Text_Generation
-
Differentially Private and Communication Efficient Large Language Model Split Inference via Stochastic Quantization and Soft Prompt
2026-02-12 03:13:16Yujie Gu, Richeng Jin, Xiaoyu Ji, Yier Jin, Wenyuan XuAbstract
Large Language Models (LLMs) have achieved remarkable performance and received significant research interest. The enormous computational demands, however, hinder the local deployment on devices with limited resources. The current prevalent LLM inference paradigms require users to send queries to the service providers for processing, which raises critical privacy concerns. Existing approaches propose to allow the users to obfuscate the token embeddings before transmission and utilize local models for denoising. Nonetheless, transmitting the token embeddings and deploying local models may result in excessive communication and computation overhead, preventing practical implementation. In this work, we propose \textbf{DEL}, a framework for \textbf{D}ifferentially private and communication \textbf{E}fficient \textbf{L}LM split inference. More specifically, an embedding projection module and a differentially private stochastic quantization mechanism are proposed to reduce the communication overhead in a privacy-preserving manner. To eliminate the need for local models, we adapt soft prompt at the server side to compensate for the utility degradation caused by privacy. To the best of our knowledge, this is the first work that utilizes soft prompt to improve the trade-off between privacy and utility in LLM inference, and extensive experiments on text generation and natural language understanding benchmarks demonstrate the effectiveness of the proposed method.
Abstract (translated)
大型语言模型（LLMs）已经取得了显著的性能，并且受到了广泛的研究兴趣。然而，巨大的计算需求阻碍了这些模型在资源有限的设备上的本地部署。目前流行的LLM推理范式要求用户将查询发送给服务提供商进行处理，这引发了重要的隐私问题。现有方法提出允许用户在传输之前对令牌嵌入进行混淆处理，并利用本地模型进行去噪。然而，传输令牌嵌入和部署本地模型可能会导致过多的通信和计算开销，从而阻碍实际应用。 为此，我们提出了**DEL框架**（Differentially private and communication Efficient LLM split inference），旨在以保护隐私的方式减少通信开销并实现有效的LLM分割推理。具体来说，我们设计了一个嵌入投影模块，并提出了一种差分私有随机量化机制来降低通信开销。为了消除对本地模型的需求，我们在服务器端采用软提示（soft prompt）来补偿由于隐私保护导致的效用下降。 据我们所知，这是首次利用软提示来改进LLM推理中隐私与性能之间的权衡的工作，并且在文本生成和自然语言理解基准上的广泛实验表明了该方法的有效性。
URL
https://arxiv.org/abs/2602.11513
PDF
-
TRACER: Trajectory Risk Aggregation for Critical Episodes in Agentic Reasoning
2026-02-11 22:23:56Sina Tayebati, Divake Kumar, Nastaran Darabi, Davide Ettori, Ranganath Krishnan, Amit Ranjan TrivediAbstract
Estimating uncertainty for AI agents in real-world multi-turn tool-using interaction with humans is difficult because failures are often triggered by sparse critical episodes (e.g., looping, incoherent tool use, or user-agent miscoordination) even when local generation appears confident. Existing uncertainty proxies focus on single-shot text generation and therefore miss these trajectory-level breakdown signals. We introduce TRACER, a trajectory-level uncertainty metric for dual-control Tool-Agent-User interaction. TRACER combines content-aware surprisal with situational-awareness signals, semantic and lexical repetition, and tool-grounded coherence gaps, and aggregates them using a tail-focused risk functional with a MAX-composite step risk to surface decisive anomalies. We evaluate TRACER on $\tau^2$-bench by predicting task failure and selective task execution. To this end, TRACER improves AUROC by up to 37.1% and AUARC by up to 55% over baselines, enabling earlier and more accurate detection of uncertainty in complex conversational tool-use settings. Our code and benchmark are available at this https URL.
