EZAI User Manual

Table of Contents

1. Introduction
2. System Requirements
3. Installation
4. Operation
   • Website URL
   • Registering Models
   • Deploying Models
   • Managing Jobs
5. Troubleshooting
6. Security
7. Support

---

Introduction

EZAI is a tool designed to manage and deploy machine learning models with ease. It uses Celery workers, SageMaker, and a structured relational database to facilitate efficient model management, allowing users to register and deploy models, track job statuses, and manage model endpoints.

Note: All the data are user generated content. Therefore you'll be shown an empty database upon hosting it.

System Requirements

• Python 3.9 or higher
• Redis and RabbitMQ for Celery
• AWS SageMaker with appropriate permissions
• PostgreSQL
• Docker (for containerization)

Installation

1. Clone the Repository:

   git clone https://github.com/fishpain/ezai.git
   cd ezai

2. Environment Configuration:

   • Copy the .env.example file to .env:

     cp .env.example .env

   • Update .env with your environment-specific variables, such as DATABASE_URI, RABBITMQ_URI, and AWS credentials for SageMaker.

   Note: Email 2302705 at sit.singaporetech.edu.sg for env file.

3. Build Application, RabbitMQ, Database through:

   cd hf_api/docker && docker compose -f docker-compose-local-x86.yml up -d

   Note: Above command is to build for X86 platform, aka windows. To build on ARM achitechture, you have to edit the requirements.txt file and choose libs/h5py-3.9.0-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl instead. Then build the docker-compose-local-arm.yml instead.

Operation

Website URL

• Website URL available at localhost:5051/
• Swagger API Available at localhost:5051/v1/docs/

Application supports Chrome and Safari. Not IE.

Logging In

• Users authenticate via a token-based system. Provide a valid token for operations that require user-specific access.

Note: Please enable cookie and javascript on your browser.

Registering Models

• Call the /register endpoint or use the register_model_worker task to register a new model with SageMaker. This will:
  • Create an entry in the MLModel table.
  • Register the model version in ModelRegistryModel.
  • Deploy the model to an endpoint on SageMaker.

Deploying Models

• Deploy registered models to SageMaker by triggering the register_model_worker task, which handles the creation and deployment to the appropriate endpoint.

Managing Jobs

• Monitor job statuses via JobsModel. The system automatically updates job statuses as tasks progress.
• View contributions and model statistics via the get_model_run_counts_with_details function.

Troubleshooting

• Database Connection Issues: Ensure that DATABASE_URI is correctly configured and that your database server is running.

• Celery Worker Not Responding:

  • Verify that RabbitMQ and Redis are accessible.
  • Check the worker logs for specific errors.

• AWS SageMaker Permissions:

  • Ensure your AWS role has sufficient permissions for deploying and managing models in SageMaker.

• 401 Unauthorized Error:

  • Ensure you provide a valid JWT token in the Authorization header.

Security

• Database Access:

  • Ensure that the worker user has restricted access to sensitive tables.
  • Use separate database users for different parts of the system for fine-grained access control.

• Token Management:

  • Store the SECRET_KEY securely.
  • Rotate the JWT secret regularly for enhanced security.

Support

For further support, please contact the development team or refer to the official documentation on GitHub at GitHub Repository.