Uploading Models

The Inferium Store has made the process of uploading models simple, intuitive, and team-based. Here's a step-by-step tutorial on using Inferium to upload and manage your AI models:

Setting Up an Account

To get started, you need to create an account on Inferium. This account will grant you access to a range of features, including versioning, branches, discoverability, and integration with multiple libraries.

Repository Management

On Inferium, repositories based on Git are used to manage models, providing strong version control. This makes it possible for developers to keep track of modifications, work together, and maintain many iterations of their models. You have the option to associate your repositories with a person or organization account.

Individual Repositories:

For personal projects, associate your model with a specific account. One possible name for a repository would be username/model_name.

Organization Repositories:

For models associated with a company, community, or library, link your repository to an organization. This way, the model will be featured on the organization’s page, and every member can contribute. For example, company/model_name.

Flexibility and Compatibility

Models from several libraries and frameworks are supported by Inferium. Regardless of whether your model is a custom model or works with well-known libraries like Transformers or Diffusers, Inferium offers the infrastructure required to efficiently integrate and maintain various models.

Custom Model Support:

Custom models, including those using trust_remote_code=True, are fully supported. This ensures that even bespoke models can leverage the platform’s features, such as download metrics and discoverability.

Library-Specific Integration:

If your model is designed for a specific library with built-in support, you can use the methods provided by that library for seamless integration.

Upload Methods

Inferium offers multiple ways to upload your models:

1. Programmatic Uploads:

   For programmatic integration, use methods provided by supported libraries. For instance, PyTorch models can use the PyTorchModelHubMixin class to add from_pretrained and push_to_hub capabilities to any nn.Module class.

2. Web Interface:

   You can also upload models directly through the Inferium web interface. This user-friendly option is perfect for those who prefer a graphical interface over command-line tools.

Enhancing Discoverability

Once your model is uploaded, it's essential to make it easily discoverable and well-documented. Inferium recommends adding a Model Card to your repository. A Model Card provides comprehensive documentation of your model, including its intended use, performance metrics, and any other relevant information. This practice not only improves discoverability but also helps users understand and effectively utilize your model.