Parameter workload_size always executing SMALL when using the databricks-agents library to update existing model serving endpoints

Update the databricks-agents library to version 0.17.0 or later.

Written by kaushal.vachhani

Last published at: March 25th, 2025

Problem

When using the databricks-agents library to update an existing model serving endpoint, calling the agents.deploy() method with the workload_size parameter set to any value other than SMALL still results in deployment with workload size SMALL.

 

Example 

The following code snippet intends to redeploy a new model on the existing serving endpoint with the MEDIUM size. However, due to this issue, the endpoint incorrectly deploys using the default size SMALL.

agents.deploy(
    <your-uc-model-name>,
    <your-new-model-version>,
    workload_size=ServedModelInputWorkloadSize.MEDIUM
)

 

Cause

The specified workload_size parameter is not correctly applied when deploying a new agent version to the existing model-serving endpoint. 

 

Solution

Update the databricks-agents library to version 0.17.0 or later. This version includes a resolution to the issue, ensuring the workload_size parameter is correctly respected during deployments on existing model serving endpoints.