Cut WAN family time-to-ready 220-304s -> 33-51s: AOT executable cache, zero-exec warmup, converted-weights cache#443

Open

syhuang22 wants to merge 1 commit into

AI-Hypercomputer:mainfrom

syhuang22:vace-load-compile

Commits on Jul 10, 2026

Cut WAN family compile 321.6s->34.8s and load waste: per-shape AOT executable cache, zero-exec warmup, VACE/animate cast+staged-put reuse, VAE decode AOT, prompt embeds cache, threaded 2.1 TE warm
syhuang22
committed