●builderYou can use this approach to reduce inference overhead in large-scale recommendation or advertisement layouts without sacrificing sequence dependency.
●researcherThis method provides a way to integrate sequential context into non-autoregressive architectures.