GEMs: Breaking the Long-Sequence Barrier in Generative Recommendation with a Multi-Stream Decoder

While generative recommendations (GR) possess strong sequential reasoning capabilities, they face significant challenges when processing extremely long user behavior sequences: the high computational cost forces practical sequence lengths to be limited, preventing models from capturing users' lifelong interests; meanwhile, the inherent "recency bias" of attention mechanisms further weakens learning from long-term history. To overcome this bottleneck, we propose GEMs (Generative rEcommendation with a Multi-stream decoder), a novel and unified framework designed to break the long-sequence barrier by capturing users' lifelong interaction sequences through a multi-stream perspective. Specifically, GEMs partitions user behaviors into three temporal streams$\unicode{x2014}$Recent, Mid-term, and Lifecycle$\unicode{x2014}$and employs tailored inference schemes for each: a one-stage real-time extractor for immediate dynamics, a lightweight indexer for cross attention to balance accuracy and cost for mid-term sequences, and a two-stage offline-online compression module for lifelong modeling. These streams are integrated via a parameter-free fusion strategy to enable holistic interest representation. Extensive experiments on large-scale industrial datasets demonstrate that GEMs significantly outperforms state-of-the-art methods in recommendation accuracy. Notably, GEMs is the first lifelong GR framework successfully deployed in a high-concurrency industrial environment, achieving superior inference efficiency while processing user sequences of over 100,000 interactions.

Kelix Technical Report

Autoregressive large language models (LLMs) scale well by expressing diverse tasks as sequences of discrete natural-language tokens and training with next-token prediction, which unifies comprehension and generation under self-supervision. Extending this paradigm to multimodal data requires a shared, discrete representation across modalities. However, most vision-language models (VLMs) still rely on a hybrid interface: discrete text tokens paired with continuous Vision Transformer (ViT) features. Because supervision is largely text-driven, these models are often biased toward understanding and cannot fully leverage large-scale self-supervised learning on non-text data. Recent work has explored discrete visual tokenization to enable fully autoregressive multimodal modeling, showing promising progress toward unified understanding and generation. Yet existing discrete vision tokens frequently lose information due to limited code capacity, resulting in noticeably weaker understanding than continuous-feature VLMs. We present Kelix, a fully discrete autoregressive unified model that closes the understanding gap between discrete and continuous visual representations.

SARM: LLM-Augmented Semantic Anchor for End-to-End Live-Streaming Ranking

Large-scale live-streaming recommendation requires precise modeling of non-stationary content semantics under strict real-time serving constraints. In industrial deployment, two common approaches exhibit fundamental limitations: discrete semantic abstractions sacrifice descriptive precision through clustering, while dense multimodal embeddings are extracted independently and remain weakly aligned with ranking optimization, limiting fine-grained content-aware ranking. To address these limitations, we propose \textbf{SARM}, an end-to-end ranking architecture that integrates natural-language semantic anchors directly into ranking optimization, enabling fine-grained author representations conditioned on multimodal content. Each semantic anchor is represented as learnable text tokens jointly optimized with ranking features, allowing the model to adapt content descriptions to ranking objectives. A lightweight dual-token gated design captures domain-specific live-streaming semantics, while an asymmetric deployment strategy preserves low-latency online training and serving. Extensive offline evaluation and large-scale A/B tests show consistent improvements over production baselines. SARM is fully deployed and serves over 400 million users daily.

SMES: Towards Scalable Multi-Task Recommendation via Expert Sparsity

Industrial recommender systems typically rely on multi-task learning to estimate diverse user feedback signals and aggregate them for ranking. Recent advances in model scaling have shown promising gains in recommendation. However, naively increasing model capacity imposes prohibitive online inference costs and often yields diminishing returns for sparse tasks with skewed label distributions. This mismatch between uniform parameter scaling and heterogeneous task capacity demands poses a fundamental challenge for scalable multi-task recommendation. In this work, we investigate parameter sparsification as a principled scaling paradigm and identify two critical obstacles when applying sparse Mixture-of-Experts (MoE) to multi-task recommendation: exploded expert activation that undermines instance-level sparsity and expert load skew caused by independent task-wise routing. To address these challenges, we propose SMES, a scalable sparse MoE framework with progressive expert routing. SMES decomposes expert activation into a task-shared expert subset jointly selected across tasks and task-adaptive private experts, explicitly bounding per-instance expert execution while preserving task-specific capacity. In addition, SMES introduces a global multi-gate load-balancing regularizer that stabilizes training by regulating aggregated expert utilization across all tasks. SMES has been deployed in Kuaishou large-scale short-video services, supporting over 400 million daily active users. Extensive online experiments demonstrate stable improvements, with GAUC gain of 0.29% and a 0.31% uplift in user watch time.

OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommendation

Live-streaming recommender system serves as critical infrastructure that bridges the patterns of real-time interactions between users and authors. Similar to traditional industrial recommender systems, live-streaming recommendation also relies on cascade architectures to support large-scale concurrency. Recent advances in generative recommendation unify the multi-stage recommendation process with Transformer-based architectures, offering improved scalability and higher computational efficiency. However, the inherent complexity of live-streaming prevents the direct transfer of these methods to live-streaming scenario, where continuously evolving content, limited lifecycles, strict real-time constraints, and heterogeneous multi-objectives introduce unique challenges that invalidate static tokenization and conventional model framework. To address these issues, we propose OneLive, a dynamically unified generative recommendation framework tailored for live-streaming scenario. OneLive integrates four key components: (i) A Dynamic Tokenizer that continuously encodes evolving real-time live content fused with behavior signal through residual quantization; (ii) A Time-Aware Gated Attention mechanism that explicitly models temporal dynamics for timely decision making; (iii) An efficient decoder-only generative architecture enhanced with Sequential MTP and QK Norm for stable training and accelerated inference; (iv) A Unified Multi-Objective Alignment Framework reinforces policy optimization for personalized preferences.

