The intermediate user
  • The advanced user
• Authentication
  • How to create an account
• AI4OS Modules
  • CI /CD pipeline
• AI4OS Modules Template
  • Create your project based on the template
  • Project structure
• DEEPaaS API
  • Integrate your model with the API
  • Running the API
• AI4OS Dashboard
  • Selecting the modules
  • Making a deployment
  • Managing the deployments
• AI4OS Storage
• AI4OS Inference

How-to’s

Use a model (basic user)

• Perform inference locally
  • 1. Choose your module
  • 2. Launch the API and predict
• Perform inference remotely
  • 1. Choose your module and deploy
  • 2. Accessing the module and make predictions

Train a model (intermediate user)

• Train a model locally
  • 1. Choose a module from the Marketplace
  • 2. Prepare your dataset
  • 3. Run your module
  • 4. Open the DEEPaaS API and train the model
  • 5. Test and export the newly trained model
  • 6. Next steps
• Train a model remotely
  • 1. Choose a module from the Marketplace
  • 2. Upload your files to Nextcloud
  • 3. Deploy with the Training Dashboard
  • 4. Go to JupyterLab and mount your dataset
  • 5. Open the DEEPaaS API and train the model
  • 6. Test and export the newly trained model
  • 7. Create a Docker repo for your new module
  • 8. Share your new module in the Marketplace
  • 9. [optional] Add your new module to the original Continuous Integration pipeline
• Use rclone
  • Installing rclone
  • Configuring rclone
  • Using rclone

Develop a model (advanced user)

• Develop a model
  • 1. Setting the framework
  • 2. Editing <project-name> code
  • 3. Editing DEEP-OC-<project-name> code
  • 4. Integrating the module in the Marketplace

Use a tool (intermediate user)

We have specific guides for each of the tools.

• Federated server
  • Deploying a Federated server
  • Federated learning training in AI4EOSC

Others

• Useful Machine Learning resources
  • Tutorials
  • Datasets
  • Models
• Video demos