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

by charles_irl | View on Hacker News