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 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




