OneTrans-GSR: A Generative Slate Ranking Paradigm Unifying Reranking and Mix-Ranking with One Autoregressive Transformer
AuthorsFeng Cheng, Yu-Hsuan Huang, Fengyi Wang, Chaofan Zhou, Zhennan Zhu, Huan He, Bo Zhang, Ruoqi Zhang, Zhanfeng Mo, Ziwei Liu, Lanyi Li, Yifeng Pan, Junhua Long, Ziqiu Zhou, Yue Lin, Shaowei Liu
AffiliationsByteDance
OneTrans-GSR uses a single learned autoregressive policy to assemble recommendation slates, reporting improved business outcomes and lower latency in online tests.
Key results
Production A/B improvement versus the Generator–Evaluator system.
Production A/B improvement versus the Generator–Evaluator system.
Production A/B improvement versus the Generator–Evaluator system.
Reduction across the reranking and mix-ranking stages.
What the paper found
ByteDance’s OneTrans-GSR replaces separate reranking and mix-ranking components with one autoregressive OneTrans policy that selects and orders product cards, videos, livestreams, and ads from each request’s candidate pool. A slate-level reward model evaluates complete slates during offline training, while constrained decoding enforces business rules at serving time. The new GMR-PO method combines normalized click, order, GMV, and long-term-value rewards using a weighted geometric mean, then incorporates soft business preferences through separately standardized advantages. In a randomized production A/B test against the existing Generator–Evaluator system, OneTrans-GSR increased GMV per user by 2.63%, advertiser value by 2.06%, and repurchase rate by 2.03%, while reducing end-to-end reranking and mix-ranking latency by 13%. Offline tests also found that slate quality improved as reward-model, policy, and sampling compute increased, supporting a learned approach that can scale beyond hand-engineered slate assembly.
Original abstract
Industrial recommender systems allocate substantial computation to pointwise candidate scoring, yet separate reranking and mix-ranking stages still assemble the final exposure slate using hand-crafted rules and specialized low-capacity models. Mix-ranking combines product cards, short videos, livestreams, and advertising cards supplied by distinct upstream business channels. Although these rules and models encode valuable domain knowledge, they are inherently limited in leveraging additional data and computation. They constrain the combinatorial space of candidate slates and determine objective trade-offs through manual tuning. To enable scalable and effective slate generation, we reformulate reranking and mix-ranking as a unified learning problem whose performance can improve with computation. Specifically, we instantiate this formulation across the reranking and mix-ranking stages and propose OneTrans-GSR, a single decoder-only policy that extends our deployed OneTrans backbone to generate the exposure slate end-to-end. During training, a reward model evaluates complete slates and guides reinforcement learning toward slate-level and long-term objectives. During decoding, business constraints are explicitly enforced. Offline evaluation shows that slate quality consistently improves as reward-model and policy compute increase. Deployed in a large-scale industrial recommendation system, OneTrans-GSR mixes heterogeneous content types within a unified slate. In online A/B tests, it improves GMV/u by 2.63%, advertiser value by 2.06%, and repurchase rate by 2.03%, while reducing end-to-end latency across the reranking and mix-ranking stages by 13%. These results mark a paradigm shift in slate generation: moving beyond human-engineered heuristics toward a general learned policy creates a scalable path for further algorithmic optimization and compute-driven improvement.
Read the original paperMore in Generative Models
Browse all 87 papers →dNaViT: Discrete Codes are General-Purpose Visual Representations
Xiaotong Li, Zhengrong Yue, Xu Huang, Jialv Zou, Minhao Jing, Wenhao Yan, Haodong Li, Hai Huang, Bo Zhao, Yan Feng, Peng Pei, Xunliang Cai
dNaViT turns images into reusable discrete codes designed to work across visual understanding, generation, and editing.
Lipschitz-Controlled Flow Matching for Robust Image Restoration
Sedjro Salomon Hotegni, Konstantin Sonntag, Paul-Christian Bürkner, Sebastian Peitz
LipsFlower aims to make flow-matching image restorations less sensitive to small input changes without sacrificing perceptual quality.
PhysLoop: Harnessing LLMs' Physics Awareness for Self-Improving Video Generation Models
Yuxuan Chen, Xiaojie Li, Hao Chen, Juncheng Ma, Kewei Zhang, Zhongjing Du, Guoqiang Wei, Fangfang Wang, Zhen Dong, Zhiyang Dou, Liming Zhao, Yan Zeng, Jianwei Zhang, Daquan Zhou
PhysLoop uses LLMs’ physics knowledge to help video generators create and repair more physically consistent videos.