Scaling LLM Workloads on TPUs with Ray: From Pre-Training to Inference | Google | Ray Summit 2026 @anyscale
Scaling LLM Workloads on TPUs with Ray: From Pre-Training to Inference | Google | Ray Summit 2026  @anyscale
Uploaded September 2026 | Updated November 2026, 3 days ago
Scaling models across pre-training, post-training, and inference introduces complex distributed TPU orchestration challenges: topology-aware placement, coordinated scheduling, and failure recovery require abstractions beyond standard Kubernetes primitives.

At Ray Summit 2026, Ryan O'Leary, Software Engineer at Google, and Lehui Liu, Software Engineer at Anyscale, explore how Ray on GKE establishes a unified operational model for end-to-end TPU workloads, featuring elastic multi-slice training, multi-host inference, and performant PyTorch RL support scaled natively on Cloud TPUs.

You'll leave with an operational model for TPU workloads on Ray.

Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo

Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale

đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
Scaling LLM Workloads on TPUs with Ray: From Pre-Training to Inference | Google | Ray Summit 2026Whats new in Ray and Anyscale: KubeRay Connect, Native RDMA, and more | Ray Summit 2026Anyscale on Azure: Build and deploy AI at scale in your own tenantWhy Current AI Cannot Truly See the World | Elorian AI | Ray Summit 2026Prompt Learning: A Reinforcement Learning-Inspired Approach to AI Optimization | Ray Summit 2025Introducing Terminal-Bench: Evaluating LLM Agents in Realistic Terminal Settings | Ray Summit 2025
Anyscale |

Scaling LLM Workloads on TPUs with Ray: From Pre-Training to Inference | Google | Ray Summit 2026

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER