Adaptive compute reduces latency by up to ~45%

Adaptive compute primitives now shave 30–50% off inference latency across video diffusion,...