Abstract (translated)
在现实世界中，AI代理与人类进行多轮工具使用交互时，估算不确定性非常困难。因为即使局部生成看起来很自信，失败也可能由稀疏的关键事件（如循环、不连贯的工具使用或用户-代理协调失误）触发。现有的不确定性的替代指标主要关注单次文本生成，因此忽略了这些轨迹层面崩溃信号。 我们引入了TRACER，这是一种针对双控制工具-代理-用户的交互的轨迹级不确定性度量方法。TRACER结合了基于内容的意外性（surprisal）、情境感知信号、语义和词汇重复以及工具相关的连贯性差距，并使用一种尾部聚焦的风险函数来汇总这些因素，同时采用MAX复合步骤风险以揭示决定性的异常情况。 我们在$\tau^2$-bench上评估了TRACER，通过预测任务失败和选择性任务执行的能力。为此，与基线相比，TRACER将AUROC提高了最多37.1%，将AUARC提高了最多55%，从而在复杂的对话工具使用环境中实现了更早且更准确的不确定性检测。 我们的代码和基准测试可在以下网址获取：[this https URL]。
URL
https://arxiv.org/abs/2602.11409
PDF
-
Beyond Confidence: The Rhythms of Reasoning in Generative Models
2026-02-11 12:58:23Deyuan Liu, Zecheng Wang, Zhanyue Qin, Zhiying Tu, Dianhui Chu, Dianbo SuiAbstract
Large Language Models (LLMs) exhibit impressive capabilities yet suffer from sensitivity to slight input context variations, hampering reliability. Conventional metrics like accuracy and perplexity fail to assess local prediction robustness, as normalized output probabilities can obscure the underlying resilience of an LLM's internal state to perturbations. We introduce the Token Constraint Bound ($\delta_{\mathrm{TCB}}$), a novel metric that quantifies the maximum internal state perturbation an LLM can withstand before its dominant next-token prediction significantly changes. Intrinsically linked to output embedding space geometry, $\delta_{\mathrm{TCB}}$ provides insights into the stability of the model's internal predictive commitment. Our experiments show $\delta_{\mathrm{TCB}}$ correlates with effective prompt engineering and uncovers critical prediction instabilities missed by perplexity during in-context learning and text generation. $\delta_{\mathrm{TCB}}$ offers a principled, complementary approach to analyze and potentially improve the contextual stability of LLM predictions.
Abstract (translated)
大型语言模型（LLM）展示了令人印象深刻的性能，但它们对输入上下文的细微变化敏感，这影响了其可靠性。传统的准确性、困惑度等评估指标无法衡量局部预测的稳健性，因为归一化的输出概率掩盖了LLM内部状态对扰动的实际抵抗力。我们引入了一种新的度量标准——Token Constraint Bound（$\delta_{\mathrm{TCB}}$），用于量化在大型语言模型的主导下一个标记预测发生显著变化之前，其内部状态能承受的最大扰动程度。该指标与输出嵌入空间几何结构内在关联，能够揭示模型内部预测承诺的稳定性。实验表明，$\delta_{\mathrm{TCB}}$ 与有效的提示工程相关，并且在上下文学习和文本生成过程中发现了困惑度所未能捕捉到的关键预测不稳定性。因此，$\delta_{\mathrm{TCB}}$ 提供了一种原则性的、互补的方法来分析并潜在地改进大型语言模型的上下文稳定性。
URL
https://arxiv.org/abs/2602.10816
PDF
-
Flow of Spans: Generalizing Language Models to Dynamic Span-Vocabulary via GFlowNets
2026-02-11 07:17:41Bo Xue, Yunchong Song, Fanghao Shao, Xuekai Zhu, Lin Chen, Luoyi Fu, Xinbing Wang, Zhouhan LinAbstract
Standard autoregressive language models generate text token-by-token from a fixed vocabulary, inducing a tree-structured state space when viewing token sampling as an action, which limits flexibility and expressiveness. Recent work introduces dynamic vocabulary by sampling retrieved text spans but overlooks that the same sentence can be composed of spans of varying lengths, lacking explicit modeling of the directed acyclic graph (DAG) state space. This leads to restricted exploration of compositional paths and is biased toward the chosen path. Generative Flow Networks (GFlowNets) are powerful for efficient exploring and generalizing over state spaces, particularly those with a DAG structure. However, prior GFlowNets-based language models operate at the token level and remain confined to tree-structured spaces, limiting their potential. In this work, we propose Flow of SpanS (FOSS), a principled GFlowNets framework for span generation. FoSS constructs a dynamic span vocabulary by segmenting the retrieved text flexibly, ensuring a DAG-structured state space, which allows GFlowNets to explore diverse compositional paths and improve generalization. With specialized reward models, FoSS generates diverse, high-quality text. Empirically, FoSS improves MAUVE scores by up to 12.5% over Transformer on text generation and achieves 3.5% gains on knowledge-intensive tasks, consistently outperforming state-of-the-art methods. Scaling experiments further demonstrate FoSS benefits from larger models, more data, and richer retrieval corpora, retaining its advantage over strong baselines.
