Model Update

Model Registry service integration

DL Streamer Pipeline Server supports MLOps capabilities through integration with the Model Registry service. This enhancement streamlines the model management process, enabling dynamic updates and ensuring that DL Streamer Pipeline Server and ensuring that DL Streamer Pipeline Server utilizes the most accurate and up-to-date models. Currently it supports following features:

Model Initialization: Upon initialization, DL Streamer Pipeline Server can retrieve and deploy machine learning models directly from a Model Registry. To leverage this feature, users must specify the necessary model details in the configuration file.

Runtime Model Update: DL Streamer Pipeline Server offers a REST endpoint that facilitates on-demand model updates during runtime. This endpoint allows users to pull the latest model version from the Model Registry and download it for immediate use within the existing DL Streamer Pipeline Server pipeline. This endpoint also provides options to dynamically restart pipeline with new model.

Refer the detailed documentation for steps on launching the model registry microservice and configurations to interact with it during DL Streamer Pipeline Server’s startup and runtime.