Dynamo Open-Sources Datacenter-Scale Distributed Inference Orchestration

Summary

Dynamo is an open-source datacenter-scale orchestration layer that coordinates SGLang, TensorRT-LLM, and vLLM across GPU clusters. It provides disaggregated prefill and decode, KV-aware routing, tiered KV caching, weight streaming, SLA-based autoscaling, topology-aware scheduling, and fault tolerance. Dynamo 1.0 also introduces beta zero-config deployment, agentic inference controls, multimodal serving, video-generation support, and Kubernetes Gateway API integration.