토스는 AI모델을 어떻게 서빙할까 | SlashLive
토스 챌린저스 · 21:06 · Watch on YouTube
Toss Serving Engineering treats model SLA as a full-stack problem: an agentic, configuration-driven workflow estimates GPU count, cost, and ROI, then deploys a mix of stock engines and in-house kernel patches so a 700...