Using artifacts with multiple architectures and runs?

This article describes how to organize artifacts when a project explores multiple model architectures across many runs, so you can track each architecture’s checkpoints independently.

You can version a model in several ways. Artifacts provide a tool for model versioning tailored to specific needs. For projects that explore multiple model architectures, a common approach separates artifacts by architecture. Consider the following steps:

1. Create a new artifact for each distinct model architecture. To provide detailed descriptions of the architecture, use the metadata attribute of artifacts, similar to how you use config for a run. Separating artifacts by architecture keeps each model’s version history isolated.
2. For each model, log checkpoints periodically with log_artifact. W&B builds a history of these checkpoints and labels the most recent one with the latest alias. To refer to the latest checkpoint for any model architecture, use [ARCHITECTURE-NAME]:latest.

This approach gives you a distinct, versioned history per architecture, with a stable latest reference for the most recent checkpoint of each.

---

Artifacts