Abstract (translated)
标准的自回归语言模型从固定词汇表中逐词生成文本，当把单词采样视为一个动作时，这种做法会产生树状结构的状态空间。这样的方法限制了灵活性和表达能力。最近的研究通过抽取检索到的文本片段来引入动态词汇表，但忽略了同一个句子可以由不同长度的片段组成这一事实，并且没有显式地对有向无环图（DAG）状态空间进行建模。这导致了组合路径探索受限并偏向于所选路径。 生成流网络（GFlowNets）对于有效探索和泛化处理具有DAG结构的状态空间非常有用。然而，基于GFlowNet的语言模型先前仅在词级别上操作，并且仍然局限于树状结构的空间中工作，从而限制了它们的潜力。在这项工作中，我们提出了SpanS流（FOSS），这是一种用于片段生成的原则性GFlowNets框架。FoSS通过灵活地分割检索到的文本构建动态片段词汇表，确保了DAG结构的状态空间，使GFlowNets能够探索多种组合路径并提高泛化能力。 借助专门设计的奖励模型，FoSS可以产生多样且高质量的文本。实证研究表明，在文本生成方面，FoSS相比Transformer提高了高达12.5%的MAUVE分数，并在知识密集型任务上取得了3.5%的改进，持续优于当前最先进的方法。规模实验进一步表明，随着模型更大、数据更多以及检索语料库更丰富时，FoSS能够保持其相对于强基线的优势。
URL
https://arxiv.org/abs/2602.10583
PDF
-
AmharicIR+Instr: A Two-Dataset Resource for Neural Retrieval and Instruction Tuning
2026-02-10 15:45:20Tilahun Yeshambel, Moncef Garouani, Josiane MotheAbstract
Neural retrieval and GPT-style generative models rely on large, high-quality supervised data, which is still scarce for low-resource languages such as Amharic. We release an Amharic data resource consisting of two datasets that supports research on (i) neural retrieval-ranking and (ii) instruction-following text generation. The retrieval-ranking dataset contains 1,091 manually verified query-positive-negative document triplets drawn from diverse Amharic sources and constructed to support contrastive training and benchmarking of neural retrievers (e.g., DPR, ColBERT-style late interaction and SPLADE-style sparse neural retrieval). Triplets are created through a combination of expert-curated queries, web-derived queries, and LLM-assisted generation, with positive/negative documents selected from the web or synthesized by LLMs and then validated by native speakers. The instruction prompt-response dataset comprises 6,285 Amharic prompt-response pairs spanning multiple domains and instruction types, generated with several LLMs and refined through manual review and correction for grammaticality, relevance, fluency, and factual plausibility. We release both datasets with standardized splits and formats (CSV,JSON,JSONL) to enable reproducible work on Amharic retrieval, ranking, and generative modelling. These datasets also come with a methodology that can be generalized to other low-resource languages.
Abstract (translated)
神经检索和GPT风格的生成模型依赖于大量高质量的监督数据，但对于像阿姆哈拉语这样的低资源语言来说，这种类型的数据仍然稀缺。我们发布了一个包含两个数据集的阿姆哈拉语数据资源，支持关于（i）神经检索排序研究以及（ii）指令跟随文本生成的研究。 **检索-排名数据集**包含了1,091个人工验证过的查询-正样本-负样本三元组，这些三元组是从多种阿姆哈拉语来源中提取并构建出来的，用于支持对比训练和神经检索器（例如DPR、ColBERT风格的后期交互以及SPLADE风格的稀疏神经检索）的基准测试。这些建立了查询-文档对的数据集通过专家策划的查询、网络衍生的查询及LLM辅助生成相结合的方法创建，而正样本/负样本文档则从互联网上选择或由LLMs合成后，再经过母语者的验证。 **指令提示响应数据集**包含6,285个阿姆哈拉语的指令-响应对，这些数据覆盖了多个领域和多种类型的指令，并通过几个大型语言模型生成并经手工审核修正以确保语法正确性、相关性、流畅性和事实上的合理性后得到完善。 我们以标准化的分割格式（CSV、JSON、JSONL）发布这两个数据集，以便于在阿姆哈拉语检索、排名及生成建模方面进行可重复的研究。此外，这些数据集还附带了一种可以推广到其他低资源语言的方法论。
URL
https://arxiv.org/abs/2602.09914
PDF
-
Schr'oMind: Mitigating Hallucinations in Multimodal Large Language Models via Solving the Schr'odinger Bridge Problem
2026-02-10 08:36:40Ziqiang Shi, Rujie Liu, Shanshan Yu, Satoshi Munakata, Koichi ShirahataAbstract
Recent advancements in Multimodal Large Language Models (MLLMs) have achieved significant success across various domains. However, their use in high-stakes fields like healthcare remains limited due to persistent hallucinations, where generated text contradicts or ignores visual input. We contend that MLLMs can comprehend images but struggle to produce accurate token sequences. Minor perturbations can shift attention from truthful to untruthful states, and the autoregressive nature of text generation often prevents error correction. To address this, we propose SchröMind-a novel framework reducing hallucinations via solving the Schrödinger bridge problem. It establishes a token-level mapping between hallucinatory and truthful activations with minimal transport cost through lightweight training, while preserving the model's original capabilities. Extensive experiments on the POPE and MME benchmarks demonstrate the superiority of Schrödinger, which achieves state-of-the-art performance while introducing only minimal computational overhead.
