●builderYou can significantly reduce input costs and training time when fine-tuning LLMs for recommendation systems.
●researcherThis architecture provides an efficient way to handle the asymmetry between long interaction histories and short target identifiers.