Anyscale announced the latest Ray 1.0 version at the first annual Ray summit in front of the attendees. Anyscale also announced a private beta version of Anyscale’s managed Ray platform.
Generally, developers have to develop applications in tandem with infrastructure for the application functioning. But the drawback is high development costs and a lack of interoperability which is associated with infrastructure development. In the case of Ray 1.0, developers can develop applications and scalable libraries without focusing on infrastructure. Ray 1.0 provides a general-purpose API which increases the set of applications that can be developed and run it in a scalable serverless fashion. This kind of service is not even provided with the latest specialized systems.
Ray aims to build the most comprehensive library ecosystem for distributed applications. Taking this forward, Ray 1.0 plans to expand the existing support of native libraries like Ray Serve, RaySGD, RLlib, and Tune. Ray 1.0 also supports popular python libraries like Hugging Face, spaCy, Mars, Horovod, PyTorch, Hyperopt, Optuna, Dask, and Modin for training, NLP, hyperparameter tuning, and other workloads.
Mr. Robert Nishihara is the co-creator of Ray and CEO of Anyscale. According to him, Ray 1.0 indicates the significant investment in the stability of the project, its maturity, and production-readiness. Due to this, Ray 1.0 release impacts the entire application ecosystem of multiple libraries developed by the Ray.
This re-definition of serverless computing frees the developer from managing the server. It also enables them to move seamlessly between their laptops and the cloud without any need for changes in codes for it. Anyscale indeed dramatically simplifies the developing process and productionisation of distributed applications. But it is offered in invitation-only beta for now.
With the vision of Ray and Anyscale platform to make developing and running distributed applications as simple as programming on one’s laptop, they begin inviting participants to try the Anyscale platform.
Jason Dai, Senior Principal Engineer, Intel said, “At Intel, we’re using Ray in many ways to support our AI applications. Not only do we leverage it through Analytics Zoo with RayOnSpark, but we also leverage Ray for hyperparameter search, model selection, and AutoML. Ray and its libraries have proven invaluable in meeting the demands of AI workloads for our users.
