Body
AI inference imposes substantial compute costs, which are not sustainable at certain stages of the AI development process. Several techniques exist to reduce compute requirements and associated costs. Here we detail decoupled pipeline stages.