OneRec Technical Report

Recommender systems have been widely used in various large-scale user-oriented platforms for many years. However, compared to the rapid developments in the AI community, recommendation systems have not achieved a breakthrough in recent years. For instance, they still rely on a multi-stage cascaded architecture rather than an end-to-end approach, leading to computational fragmentation and optimization inconsistencies, and hindering the effective application of key breakthrough technologies from the AI community in recommendation scenarios. To address these issues, we propose OneRec, which reshapes the recommendation system through an end-to-end generative approach and achieves promising results. Firstly, we have enhanced the computational FLOPs of the current recommendation model by 10 $\times$ and have identified the scaling laws for recommendations within certain boundaries. Secondly, reinforcement learning techniques, previously difficult to apply for optimizing recommendations, show significant potential in this framework. Lastly, through infrastructure optimizations, we have achieved 23.7% and 28.8% Model FLOPs Utilization (MFU) on flagship GPUs during training and inference, respectively, aligning closely with the LLM community. This architecture significantly reduces communication and storage overhead, resulting in operating expense that is only 10.6% of traditional recommendation pipelines. Deployed in Kuaishou/Kuaishou Lite APP, it handles 25% of total queries per second, enhancing overall App Stay Time by 0.54% and 1.24%, respectively. Additionally, we have observed significant increases in metrics such as 7-day Lifetime, which is a crucial indicator of recommendation experience. We also provide practical lessons and insights derived from developing, optimizing, and maintaining a production-scale recommendation system with significant real-world impact.

E-Tree Learning: A Novel Decentralized Model Learning Framework for Edge AI

Traditionally, AI models are trained on the central cloud with data collected from end devices. This leads to high communication cost, long response time and privacy concerns. Recently Edge empowered AI, namely Edge AI, has been proposed to support AI model learning and deployment at the network edge closer to the data sources. Existing research including federated learning adopts a centralized architecture for model learning where a central server aggregates the model updates from the clients/workers. The centralized architecture has drawbacks such as performance bottleneck, poor scalability and single point of failure. In this paper, we propose a novel decentralized model learning approach, namely E-Tree, which makes use of a well-designed tree structure imposed on the edge devices. The tree structure and the locations and orders of aggregation on the tree are optimally designed to improve the training convergency and model accuracy. In particular, we design an efficient device clustering algorithm, named by KMA, for E-Tree by taking into account the data distribution on the devices as well as the the network distance. Evaluation results show E-Tree significantly outperforms the benchmark approaches such as federated learning and Gossip learning under NonIID data in terms of model accuracy and convergency.