Skip to content

NVIDIA-Merlin/gcp-ml-ops

Repository files navigation

Merlin - MLOps on GKE

Introduction

NVIDIA Merlin is an open-source application framework that facilitates the development and deployment of large-scale deep recommender systems on GPUs.

The figure below shows the architecture of a recommendation system example using NVIDIA Merlin on a Kubeflow pipeline.

Through this, we intend to show an end-to-end reference architecture, all the way from data preparation, to model deployment, with features like continuous and fast re-training, autoscaling, and model monitoring.

Merlin-Kubeflow Architecture

For this example, we use the Criteo 1TB Click Logs dataset, a large publicly available dataset for recommender systems. It contains feature values and click feedback for millions of display ads. It is divided into 24 files, each one corresponding to one day of data.

Running the example

Please follow the User Guide available here!

About

MLOps pipeline for NVIDIA Merlin on GKE

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published