On Wednesday at the PyTorch Conference in San Francisco, the PyTorch Foundation announced that it will now host Ray, the popular open source distributed computing framework developed by Anyscale. The contribution of Ray by Anyscale to the PyTorch Foundation represents a milestone in the PyTorch Foundation's goal to provide developers with the necessary tools to efficiently build and scale AI applications without depending on fragmented workflows or locking themselves into proprietary tools.

Ray enables developers to cut down the complexity associated with efficiently scaling applications from a single node development environment (i.e. a laptop) to a multi-node environment suitable for handling the demands of AI and machine learning workloads. Ray already enjoys significant adoption by developers, as demonstrated by its over 39,000 GitHub stars and more than 237 million downloads, both accumulated since the project was first developed at UC Berkley.

Some of Ray's most notable features include: multimodal data processing, the possibility of scaling a variety of frameworks (such as PyTorch) for both pre-training and post-tuning, and distributed inference to ensure high throughput and low latency in models served for production. By incorporating Ray under its umbrella, the PyTorch Foundation will now be able to offer an integrated foundation for their AI and ML projects: PyTorch for model development, vLLM for inference, and Ray for distributed execution.

The involved parties commented on how the contribution of Ray to the PyTorch Foundation represents a step forward for the open source community. Robert Nishihara, co-founder of Anyscale highlighted how the PyTorch Foundation will help ensure Ray stays a critical community-driven toolkit for distributed computing. Similarly, Matt White, Executive Director of the PyTorch Foundation, said: "By bringing Ray under the PyTorch Foundation umbrella, alongside projects like vLLM and DeepSpeed, we are uniting the critical components needed to build next-generation AI systems."