Distributed computing startup Anyscale Inc. said at AWS re:Invent today it’s introducing a number of updates to its platform aimed at making Python-based artificial intelligence and machine learning workload development and scaling easier for developers.
Anyscale is the company behind the open-source Python framework Ray, which is used to run distributed computing projects. Ray includes both a universal serverless compute application programming interface and an expanded ecosystem of libraries. They enable developers to build scalable applications that can run on multicloud platforms without needing to worry about the underlying infrastructure.
One of the key advantages of Ray is it eliminates the need for in-house distributed computing expertise.
The Anyscale cloud platform, meanwhile, is a managed version of Ray aimed at making it more accessible. The Ray platform requires a fair amount of expertise that typically only a few high-level developers and computing specialists possess. Anyscale’s platform, which runs on AWS, solves the difficulty of taking an artificial intelligence prototype built on a laptop and scaling that model across hundreds of machines in the cloud.
The new capabilities announced at re:Invent include the early access availability of the new Anyscale Workspaces environment, which is said to provide a unified and more seamless experience for developers as they scale machine learning workloads from a laptop to the cloud, without making any significant code changes. Developers now have a single environment to build machine learning workloads and move them to production, Anyscale said.
One of the main advantages of Anyscale Workspaces is that it allows developers to use the same set of familiar tools throughout the process, including VS Code and Jupyter, while reducing context-switching as they bring new machine learning models to cloud scale.
In a second update, the Anyscale Platform gains the ability to start up clusters up to five-times faster than is possible with the open-source Ray platform. Developers can therefore accelerate the iteration, experimentation and deployment of machine learning models, Anyscale said. Finally, Anyscale is adding new job scheduling automation capabilities. With this, developers now have a native way to schedule jobs and integrate them with third-party orchestration tools such as Airflow and Prefect, with auto-scaling, alerting, auto-retries and other capabilities available.
The updates are all about making machine learning developers faster and more productive, and early testers say they have the intended effect.
“In the same time that it took to actually run our original workload – a week – we were able to effortlessly migrate all our Python workloads to the Anyscale Platform, quickly fine tune jobs for scaling, and move to production at scale effortlessly,” said Jake Carter director of data, machine learning and technology at Biolexis Therapeutics LLC. “It was remarkable and literally saved us a week end-to-end.”
AWS Vice President and Global Head of Startups Howard Wright said enabling innovations such as the Anyscale Platform is exactly what the AWS cloud was designed for. “Making it easier for companies to build machine learning models that are mature, reliable and scalable with as little as two lines of code, is the type of added value that we are excited to help bring to the market with Anyscale and Ray,” he said.