Leon Gao

Jiaqi Zhai, Lucy Liao, Xing Liu et al.

Large-scale recommendation systems are characterized by their reliance on high cardinality, heterogeneous features and the need to handle tens of billions of user actions on a daily basis. Despite being trained on huge volume of data with thousands of features, most Deep Learning Recommendation Models (DLRMs) in industry fail to scale with compute. Inspired by success achieved by Transformers in language and vision domains, we revisit fundamental design choices in recommendation systems. We reformulate recommendation problems as sequential transduction tasks within a generative modeling framework ("Generative Recommenders"), and propose a new architecture, HSTU, designed for high cardinality, non-stationary streaming recommendation data. HSTU outperforms baselines over synthetic and public datasets by up to 65.8% in NDCG, and is 5.3x to 15.2x faster than FlashAttention2-based Transformers on 8192 length sequences. HSTU-based Generative Recommenders, with 1.5 trillion parameters, improve metrics in online A/B tests by 12.4% and have been deployed on multiple surfaces of a large internet platform with billions of users. More importantly, the model quality of Generative Recommenders empirically scales as a power-law of training compute across three orders of magnitude, up to GPT-3/LLaMa-2 scale, which reduces carbon footprint needed for future model developments, and further paves the way for the first foundational models in recommendations.

Juhee Hong, Meng Liu, Shengzhi Wang et al.

Large-scale recommendations commonly adopt a multi-stage cascading ranking system paradigm to balance effectiveness and efficiency. Early Stage Ranking (ESR) systems utilize the "user-item decoupling" approach, where independently learned user and item representations are only combined at the final layer. While efficient, this design is limited in effectiveness, as it struggles to capture fine-grained user-item affinities and cross-signals. To address these, we propose the Generative Early Stage Ranking (GESR) paradigm, introducing the Mixture of Attention (MoA) module which leverages diverse attention mechanisms to bridge the effectiveness gap: the Hard Matching Attention (HMA) module encodes explicit cross-signals by computing raw match counts between user and item features; the Target-Aware Self Attention module generates target-aware user representations conditioned on the item, enabling more personalized learning; and the Cross Attention modules facilitate early and more enriched interactions between user-item features. MoA's specialized attention encodings are further refined in the final layer through a Multi-Logit Parameterized Gating (MLPG) module, which integrates the newly learned embeddings via gating and produces secondary logits that are fused with the primary logit. To address the efficiency and latency challenges, we have introduced a comprehensive suite of optimization techniques. These span from custom kernels that maximize the capabilities of the latest hardware to efficient serving solutions powered by caching mechanisms. The proposed GESR paradigm has shown substantial improvements in topline metrics, engagement, and consumption tasks, as validated by both offline and online experiments. To the best of our knowledge, this marks the first successful deployment of full target-aware attention sequence modeling within an ESR stage at such a scale.

Liang Guo, Wei Li, Lucy Liao et al.

Deep Learning Recommendation Models (DLRMs) represent one of the largest machine learning applications on the planet. Industry-scale DLRMs are trained with petabytes of recommendation data to serve billions of users every day. To utilize the rich user signals in the long user history, DLRMs have been scaled up to unprecedented complexity, up to trillions of floating-point operations (TFLOPs) per example. This scale, coupled with the huge amount of training data, necessitates new storage and training algorithms to efficiently improve the quality of these complex recommendation systems. In this paper, we present a Request-Only Optimizations (ROO) training and modeling paradigm. ROO simultaneously improves the storage and training efficiency as well as the model quality of recommendation systems. We holistically approach this challenge through co-designing data (i.e., request-only data), infrastructure (i.e., request-only based data processing pipeline), and model architecture (i.e., request-only neural architectures). Our ROO training and modeling paradigm treats a user request as a unit of the training data. Compared with the established practice of treating a user impression as a unit, our new design achieves native feature deduplication in data logging, consequently saving data storage. Second, by de-duplicating computations and communications across multiple impressions in a request, this new paradigm enables highly scaled-up neural network architectures to better capture user interest signals, such as Generative Recommenders (GRs) and other request-only friendly architectures.

Dheevatsa Mudigere, Yuchen Hao, Jianyu Huang et al.

Deep learning recommendation models (DLRMs) are used across many business-critical services at Facebook and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper we discuss the SW/HW co-designed solution for high-performance distributed training of large-scale DLRMs. We introduce a high-performance scalable software stack based on PyTorch and pair it with the new evolution of Zion platform, namely ZionEX. We demonstrate the capability to train very large DLRMs with up to 12 Trillion parameters and show that we can attain 40X speedup in terms of time to solution over previous systems. We achieve this by (i) designing the ZionEX platform with dedicated scale-out network, provisioned with high bandwidth, optimal topology and efficient transport (ii) implementing an optimized PyTorch-based training stack supporting both model and data parallelism (iii) developing sharding algorithms capable of hierarchical partitioning of the embedding tables along row, column dimensions and load balancing them across multiple workers; (iv) adding high-performance core operators while retaining flexibility to support optimizers with fully deterministic updates (v) leveraging reduced precision communications, multi-level memory hierarchy (HBM+DDR+SSD) and pipelining. Furthermore, we develop and briefly comment on distributed data ingestion and other supporting services that are required for the robust and efficient end-to-end training in production environments.