Scaling Multimodal Data and Reinforcement Learning with Ray

AI workloads are growing in complexity and require increasing scale of both data and compute, as well as significant heterogeneity across models, data types, and hardware accelerators. As a result, the software stack for running compute-intensive AI workloads is fragmented and rapidly evolving. However, within the fragmented landscape, common patterns are beginning to emerge. This talk describes a popular software stack combining Kubernetes, Ray, PyTorch, and vLLM designed to support a variety of emerging AI workloads including multimodal data processing and reinforcement learning. It describes the role of each of these frameworks and how they operate together.