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

78 points

by charles_irl

10 hours ago

18 comments