Teacher Retains Full Tokens, Student Merges Efficiently: TM20K for E-Commerce Sequence Modeling in Ad Recommendation

Benefiting from ultra-long behavior sequence modeling, existing recommender systems bring users a better experience via simultaneously considering their long-term and short-term interests. Nevertheless, extended sequence lengths introduce substantial burdens on training efficiency and serving throughput. Prior approaches typically utilize search-based or cluster-based compression on ultra-long sequences at the cost of fine-grained information, or rely on various lightweight target attention structures incapable of sufficient sequential feature extraction. In this paper, we balance the effectiveness and efficiency for ultra-long sequence modeling via full transformer modeling accompanied with a two-stage knowledge distillation framework. First, both teacher and student models take the full attention mechanism rather than pure target-sequence attention for effective sequence scaling. For student models, we propose several simple yet well-motivated token merge approaches, significantly compressing the sequence length while maintaining an acceptable performance. Then, a one-time teacher is heavily trained with full sequence tokens, further boosting the performance of student models via knowledge distillation. The proposed paradigm named TM20K has been successfully deployed in ByteDance's e-commerce advertising recommender system that extends the e-commerce sequence length to 20K, delivering substantial improvements in key business metrics (e.g., ADSS +1.036\%) while keeping the training and serving cost nearly the same as the online state-of-the-art model (e.g., serving latency only +5.6\%).

Rec-Distill: An Industrial Distillation Pipeline for Large-Scale Recommendation Models

Large recommendation models have demonstrated substantial potential gains under scaling laws, yet these gains are difficult to realize in industrial recommendation systems because real-world deployment requires lightweight models with strict serving efficiency and latency guarantees. This creates a fundamental gap between offline model scaling and online deployment. In this work, we present Rec-Distill, an industrial distillation pipeline that transfers the performance gains of large-scale recommendation modeling to efficient serving models. Rec-Distill combines large-teacher scaling with student-side transfer optimization through decoupled training, black-box distillation, debiasing mechanism, and a hybrid batch-streaming pipeline for dynamic recommendation environments. Across multiple recommendation and advertising scenarios on real-world platforms, our framework scales teacher models up to 24B dense parameters and 20K behavior sequence length, while enabling lightweight students to recover a substantial portion of teacher gains, with distillation transferability exceeding 60% in the best setting. Extensive offline and online experiments further show that these transferred gains consistently translate into measurable business improvements under industrial constraints. These results demonstrate that Rec-Distill provides a practical framework for distilling large-scale recommendation models into deployable, cost-efficient serving systems, while also establishing a reliable path toward scaling recommendation models to even larger regimes in the future.

IAT: Instance-As-Token Compression for Historical User Sequence Modeling in Industrial Recommender Systems

Although sophisticated sequence modeling paradigms have achieved remarkable success in recommender systems, the information capacity of hand-crafted sequential features constrains the performance upper bound. To better enhance user experience by encoding historical interaction patterns, this paper presents a novel two-stage sequence modeling framework termed Instance-As-Token (IAT). The first stage of IAT compresses all features of each historical interaction instance into a unified instance embedding, which encodes the interaction characteristics in a compact yet informative token. Both temporal-order and user-order compression schemes are proposed, with the latter better aligning with the demands of downstream sequence modeling. The second stage involves the downstream task fetching fixed-length compressed instance tokens via timestamps and adopting standard sequence modeling approaches to learn long-range preferences patterns. Extensive experiments demonstrate that IAT significantly outperforms state-of-the-art methods and exhibits superior in-domain and cross-domain transferability. IAT has been successfully deployed in real-world industrial recommender systems, including e-commerce advertising, shopping mall marketing, and live-streaming e-commerce, delivering substantial improvements in key business metrics.