Managing massive deployments of accelerators for AI and high performance computing (HPC) workloads can be complex. This talk dives into running AI-optimized Google Kubernetes Engine (GKE) clusters that streamline infrastructure provisioning, workload orchestration, and ongoing operations for tens of thousands of accelerators. Learn how topology-aware scheduling, maintenance controls, and advanced networking capabilities enable ultralow latency and maximum performance by default for demanding workloads like AI pretraining, fine-tuning, inference, and HPC.

In this session, you’ll learn how to deploy a fully-functional Retrieval-Augmented Generation (RAG) application to Google Cloud using open-source tools and models from Ray, HuggingFace, and LangChain. You’ll learn how to augment it with your own data using Ray on Google Kubernetes Engine (GKE) and Cloud SQL’s pgvector extension, deploy any model from HuggingFace to GKE, and rapidly develop your LangChain application on Cloud Run. After the session, you’ll be able to deploy your own RAG application and customize it to your needs.

Click the blue “Learn more” button above to tap into special offers designed to help you implement what you are learning at Google Cloud Next 25.