aelheloudb / databricks-model-serving

SRTI demo for the field

Databricks Model Serving Endpoints Python Client

---

---

Introduction

• This is a Python framework which wraps Databricks Model Serving Endpoints API functionality.
• With a few lines of code, you can:
  • Deploy realtime models
  • Distribute traffic across two or more models running under the same endpoint (e.g. for A/B testing)
  • Inspect model build and server logs

Getting Started

To get started, simply install the package from this repo:

pip install git+https://github.com/sebrahimi1988/databricks-model-serving

Once the package is installed, you can leverage different functions in the EndpointClient class to list, create and update endpoints, amongst others. For instance, to list all model serving endpoints from a particular workspace:

from databricks.model_serving.client import EndpointClient

client = EndpointClient(databricks_url, databricks_token)
client.list_inference_endpoints()

Examples

In the notebooks folder you can find an example use case, where we train a model, register it in Model Registry and deploy it using the framework.

• 01_train trains two models and registers them in model registry prior to serving.

• 02_serve demonstrates how to serve models using Model Serving.

• 03_debug shows you how to investigate logs of the endpoint for debugging purposes.