Running DE & ML Workloads at Scale

What is Ray? A gentle introduction to Ray Core by example.

Ray provides flexible, distributed Python for developers and ML practitioners. It is powered by the concise Ray Core API, whose six main API calls you can quickly pick up during this talk. We'll look at some examples together that show you what you can build with just Ray Core alone. Along the way we touch on the basic architecture of Ray clusters and discuss how Ray's ML libraries are built on top of Core.

How Ray powers large-scale apps like ChatGPT. An overview of the AI Runtime (AIR)

OpenAI trained ChatGPT on Ray, and companies like Uber or Shopify bet on Ray to build their ML platforms on top of it. We briefly discuss all components of the Ray AI Runtime (AIR), from large-scale ML ingest using Datasets, to model deployments with Ray Serve, that can be used in complex ML applications. We then give you a brief overview of the technical intricacies behind systems like ChatGPT and how you can use Ray as the single distributed system for ML projects at scale.

An architecture for scaling data workloads and data teams

Data architectures are required to cover many different aspects for leveraging the underlying data to generate business value. Two of the most crucial ones are

• Have your businesses data available to enable data driven use cases
• Implementing use cases on that data which derive insights for the business

These two aspects are dependent on each other. Both in terms of technology, but also regarding engineering resources: If no data is available, no insights can be derived. Likewise, if no insights will be derived, the data is of no value.
This talk showcases an architecture that aims for scaling DE & ML workloads in both the technical sense (speedup,...) but also in terms of keeping the barrier between DE & ML workloads low while still being able to iterate independently between these workloads.
Concretely Ray & Spark will be leveraged for demonstrating that architecture.

Book Signing

Book Signing for the recently published book "Learning Ray" with author Max Pumperla will be available in breaks and after the meetup.

Speaker Bio

Max is a data science professor and software engineer located in Hamburg, Germany. He’s an active open source contributor, maintainer of several Python packages, and author of machine learning books. He currently works as a software engineer at Anyscale. As head of product research at Pathmind Inc. he was developing reinforcement learning solutions for industrial applications at scale using Ray RLlib, Serve and Tune. Max is a Ray contributor, has been a core developer of DL4J at Skymind, and helped grow and extend the Keras ecosystem.

Kolja is a Data & ML Engineer at inovex GmbH. He designs and implements data products that cover the whole data engineering lifecycle to solve business needs. His focus lies on data architectures and shipping data applications like Streaming- and ML applications to production.

Agenda

The time already contains time for Q&A.
Whether the MeetUp is in English or German, we decide on the spot and according to the participants.

18:00 Uhr | Doors open
18:30 Uhr | Talk: What is Ray? A gentle introduction to Ray Core by example.
19:00 Uhr | Short Break
19:15 Uhr | Talk: How Ray powers large-scale apps like ChatGPT. An overview of the AI Runtime (AIR)
19:45 Uhr | Talk: An architecture for scaling data workloads and data teams
20:30 Uhr | Closing, drinks & Book Signing