Model Management

Models on OpenGradient are managed through individual model repositories. As models evolve, a repository allows multiple versions and iterations of the same model to be published and consumed without breaking existing consumers.

A model repository has a description, name, task, and other high-level information that helps define the overall goal of the model. Within each repository, multiple releases of the same model can be developed and released over time. Each model release contains a set of files that are associated with the model. In the simplest case, this could only be the model file (e.g., in onnx format), however we also support more complex architectures such as LLMs where a model is broken down into multiple files. In addition, you can upload example scripts, readme's, and anything else that might be useful to consumers.

To sum up, our Model Hub organizes models using the following concepts:

• Model Repository: a high-level container for a model that contains multiple iterations and releases
• Model Release: a single release or version of the model that is usable on its own
• Model File: files that represents the actual model, each file must belong to a particular release (eg. onnx file)

TIP

You can upload and manage models using the Hub website or the Python SDK.