Deploying Models
Overview​
After building a model in Abacus.AI, the next step is to deploy it. Deploying a model makes it available for predictions and enables integration into your applications, regardless of the use case.
What You'll Learn​
In this guide, you will learn how to:
- Deploy a model
- Revert to a previous version or change the model completely
Deploying Your Model​
To deploy a model, navigate to the Model Training Page and click the Deploy Model button.
Deployment Types​
When deploying, you'll need to choose between two deployment types:
- Offline Deployment: Suitable for models that don't require real-time predictions. Ideal for ML models that run on input data using scheduled cron jobs.
- Online Deployment: Designed for models requiring real-time predictions, such as GenAI models or ML models integrated into live applications.
Monitoring Your Deployment​
For both deployment types, you can monitor the deployment status and view logs directly from the Abacus.AI platform.
Once deployed, you can view the deployment status from the deployment page:
Accessing the Prediction API​
To integrate your deployment externally, click on Prediction API to view the API documentation and calling instructions.
Key Deployment Settings​
Here is an overview of all important settings:
Auto Deployment Toggle​
- ON: When the model connected to this deployment is retrained, the deployment automatically updates to use the new model version.
- OFF: The deployment continues using the current model version even after retraining. You must explicitly update the deployment to point to the new model version.
Switching Model Versions​
You can select which model or version to deploy using the Switch Version Algorithm button:
Understanding Deployments and Models​
Key Concepts​
- Deployment-Model Relationship: Each deployment is connected to a particular model version, but you can change the version or the model itself at any point in time.
- Multiple Deployments: Each model can have multiple deployments, but each deployment is only connected to a single model version.
- Separation: Deployments are completely separated from the model, providing flexibility in version management and deployment strategies.