TimeChat-Captioner: Scripting Multi-Scene Videos with Time-Aware and Structural Audio-Visual Captions

This paper proposes Omni Dense Captioning, a novel task designed to generate continuous, fine-grained, and structured audio-visual narratives with explicit timestamps. To ensure dense semantic coverage, we introduce a six-dimensional structural schema to create "script-like" captions, enabling readers to vividly imagine the video content scene by scene, akin to a cinematographic screenplay. To facilitate research, we construct OmniDCBench, a high-quality, human-annotated benchmark, and propose SodaM, a unified metric that evaluates time-aware detailed descriptions while mitigating scene boundary ambiguity. Furthermore, we construct a training dataset, TimeChatCap-42K, and present TimeChat-Captioner-7B, a strong baseline trained via SFT and GRPO with task-specific rewards. Extensive experiments demonstrate that TimeChat-Captioner-7B achieves state-of-the-art performance, surpassing Gemini-2.5-Pro, while its generated dense descriptions significantly boost downstream capabilities in audio-visual reasoning (DailyOmni and WorldSense) and temporal grounding (Charades-STA). All datasets, models, and code will be made publicly available at https://github.com/yaolinli/TimeChat-Captioner.

QARM V2: Quantitative Alignment Multi-Modal Recommendation for Reasoning User Sequence Modeling

With the evolution of large language models (LLMs), there is growing interest in leveraging their rich semantic understanding to enhance industrial recommendation systems (RecSys). Traditional RecSys relies on ID-based embeddings for user sequence modeling in the General Search Unit (GSU) and Exact Search Unit (ESU) paradigm, which suffers from low information density, knowledge isolation, and weak generalization ability. While LLMs offer complementary strengths with dense semantic representations and strong generalization, directly applying LLM embeddings to RecSys faces critical challenges: representation unmatch with business objectives and representation unlearning end-to-end with downstream tasks. In this paper, we present QARM V2, a unified framework that bridges LLM semantic understanding with RecSys business requirements for user sequence modeling.

GLASS: A Generative Recommender for Long-sequence Modeling via SID-Tier and Semantic Search

Leveraging long-term user behavioral patterns is a key trajectory for enhancing the accuracy of modern recommender systems. While generative recommender systems have emerged as a transformative paradigm, they face hurdles in effectively modeling extensive historical sequences. To address this challenge, we propose GLASS, a novel framework that integrates long-term user interests into the generative process via SID-Tier and Semantic Search. We first introduce SID-Tier, a module that maps long-term interactions into a unified interest vector to enhance the prediction of the initial SID token. Unlike traditional retrieval models that struggle with massive item spaces, SID-Tier leverages the compact nature of the semantic codebook to incorporate cross features between the user's long-term history and candidate semantic codes. Furthermore, we present semantic hard search, which utilizes generated coarse-grained semantic ID as dynamic keys to extract relevant historical behaviors, which are then fused via an adaptive gated fusion module to recalibrate the trajectory of subsequent fine-grained tokens. To address the inherent data sparsity in semantic hard search, we propose two strategies: semantic neighbor augmentation and codebook resizing. Extensive experiments on two large-scale real-world datasets, TAOBAO-MM and KuaiRec, demonstrate that GLASS outperforms state-of-the-art baselines, achieving significant gains in recommendation quality. Our codes are made publicly available to facilitate further research in generative recommendation.

Semantic Routing: Exploring Multi-Layer LLM Feature Weighting for Diffusion Transformers

Recent DiT-based text-to-image models increasingly adopt LLMs as text encoders, yet text conditioning remains largely static and often utilizes only a single LLM layer, despite pronounced semantic hierarchy across LLM layers and non-stationary denoising dynamics over both diffusion time and network depth. To better match the dynamic process of DiT generation and thereby enhance the diffusion model's generative capability, we introduce a unified normalized convex fusion framework equipped with lightweight gates to systematically organize multi-layer LLM hidden states via time-wise, depth-wise, and joint fusion. Experiments establish Depth-wise Semantic Routing as the superior conditioning strategy, consistently improving text-image alignment and compositional generation (e.g., +9.97 on the GenAI-Bench Counting task). Conversely, we find that purely time-wise fusion can paradoxically degrade visual generation fidelity. We attribute this to a train-inference trajectory mismatch: under classifier-free guidance, nominal timesteps fail to track the effective SNR, causing semantically mistimed feature injection during inference. Overall, our results position depth-wise routing as a strong and effective baseline and highlight the critical need for trajectory-aware signals to enable robust time-dependent conditioning.

ALPBench: A Benchmark for Attribution-level Long-term Personal Behavior Understanding

Recent advances in large language models have highlighted their potential for personalized recommendation, where accurately capturing user preferences remains a key challenge. Leveraging their strong reasoning and generalization capabilities, LLMs offer new opportunities for modeling long-term user behavior. To systematically evaluate this, we introduce ALPBench, a Benchmark for Attribution-level Long-term Personal Behavior Understanding. Unlike item-focused benchmarks, ALPBench predicts user-interested attribute combinations, enabling ground-truth evaluation even for newly introduced items. It models preferences from long-term historical behaviors rather than users' explicitly expressed requests, better reflecting enduring interests. User histories are represented as natural language sequences, allowing interpretable, reasoning-based personalization. ALPBench enables fine-grained evaluation of personalization by focusing on the prediction of attribute combinations task that remains highly challenging for current LLMs due to the need to capture complex interactions among multiple attributes and reason over long-term user behavior sequences.