Registries

The LLMOS platform supports private registries for storing models and datasets. Registries provide a secure and centralized way to manage your machine learning assets, making them easily accessible for model services and other applications within your LLMOS cluster.

Overview

A registry is a private repository used to store models or datasets. The current backend storage only supports object storage solutions, providing scalable and reliable storage for your machine learning assets.

note

Important Requirements:

• Object storage must have CORS (Cross-Origin Resource Sharing) configured
• Only HTTPS endpoints are supported for security reasons

Creating a Registry

You can create one or more registries from the AI Infrastructure Management > Model Management > Registries page.

General Configuration

1. Name: Enter a unique name for your registry.
2. Description: Provide a description that better describes this resource.
3. Backend Type: Select S3 as the storage backend (currently the only supported option).

Configure S3 Backend

When configuring the S3 backend, you need to provide the following information:

1. Endpoint: The S3-compatible endpoint URL (e.g., s3.registry.com:5000)

   • Must s3 server use HTTPS protocol to ensure secure communication
   • Verify that the endpoint is reachable from your LLMOS cluster

2. Bucket: The name of the S3 bucket where your models and datasets will be stored (e.g., bucket)

3. Use Https: This checkbox is enabled by default and cannot be disabled, as only HTTPS connections are supported.

4. Secret Name: Select an existing secret that contains your S3 credentials, or create a new one.

   • The secret should contain your S3 access key and secret key
   • See Adding S3 Credentials for detailed instructions

Managing Registries

Once created, registries appear in the registries list with the following information:

• State: Shows whether the registry is active and accessible
• Name: The registry name you specified
• Endpoint: The configured S3 endpoint
• Bucket: The associated S3 bucket
• Default: Indicates if this is the default registry for the namespace

Registry Actions

For each registry, you can:

• Edit: Modify registry configuration
• Delete: Remove the registry (ensure no models or datasets are dependent on it)
• Set as Default: Make this registry the default for new model and dataset operations

Using Registries

Once configured, registries can be used to:

1. Store Models: Upload and manage machine learning models
2. Store Datasets: Organize and version your training and validation datasets

CORS Configuration

warning

CORS Setup Required

Your S3-compatible object storage must be configured with appropriate CORS settings to allow LLMOS to access the registry. Without proper CORS configuration, upload and download operations for models and datasets will fail.

Example CORS configuration for your S3 bucket:

[
  {
    "AllowedHeaders": ["*"],
    "AllowedMethods": ["GET", "PUT", "POST", "DELETE", "HEAD"],
    "AllowedOrigins": ["https://your-llmos-domain.com"],
    "ExposeHeaders": ["ETag"]
  }
]