04:00
2026-09-28
machinebrief.com
machine-learning
KuaFu: Compressing Long User Behavior into Understanding at Billion Scale
Tencent researchers presented KuaFu, a unified behavior-compression layer that compresses each user behavior item into 2-4 tokens of width 128-256, cutting per-item cache from 10 KB to 0.5 KB, accordi…