让小模型高效推荐：通过多层出口融合提升紧凑大模型性能
Empowering Compact LLMs with Fusion of Layer-wise Exits for Recommendation
- 在Transformer多层插入预测头，动态融合各层输出结果
- 在三个真实数据集上超越现有方法，1.7B/3B模型仍保持高精度
- 适合资源受限场景下的高效推荐系统研发
基于大语言模型的推荐系统（LLM-RS）虽能力突出，但计算开销大，难以落地。紧凑型LLM虽更实用，但容量有限，常需依赖推理或知识蒸馏，导致延迟上升或依赖大模型。结合自回归生成的方法还面临严重可扩展性瓶颈。相比之下，判别式LLM-RS通过嵌入相似度实现全库高效排序，但紧凑主干网络表达力与结构适应性不足。本文提出FLEXRec框架，在多个Transformer层插入预测头（即出口），并自适应融合其得分分布。采用自适应连续路由（AC-Router）动态选择每条用户序列的出口数量与身份，引入新型目标-阈值铰链损失以调控路由稀疏性。在Qwen 3 1.7B和Llama 3.2 3B模型下，三个真实数据集实验表明，FLEXRec在保持高度效率的同时，达到当前最优推荐准确率。
原文摘要 · Abstract (English)
Large language model-based recommender systems (LLM-RSs) have demonstrated remarkable capabilities, but are computationally unsustainable for many real-world applications. Compact LLMs offer a practical alternative, yet their reduced capacity often requires reasoning or knowledge distillation methods that increase latency or depend on larger models. Combined with autoregressive generation, these approaches face severe scalability bottlenecks. In contrast, discriminative LLM-RSs enable efficient full-corpus ranking through embedding similarity, but compact backbones remain limited in expressiveness and structural adaptivity. We propose the Fusion of Layer-wise Exits for Sequential Recommendation (FLEXRec), a discriminative framework that enhances compact LLMs while retaining scalable full-corpus ranking. FLEXRec inserts prediction heads (i.e., exits) at multiple transformer layers and adaptively fuses their score distributions. An adaptive continuous router (AC-Router) dynamically selects both the number and identity of exits for each user sequence, while a novel target-k hinge loss regulates routing sparsity. Experiments on three real-world datasets with Qwen 3 1.7B and Llama 3.2 3B show that FLEXRec achieves state-of-the-art accuracy among competing methods while remaining highly efficient. Code: https://github.com/xurong-liang/FLEXRec
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。