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

 (modal.com)

89 points | by charles_irl 3 days ago

18 comments