OneLatent: Latent Reasoning for Efficient Foundation Recommendation Models
AuthorsHao Jiang, Peiru Du, Pengfei Yao, Mengting Li, Siyuan Lou, Kuo Cai, Sheng Yu, Qiang Luo, Jian Liang, Ruiming Tang, Fei Pan, Peng Jiang, Wenwu Ou
AffiliationsKuaishou Technology
OneLatent replaces lengthy reasoning traces with compact learned tokens to make foundation recommendation models faster and more effective.
Key results
Improvement on the industrial dataset over the matched explicit-reasoning Think baseline.
Improvement on the industrial dataset over the matched direct-prediction baseline.
Times the throughput of explicit CoT inference.
Reduction in average generated reasoning length.
Estimated lift in the Kuaishou local-services advertising A/B test.
What the paper found
OneLatent replaces lengthy chain-of-thought generation in foundation recommendation models with three fixed, learnable latent tokens, so the model can reason internally and then predict an item’s semantic ID. Built on Qwen3.5-0.8B, it is trained through staged alignment using Multi-View Adaptive CoT: teacher-generated reasoning covers intent exploration, positive-item evaluation, and interaction attribution, with reasoning depth adapted to each case. On the industrial Kuaishou dataset, OneLatent improves SID@64 by 17.44% over the matched Think baseline and 9.33% over No-Think, while delivering 17.49× the throughput of explicit CoT. Adaptive CoT also reduces average reasoning length by 35% without materially losing semantic content on complex cases. In a Kuaishou local-services advertising A/B test, asynchronous serving of cached recommendations produced an estimated 9.6% revenue lift over strong online baselines including OneRec and OneReason. The results show how explicit reasoning can supervise a compact inference-time interface without requiring the recommender to generate a rationale for every prediction.
Original abstract
Large language models (LLMs) have demonstrated strong reasoning capabilities, motivating their use as the backbone of foundation recommendation models (FRMs). Existing methods enhance recommendations through explicit Chain-of-Thought (CoT) reasoning under a Think-then-Answer paradigm. However, explicit CoT incurs substantial inference overhead by generating lengthy reasoning traces and relies on manually designed templates that struggle to capture diverse, dynamic user interests. We propose OneLatent, an efficient latent reasoning framework that compresses explicit reasoning traces into several learnable latent tokens, enabling Latent-Reason-then-Answer inference without generating verbose traces. OneLatent first introduces Multi-View Adaptive CoT (MV-ACoT), which creates diverse, high-quality teacher-generated supervision by exploring user interests from multiple perspectives and automatically adapting reasoning complexity to each instance. Building on pretrained FRMs, it then uses a three-stage latent-token alignment paradigm to progressively internalize CoT traces into learnable latent tokens. Finally, a multistage curriculum-based post-training strategy activates latent-token reasoning for downstream recommendation tasks. Experiments on an industrial-scale Kuaishou dataset and the public Kuaishou LLM-Rec benchmark show that OneLatent consistently outperforms explicit CoT-based methods and traditional baselines. Compared with the Think and No-Think variants of FRMs, OneLatent improves SID@64 by 17.44% and 9.33%, respectively, while achieving over 17x higher online inference throughput. We further develop a production serving system for scalable, real-time FRM inference. An online A/B test in Kuaishou's local-services advertising scenario shows that deploying OneLatent with this system yields an estimated 9.6% revenue lift over strong online baselines, including OneRec and OneReason.
Read the original paperMore in AI Reasoning
Browse all 48 papers →Amortizing Sequential Experimental Design in Reasoning LLMs
Anita Kriz, Antonin Berthon, Nicolás Astorga, Mihaela van der Schaar
The paper teaches LLMs to ask more informative questions, making sequential information gathering cheaper and more transferable.
CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation
Zinan Tang, Zhuoshi Pan, Kairui Chen, Ze Yan, Shaomian Zheng, Yukun Zhang, Gang Wang, Biqing Huang
CoTranslator turns cheaper answer-conditioned reasoning traces into useful student training data while approaching the performance of more expensive rejection sampling.
Inducing Process Supervision from Outcome-Only Reinforcement Learning
Shengda Fan, Xin Cong, Zhong Zhang, Haotian Chen, Yankai Lin
TIPS trains models to judge their own reasoning steps using only outcome feedback, avoiding the need for costly step-by-step annotations.