Automated CI/CD workflows triggered by W&B Automations. When a model is promoted in the W&B Registry (e.g., the production alias is added), a webhook fires and triggers a GitHub Actions workflow that creates a review issue.
- A model artifact is promoted in the W&B Model Registry (alias
productionis added) - W&B fires a webhook automation with artifact metadata (version, collection, author, project)
- The webhook triggers a GitHub Actions
repository_dispatchevent - The GitHub Action creates a review issue from a template, populated with the promotion details
- A W&B account with access to Automations (Pro or Enterprise plan)
- A GitHub Personal Access Token (PAT) with
reposcope or fine-grained permissions for Issues (read/write) and Contents (read)
- Add a repository secret called
ACTION_PATwith your GitHub PAT value (Settings -> Secrets and variables -> Actions)
- Create a Secret in W&B Team Settings: name it
GITHUB_TOKEN, paste your GitHub PAT - Create a Webhook in W&B Team Settings:
- Name:
github-model-cicd - Endpoint:
https://api.github.com/repos/<OWNER>/<REPO>/dispatches - Access token: select the
GITHUB_TOKENsecret
- Name:
- Create an Automation on the Registry or Project:
- Event: "An artifact alias is added"
- Alias regex:
^production$ - Action: Webhook ->
github-model-cicd - Payload:
{ "event_type": "webhook", "client_payload": { "event_author": "${event_author}", "artifact_version": "${artifact_version}", "artifact_version_string": "${artifact_version_string}", "artifact_collection_name": "${artifact_collection_name}", "project_name": "${project_name}", "entity_name": "${entity_name}" } }
- In W&B Team Settings, find your webhook and click the
...menu -> Test - Paste the payload above (replace
${}variables with sample values) and click "Test webhook" - Check the GitHub Actions tab — a workflow should run and create an issue
.github/
PROMOTION_TEMPLATE.md # Issue template with artifact metadata
workflows/
model_promotion_ci_cd.yml # GitHub Actions workflow triggered by webhook
This example creates a GitHub Issue for human review. In production, the same pattern can trigger:
- Model evaluation on a held-out benchmark
- Deployment to a serving endpoint (SageMaker, Vertex AI, etc.)
- Slack/Teams notifications to the ML platform team
- A PR that updates a model config file in a deployment repo
The webhook payload passes all the artifact metadata needed to identify and fetch the model programmatically via the W&B Public API.