Deployment types
There are three types of Deployments: managed, external, and registration.
Managed Deployments
A model, along with an optional explainer and/or transformer, is deployed on Deeploy using one of the available deployment services (KServe, SageMaker, or Azure Machine Learning). Once deployed, Deeploy provides API endpoints to do inference and other operations.
External Deployments
A model deployed on external infrastructure is accessed through an external endpoint. Deeploy provides an endpoint for external Deployments to retrieve predictions with, but some functionalities, specifically explanation requests and container logs, are not supported for this type.
For examples, check out the available integrations with Azure OpenAI and IBM Watsonx
Registration Deployments
Registration Deployments can be used to register any model, or even to register and document ideas prior to development. Deeploy provides an endpoint for registration Deployments, which can be used to send monitoring data to the Deployment. Select functionalities are available for registration Deployments, including model and data cards, metadata, and compliance documentation.
Available features per type
The following table provides an overview of which features are available for each Deployment type:
| Managed | External | Registration | |
|---|---|---|---|
| Documentation | ✓ | ✓ | ✓ |
| Model & data card | ✓ | ✓ | ✓ |
| Metadata | ✓ | ✓ | ✓ |
| Events | ✓ | ✓ | ✓ |
| Monitoring | ✓ | ✓ | ✓* |
| Predict endpoint | ✓ | ✓ | |
| Evaluations | ✓ | ✓ | |
| Actuals | ✓ | ✓ | |
| Alerts | ✓ | ✓ | |
| Tokens | ✓ | ✓ | |
| Test | ✓ | ✓ | |
| Prediction logs | ✓ | ✓ | |
| Code snippets | ✓ | ✓ | |
| Explain endpoint | ✓ | ||
| Job schedules | ✓ |
- Only tracing & custom metric monitoring are available for registration Deployments
Upgrading Deployments
Upgrade your Deployment to a different Deployment type to unlock more features. See upgrading Deployments for more information.