In this article, we’ll introduce distributed training and how it works by parallelizing the workload across multiple processors (data parallelism or model parallelism). Then, we’ll discuss how to choose between distributed machine learning tools.
Ray 1.12 is here! This release includes Ray AI Runtime (alpha), a new unified experience for Ray libraries. In addition, we’ve implemented a lightweight usage statistics and data collection mechanism in Ray (turned off by default).
Ray Summit, the annual Ray user conference, is back this August, and the Call for Papers is open until April 18! In this blog post, we’ll give you all the information you need to prepare a stellar talk proposal.
We’re holding the first-ever Production RL Summit on March 29. Read more about our featured speakers and sessions and learn why you should join if you’re a reinforcement learning practitioner — or even if you have just a passing interest in RL.
Ray 1.11 is here! Ray no longer starts Redis by default, opening up the possibility for better support of fault tolerance and high availability in future releases. Plus, there’s a new, more intuitive design for the Ray docs.
In this article, we’ll cover how to deploy XGBoost with two frameworks: Flask and Ray Serve. We’ll also highlight the advantages of Ray Serve over other serving solutions when comparing models in production.
In this article, we will highlight the options available for serving a PyTorch model into production and deploying it with several frameworks, such as TorchServe, Flask, and FastAPI.
In a previous blog post, we covered the advantages and disadvantages of several approaches for speeding up XGBoost model training. In this article, we’ll dive into three different approaches, with code snippets so you can follow along.
Ray 1.10 is here! The highlights include Windows support — now in beta — and enhancements to Ray job submission, including log streaming and custom headers for authentication.