FuXi-β: Towards a Lightweight and Fast Large-Scale Generative Recommendation Model

• URL: http://arxiv.org/abs/2508.10615v1
• Date: Thu, 14 Aug 2025 13:12:29 GMT
• Title: FuXi-β: Towards a Lightweight and Fast Large-Scale Generative Recommendation Model
• Authors: Yufei Ye, Wei Guo, Hao Wang, Hong Zhu, Yuyang Ye, Yong Liu, Huifeng Guo, Ruiming Tang, Defu Lian, Enhong Chen,
• Abstract summary: We propose a new framework for Transformer-like recommendation models.<n>FuXi-$beta$ outperforms previous state-of-the-art models and achieves significant acceleration.<n>Our code is available in a public repository: https://github.com/USTC-StarTeam/FuXi-beta$.
• Score: 87.38823851271758
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Scaling laws for autoregressive generative recommenders reveal potential for larger, more versatile systems but mean greater latency and training costs. To accelerate training and inference, we investigated the recent generative recommendation models HSTU and FuXi-$\alpha$, identifying two efficiency bottlenecks: the indexing operations in relative temporal attention bias and the computation of the query-key attention map. Additionally, we observed that relative attention bias in self-attention mechanisms can also serve as attention maps. Previous works like Synthesizer have shown that alternative forms of attention maps can achieve similar performance, naturally raising the question of whether some attention maps are redundant. Through empirical experiments, we discovered that using the query-key attention map might degrade the model's performance in recommendation tasks. To address these bottlenecks, we propose a new framework applicable to Transformer-like recommendation models. On one hand, we introduce Functional Relative Attention Bias, which avoids the time-consuming operations of the original relative attention bias, thereby accelerating the process. On the other hand, we remove the query-key attention map from the original self-attention layer and design a new Attention-Free Token Mixer module. Furthermore, by applying this framework to FuXi-$\alpha$, we introduce a new model, FuXi-$\beta$. Experiments across multiple datasets demonstrate that FuXi-$\beta$ outperforms previous state-of-the-art models and achieves significant acceleration compared to FuXi-$\alpha$, while also adhering to the scaling law. Notably, FuXi-$\beta$ shows an improvement of 27% to 47% in the NDCG@10 metric on large-scale industrial datasets compared to FuXi-$\alpha$. Our code is available in a public repository: https://github.com/USTC-StarTeam/FuXi-beta

Related papers

• Light Forcing: Accelerating Autoregressive Video Diffusion via Sparse Attention [28.598033369607723]
  textscLight Forcing is a textitfirst sparse attention solution tailored for AR video generation models.<n>It incorporates a textitChunk-Aware Growth mechanism to quantitatively estimate the contribution of each chunk.<n>We also introduce a textit Sparse Attention to capture informative historical and local context in a coarse-to-fine manner.
  arXiv  Detail & Related papers  (2026-02-04T17:41:53Z)
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free [81.65559031466452]
  We conduct experiments to investigate gating-augmented softmax attention variants.<n>We find that a simple modification-applying a head-specific sigmoid gate after the Scaled Dot-Product Attention (SDPA)-consistently improves performance.
  arXiv  Detail & Related papers  (2025-05-10T17:15:49Z)
• FuXi-$α$: Scaling Recommendation Model with Feature Interaction Enhanced Transformer [81.12174905444229]
  Recent advancements have shown that expanding sequential recommendation models to large-scale recommendation models can be an effective strategy.<n>We propose a new model called FuXi-$alpha$ to address these issues.<n>Our model outperforms existing models, with its performance continuously improving as the model size increases.
  arXiv  Detail & Related papers  (2025-02-05T09:46:54Z)
• More Expressive Attention with Negative Weights [36.40344438470477]
  We propose a novel attention mechanism, named Cog Attention, that enables attention weights to be negative for enhanced expressiveness.<n>Our approach suggests a promising research direction for rethinking and breaking the entrenched constraints of traditional softmax attention.
  arXiv  Detail & Related papers  (2024-11-11T17:56:28Z)
• Long-Sequence Recommendation Models Need Decoupled Embeddings [49.410906935283585]
  We identify and characterize a neglected deficiency in existing long-sequence recommendation models.<n>A single set of embeddings struggles with learning both attention and representation, leading to interference between these two processes.<n>We propose the Decoupled Attention and Representation Embeddings (DARE) model, where two distinct embedding tables are learned separately to fully decouple attention and representation.
  arXiv  Detail & Related papers  (2024-10-03T15:45:15Z)
• Simple linear attention language models balance the recall-throughput tradeoff [60.06020449520365]
  We propose BASED, a simple architecture combining linear and sliding window attention.<n>We train language models up to 1.3b parameters and show that BASED matches the strongest sub-quadratic models in perplexity and outperforms them on real-world recall-intensive tasks by 6.22 accuracy points.
  arXiv  Detail & Related papers  (2024-02-28T19:28:27Z)
• SparseBERT: Rethinking the Importance Analysis in Self-attention [107.68072039537311]
  Transformer-based models are popular for natural language processing (NLP) tasks due to its powerful capacity. Attention map visualization of a pre-trained model is one direct method for understanding self-attention mechanism. We propose a Differentiable Attention Mask (DAM) algorithm, which can be also applied in guidance of SparseBERT design.
  arXiv  Detail & Related papers  (2021-02-25T14:13:44Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.