Robert Nishihara, Anyscale | CUBE Conversation
summary
Anyscale runs an open source Ray runtime that companies use to scale training and serving across hundreds to thousands of GPUs, with the company ~100 people and customers like Shopify, Spotify, Netflix and OpenAI (Ray helps train models such as ChatGPT). Problems solved include poor GPU utilization, fast autoscaling, spot instance cost savings, and cutting the prototype‑to‑production gap often reported as 8–12 weeks. The product is a managed service around Ray that unifies training, fine‑tuning, serving and batch inference.
AI-written summary · about
