Anyscale is the company behind Ray, an open source unified compute framework for scaling machine learning or Python workloads
Ray was built to make it easy to scale AI and Python workloads — from reinforcement learning to deep learning to tuning and model serving.
So then, it’s actually not a Shame About Ray.
The company used the end of last year to detail several new advancements on its branded Anyscale Platform.
The new capabilities extend beyond the advantages of Ray open source to make AI/ML and Python workload development, experimentation and scaling even easier for developers.
For fast development and iteration, the new Anyscale WorkspacesM environment is now available for early access.
Workspaces provides a unified developer experience to scale ML workloads from a laptop to the cloud with no code changes. Inside a single environment, developers can now build and move workloads to production while still leveraging familiar tools.
For accelerated development and rapid iteration, the Anyscale Platform adds the ability to startup clusters up to 5x faster than Ray open source.
Is that unfair to the open source community? Perhaps not, this is the commercial iteration of the product after all.
This means developers can speed up iteration, experimentation and deployments; job scheduling automation including auto-scaling, alerting, auto-retries etc. and custom cluster environments.
“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 for data, ML and technology at Biolexis Therapeutics.
Anyscale Workspaces provides a laptop like development experience to build and scale ML workloads. Workspaces enable developers to continue using the tools they are familiar with, including VS Code, Jupyter, the terminal etc. but leverage the scale and flexibility of the cloud.
Single script solution
With a single script a developer can prepare data, tune, train and deploy workloads at any scale.
According to Robert Nishihara, CEO and co-founder of Anyscale, Machine Learning model training and tuning is now inherently iterative and each iteration often requires cluster startup, tuning and shutdown events. “Anyscale shortens iteration cycles by taking cluster startup events down to under 2 minutes, up to 5X faster than Ray open source,” he said.
Organisations can now deploy their own custom docker images as Anyscale cluster environments and leverage their existing CI/CD pipelines to build and manage workloads running on Anyscale and Ray.
This includes launching Anyscale Workspaces, jobs and services while using their own docker tooling and infrastructure.