Building Frontier AI Infra: SGLang and Miles
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Describes SGLang and Miles as infrastructure for serving and post-training of frontier AI models, focusing on agent-style workloads. Covers specific technologies: Unified Radix Cache and HiCache for cache reuse, parallelism strategies for long-context serving, HiSparse for sparsity, and how Miles uses SGLang for efficient, scalable reinforcement learning with high-performance rollouts and fully asynchronous training.
AI-written summary · about
related talks
32mEnd-to-End Optimizing Multi-Turn RL and High-Performance Inference in Agents with...
36mProduction-Ready AI Agents: From LLMs to Small Language Models
35mThe Fight for the Future of Artificial Intelligence
29mMiles: Enterprise-Facing Agentic RL Framework
44mIntroduction to LLM serving with SGLang
28mOptimizing Long-Tail and MoE Challenges in Reinforcement Learning with SGLang
20mSGLang: An Efficient Open-Source Framework for Large-Scale LLM Serving
40mSurviving the Agentic Hype with Small Language Models