Keynote: Scaling Intelligence: Infrastructure as the Foundation
summary
Walks through concrete infrastructure pain points and the engineering responses: why fine-tuning pipelines break at scale and how data/compute orchestration shortens model-update cycles, what a production serving stack actually shards and caches to meet 99th-percentile latency, and how a Meta custom silicon effort ties topology to kernel optimizations. Covers benchmarked tradeoffs between throughput and latency, the tooling used to move checkpoints into production, and early lessons from deployed prototypes.
AI-written summary · about
related talks
10mKeynote: Enabling AI with Cloud Native
12mKeynote: AI Infrastructure
11mEngineering Scalable ML Infrastructure for Modern Apps
20mKeynote Welcome
34mKeynote: Building for AI at Scale: Infra, Integrity, & Innovation
29mLive Keynote
16mKeynote: An Open-Source Stack for AI Compute: Kubernetes + Ray + PyTorch + vLLM
30mKeynote Panel: Hardwares & Accelerators