Cutting inference cold starts by 40x with LP, FUSE, C/R, and CUDA-checkpoint

charles_irl 18 May 2026, 17:56 UTC (3 weeks ago) Hacker News — Top Stories free api