Xing Liu

Hiwot Tadese Kassa, Paul Johnson, Jason Akers et al.

Recommendation models are very large, requiring terabytes (TB) of memory during training. In pursuit of better quality, the model size and complexity grow over time, which requires additional training data to avoid overfitting. This model growth demands a large number of resources in data centers. Hence, training efficiency is becoming considerably more important to keep the data center power demand manageable. In Deep Learning Recommendation Models (DLRM), sparse features capturing categorical inputs through embedding tables are the major contributors to model size and require high memory bandwidth. In this paper, we study the bandwidth requirement and locality of embedding tables in real-world deployed models. We observe that the bandwidth requirement is not uniform across different tables and that embedding tables show high temporal locality. We then design MTrainS, which leverages heterogeneous memory, including byte and block addressable Storage Class Memory for DLRM hierarchically. MTrainS allows for higher memory capacity per node and increases training efficiency by lowering the need to scale out to multiple hosts in memory capacity bound use cases. By optimizing the platform memory hierarchy, we reduce the number of nodes for training by 4-8X, saving power and cost of training while meeting our target training performance.

Jiaqi Zhai, Zhaojie Gong, Yueming Wang et al.

Retrieval finds a small number of relevant candidates from a large corpus for information retrieval and recommendation applications. A key component of retrieval is to model (user, item) similarity, which is commonly represented as the dot product of two learned embeddings. This formulation permits efficient inference, commonly known as Maximum Inner Product Search (MIPS). Despite its popularity, dot products cannot capture complex user-item interactions, which are multifaceted and likely high rank. We hence examine non-dot-product retrieval settings on accelerators, and propose \textit{mixture of logits} (MoL), which models (user, item) similarity as an adaptive composition of elementary similarity functions. This new formulation is expressive, capable of modeling high rank (user, item) interactions, and further generalizes to the long tail. When combined with a hierarchical retrieval strategy, \textit{h-indexer}, we are able to scale up MoL to 100M corpus on a single GPU with latency comparable to MIPS baselines. On public datasets, our approach leads to uplifts of up to 77.3\% in hit rate (HR). Experiments on a large recommendation surface at Meta showed strong metric gains and reduced popularity bias, validating the proposed approach's performance and improved generalization.

Rengan Xu, Junjie Yang, Yifan Xu et al.

The integration of hardware accelerators has significantly advanced the capabilities of modern recommendation systems, enabling the exploration of complex ranking paradigms previously deemed impractical. However, the GPU-based computational costs present substantial challenges. In this paper, we demonstrate our development of an efficiency-driven approach to explore these paradigms, moving beyond traditional reliance on native PyTorch modules. We address the specific challenges posed by ranking models' dependence on categorical features, which vary in length and complicate GPU utilization. We introduce Jagged Feature Interaction Kernels, a novel method designed to extract fine-grained insights from long categorical features through efficient handling of dynamically sized tensors. We further enhance the performance of attention mechanisms by integrating Jagged tensors with Flash Attention. Our novel Jagged Flash Attention achieves up to 9x speedup and 22x memory reduction compared to dense attention. Notably, it also outperforms dense flash attention, with up to 3x speedup and 53% more memory efficiency. In production models, we observe 10% QPS improvement and 18% memory savings, enabling us to scale our recommendation systems with longer features and more complex architectures.

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.

Yue Dong, Han Li, Shen Li et al.

Large-scale recommendation systems are pivotal to process an immense volume of daily user interactions, requiring the effective modeling of high cardinality and heterogeneous features to ensure accurate predictions. In prior work, we introduced Hierarchical Sequential Transducers (HSTU), an attention-based architecture for modeling high cardinality, non-stationary streaming recommendation data, providing good scaling law in the generative recommender framework (GR). Recent studies and experiments demonstrate that attending to longer user history sequences yields significant metric improvements. However, scaling sequence length is activation-heavy, necessitating parallelism solutions to effectively shard activation memory. In transformer-based LLMs, context parallelism (CP) is a commonly used technique that distributes computation along the sequence-length dimension across multiple GPUs, effectively reducing memory usage from attention activations. In contrast, production ranking models typically utilize jagged input tensors to represent user interaction features, introducing unique CP implementation challenges. In this work, we introduce context parallelism with jagged tensor support for HSTU attention, establishing foundational capabilities for scaling up sequence dimensions. Our approach enables a 5.3x increase in supported user interaction sequence length, while achieving a 1.55x scaling factor when combined with Distributed Data Parallelism (DDP).