Skip to main content
Use a lineage graph to visualize a linked artifact’s history. Audit a collection’s history to track changes made to artifacts in that collection.

Lineage graphs

Within a collection in the Registry, you can view a history of the artifacts that an ML experiment uses. This history is called a lineage graph. A lineage graph shows: In other words, a lineage graph shows the input and output of a run. For example, the following image shows a typical lineage graph for artifacts created and used throughout an ML experiment:
Registry lineage
From left to right, the image shows:
  1. Multiple runs log the split_zoo_dataset:v0 artifact.
  2. The astral-deulge-3 run uses the split_zoo_dataset:v0 artifact for training.
  3. The output of the astral-deulge-3 run is a model artifact called zoo-l4oqundu:v0.
  4. A run called denim-night-4 uses the model artifact zoo-l4oqundu:v0 to evaluate the model.
To view a lineage graph for an artifact in a collection:
  1. Navigate to Registry.
  2. Select the collection that contains the artifact.
  3. From the dropdown, select the artifact version you want to view its lineage graph.
  4. Select the Lineage tab.
  5. Select a node to view detailed information about the run or artifact.
See Enable lineage graph tracking to learn how to track the input and output of a run using the W&B Python SDK.
The following image shows the expanded detailed view of a run (astral-deluge-3) when you select a node in the lineage graph:
Expanded lineage node
The following image shows the expanded detailed view of an artifact (zoo-l4oqundu:v0) when you select an artifact node in the lineage graph:
Expanded artifact node details
The Created by run shows the run that created the artifact. In this example, it is the astral-deulge-3 run.
You can also view lineage graphs for artifacts you log to W&B that are not part of a collection. See Explore artifact lineage graphs for more information.

Create a custom view

Click Custom in the top-right corner of the lineage graph to create a custom view. You can filter and customize the lineage graph with the following options:
  • Filter by artifact type: Filter by artifact type values logged to W&B. For example, if you log an artifact with the type “dataset”, then “dataset” is available as a filter value.
  • Filter by run job type: Filter by run job type values logged to W&B. For example, if you log a run with the job type “training”, then “training” is available as a filter value.
  • Include extended lineage: Display items that are not in the direct lineage for the base version.
  • Include generated artifacts: Items created programmatically will show.
  • Expand clusters: Similar items with similar connections will not be grouped together.
  • Downstream hops: Descendant generations relative to the active node.
  • Upstream hops: Ancestor generations relative to the active node.
To reset the lineage graph to its default view, click the backward arrow button.

Audit a collection’s history

View actions that members of your organization take on that collection. You can view:
  • If an alias was added or removed from an artifact version.
  • If an artifact version was added or removed from a collection.
For both actions, you can view the user that performed the action and the date the action occurred. To view a collection’s action history:
  1. Navigate to Registry.
  2. Select the collection you want to view its action history.
  3. Select the dropdown menu next to the collection name.
  4. Select the Action History option.
Last modified on September 30, 2026