MLOps with Ray: Best Practices and Strategies for Adopting Machine Learning Operations

© Hien Luu, Max Pumperla and Zhe Zhang 2024

H. Luu et al.MLOps with Rayhttps://doi.org/10.1007/979-8-8688-0376-5_7

7. Ray Core

Hien Luu¹  , Max Pumperla² and Zhe Zhang³

(1)

Santa Clara, CA, USA

(2)

Bad Segeberg, Germany

(3)

Sunnyvale, CA, USA

 

As discussed in previous chapters, MLOps is becoming essential in today’s AI landscape. A central challenge in MLOps is to manage ML-related computation, including loading data from external storage, performing necessary last-mile transformations, training ML models, tuning hyperparameters, and generating inference results.

Many companies that begin their machine learning journey often start with a single machine to perform these computation tasks. For example, a single p4d.24xlarge instance on AWS EC2 has 8 NVIDIA ...