Delete runs

Delete runs from a W&B project using the W&B App or the Public API, and learn how deleted run data is removed from storage on Dedicated Cloud and Self-Managed deployments.

Delete runs

Delete runs from a project with the W&B App or the Python API.

W&B App

1. Navigate to the project that contains the runs you want to delete.
2. Select the Runs tab.
3. Select the checkbox next to the runs you want to delete.
4. Choose the Delete button (trash can icon) above the table.
5. From the drawer that appears, choose Delete.

For projects that contain a large number of runs, you can use either the search bar to filter runs you want to delete using Regex or the filter button to filter runs based on their status, tags, or other properties.

Python

You can delete runs programmatically with Run.delete(). Set delete_artifacts=True if you also want to remove artifacts associated with the run.

import wandb

api = wandb.Api()
runs = api.runs("<entity>/<project>")
for run in runs:
    if run.state == "finished":  # Replace with your own condition
        run.delete(delete_artifacts=False)

For the full method signature and behavior, see the Run.delete reference.

Individual files attached to a run (for example logged media) can be removed with File.delete() after you obtain file handles from Run.files().

A run ID cannot be reused, even after the run is deleted. Instead, the run will fail with an error.

When you delete a run and choose to delete associated artifacts, the artifacts are permanently removed and can’t be recovered, even if the run is restored later. This includes artifacts linked to the Registry.

Run deletion flowchart

The following diagram illustrates the complete run deletion process, including the handling of associated artifacts and Registry links:

When deleted run data is removed from storage

On W&B Dedicated Cloud and W&B Self-Managed, the GORILLA_DATA_RETENTION_PERIOD environment variable controls how long deleted run data is retained before it can be permanently removed from object storage. Artifacts are not removed by this setting; they follow the artifact deletion and garbage collection flow described in Delete an artifact.

Setting or changing GORILLA_DATA_RETENTION_PERIOD is irreversible for data past the retention window. Back up your database and bucket before enabling or tightening retention. See Configure environment variables for the reference table and warnings.

Even after run or file deletion and retention processing, bucket usage can lag while background jobs catch up. W&B does not guarantee immediate reclamation of object storage. For a full overview of artifacts versus run data, timing expectations, and optional operator actions, see Manage bucket storage and costs.

If deletions do not appear as expected in the W&B App when using the Public API, upgrade the W&B Python SDK to a current release and retry.