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

• Posted 1 hour ago by charles_irl
• 21 points