Ray, an open-source machine learning (ML) technology, has just made a giant leap forward with the release of version 2.4. It streamlines the process of deploying and scaling AI workloads, specifically focusing on accelerating generative AI workloads. Thanks to a strong open-source community of contributors and the help of commercial vendor Anyscale, Ray is one of the most trusted and widely used ML technologies. Its popularity has been further cemented by OpenAI, a vendor behind some of the most cutting-edge machine learning projects such as the GPT-4 and ChatGPT. Ray is not only used for training, but also for AI inference.
In August 2022, Ray 2.x paved the way to a more efficient and dynamic ML framework. To build on the momentum, the Ray 2.2 release set its sights on improving the observability of the system. With the new Ray 2.4 upgrade now available, users can get up and running with generative AI right away. That’s because the release comes with a set of pre-built scripts and configurations that save time and eliminate the need to configure and script each and every type of deployment manually. For example, Ray 2.4 integrates models from Hugging Face including GPT-J for text, Stable Diffusion for image generation, and the popular LangChain tool. It also offers improved performance through code optimization and the Ray AI Runtime (AIR) feature, with newer integrated trainers for ML training frameworks such as Hugging Face Accelerate and DeepSpeed, as well as PyTorch Lightning.
Anyscale is a commercial vendor that powers the Ray platform and contributes to the open source community. It was co-founded in 2020 by Robert Nishihara who serves as the company’s CEO. He explains that the goal of Ray 2.4 is to enable businesses to "integrate AI into their products" by "reducing the level of expertise" needed to successfully deploy and use generative AI models. All it takes is a single click to set up a working LLM that is already performing well and can be further modified and optimized to fit individual needs.
Ray is a great open-source tool that provides viable solutions in the AI space. It has continuously evolved during the past few years with the release of improved versions. Developers and AI enthusiasts around the world have much to look forward to with the Ray 2.4 upgrade and its ability to speed up generative AI model deployment.
