Token Factory: Efficiently Integrating Diverse Signals into Large Recommendation Models

The article introduces "Token Factory," a framework aimed at efficiently integrating traditional signals into Large Recommendation Models (LRMs) by converting these signals into "soft tokens." This method addresses the challenges of long prompts and high computational costs associated with conventional approaches, enhancing model performance while reducing memory usage. Experimental results demonstrate its effectiveness in a production-scale recommendation context, making it a significant advancement for practitioners working with LRM architectures.