Ziyan Gong

Haoran Ding, Wenlin Zhao, Yuchen Jiang et al.

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.

Hui Lu, Zheng Chai, Shipeng Bai et al.

Driven by scaling laws, recommender systems increasingly rely on large-scale models to capture complex feature interactions and user behaviors, but this trend also leads to prohibitive training and inference costs. While long-sequence models(e.g., LONGER) can reuse user-side computation through KV caching, such reuse is difficult in dense feature interaction architectures(e.g., RankMixer), where user and group (candidate item) features are deeply entangled across layers. In this work, we propose User-Group Separation (UG-Sep), a novel framework that enables reusable user-side computation in dense interaction models for the first time. UG-Sep introduces a masking mechanism that explicitly disentangles user-side and item-side information flows within token-mixing layers, ensuring that a subset of tokens to preserve purely user-side representations across layers. This design enables corresponding token computations to be reused across multiple samples, significantly reducing redundant inference cost. To compensate for potential expressiveness loss induced by masking, we further propose an Information Compensation strategy that adaptively reconstructs suppressed user-item interactions. Moreover, as UG-Sep substantially reduces user-side FLOPs and exposes memory-bound components, we incorporate W8A16 (8-bit weight, 16-bit activation) weight-only quantization to alleviate memory bandwidth bottlenecks and achieve additional acceleration. We conduct extensive offline evaluations and large-scale online A/B experiments at ByteDance, demonstrating that UG-Sep reduces inference latency by up to 20 percent without degrading online user experience or commercial metrics across multiple business scenarios, including feed recommendation and advertising systems.

Wei Zeng, Xiaozhe Ren, Teng Su et al.

Large-scale Pretrained Language Models (PLMs) have become the new paradigm for Natural Language Processing (NLP). PLMs with hundreds of billions parameters such as GPT-3 have demonstrated strong performances on natural language understanding and generation with \textit{few-shot in-context} learning. In this work, we present our practice on training large-scale autoregressive language models named PanGu-$α$, with up to 200 billion parameters. PanGu-$α$ is developed under the MindSpore and trained on a cluster of 2048 Ascend 910 AI processors. The training parallelism strategy is implemented based on MindSpore Auto-parallel, which composes five parallelism dimensions to scale the training task to 2048 processors efficiently, including data parallelism, op-level model parallelism, pipeline model parallelism, optimizer model parallelism and rematerialization. To enhance the generalization ability of PanGu-$α$, we collect 1.1TB high-quality Chinese data from a wide range of domains to pretrain the model. We empirically test the generation ability of PanGu-$α$ in various scenarios including text summarization, question answering, dialogue generation, etc. Moreover, we investigate the effect of model scales on the few-shot performances across a broad range of Chinese NLP tasks. The experimental results demonstrate the superior capabilities of PanGu-$α$ in performing various tasks under few-shot or zero-shot settings.