Abstract (translated)
最近，在多模态大型语言模型（MLLM）领域的进展在多个领域取得了显著的成功。然而，这些模型在医疗保健等高风险行业中的应用仍然受到限制，原因在于它们存在持续的“幻觉”问题，即生成的文字内容与其视觉输入相矛盾或忽视了视觉信息。我们主张，尽管MLLM能够理解图像，但在产生准确的令牌序列方面仍存在问题。微小的变化可能会导致注意力从真实状态转向虚假状态，并且文本生成的自回归特性通常会阻止错误修正。 为了解决这些问题，我们提出了一种名为SchröMind的新框架，通过解决薛定谔桥问题来减少幻觉现象。该框架在轻量级训练过程中建立了幻觉激活和真实激活之间的令牌级别映射，同时尽量减少了传输成本，并保持了模型原有的能力。大量的实验表明，在POPE和MME基准测试上，Schrödinger方法展示了其优越性：它不仅实现了最先进的性能，还仅引入了最小的计算开销。
URL
https://arxiv.org/abs/2602.09528
PDF
-
Generalizing Sports Feedback Generation by Watching Competitions and Reading Books: A Rock Climbing Case Study
2026-02-09 18:41:43Arushi Rai, Adriana KovashkaAbstract
While there is rapid progress in video-LLMs with advanced reasoning capabilities, prior work shows that these models struggle on the challenging task of sports feedback generation and require expensive and difficult-to-collect finetuning feedback data for each sport. This limitation is evident from the poor generalization to sports unseen during finetuning. Furthermore, traditional text generation evaluation metrics (e.g., BLEU-4, METEOR, ROUGE-L, BERTScore), originally developed for machine translation and summarization, fail to capture the unique aspects of sports feedback quality. To address the first problem, using rock climbing as our case study, we propose using auxiliary freely-available web data from the target domain, such as competition videos and coaching manuals, in addition to existing sports feedback from a disjoint, source domain to improve sports feedback generation performance on the target domain. To improve evaluation, we propose two evaluation metrics: (1) specificity and (2) actionability. Together, our approach enables more meaningful and practical generation of sports feedback under limited annotations.
Abstract (translated)
尽管具备高级推理能力的视频大语言模型（video-LLMs）取得了快速进展，但先前的研究表明，这些模型在生成体育反馈这一具有挑战性的任务上表现不佳，并且需要昂贵且难以收集的微调数据来针对每项运动进行优化。这种局限性明显体现在模型对微调过程中未见过的运动项目的泛化能力差的问题上。此外，传统的文本生成评估指标（如BLEU-4、METEOR、ROUGE-L和BERTScore），最初为机器翻译和摘要任务设计，无法捕捉体育反馈质量的独特方面。 为了应对第一个问题，我们以攀岩作为案例研究，提出了一种利用目标领域中辅助的免费网络数据的方法，例如比赛视频和教练手册，并结合来自独立源领域的现有体育反馈来提高针对特定运动项目的体育反馈生成性能。为了改进评估方法，我们提出了两个新的评价指标：（1）具体性；（2）可操作性。 我们的方法能够更有效地在注释有限的情况下生成有意义且实用的体育反馈。
URL
https://arxiv.org/abs/2602.08996
PDF
-
Equalized Generative Treatment: Matching f-divergences for Fairness in Generative Models
2026-02-09 13:52:36Alexandre Verine, Rafael Pinot, Florian Le BronnecAbstract
Fairness is a crucial concern for generative models, which not only reflect but can also amplify societal and cultural biases. Existing fairness notions for generative models are largely adapted from classification and focus on balancing the probability of generating samples from each sensitive group. We show that such criteria are brittle, as they can be met even when different sensitive groups are modeled with widely varying quality. To address this limitation, we introduce a new fairness definition for generative models, termed as equalized generative treatment (EGT), which requires comparable generation quality across all sensitive groups, with quality measured via a reference f-divergence. We further analyze the trade-offs induced by EGT, demonstrating that enforcing fairness constraints necessarily couples the overall model quality to that of the most challenging group to approximate. This indicates that a simple yet efficient min-max fine-tuning method should be able to balance f-divergences across sensitive groups to satisfy EGT. We validate this theoretical insight through a set of experiments on both image and text generation tasks. We demonstrate that min-max methods consistently achieve fairer outcomes compared to other approaches from the literature, while maintaining competitive overall performance for both tasks.
Abstract (translated)
公平性是生成模型中的一个关键问题，因为这些模型不仅会反映社会和文化的偏见，还可能放大这些问题。现有的针对生成模型的公平概念大多借鉴自分类方法，并主要集中在平衡从每个敏感群体中生成样本的概率上。我们证明了这种标准较为脆弱，即使当不同敏感群体的质量相差甚远时，也仍然可以满足这些标准。 为了解决这一局限性，我们引入了一种新的用于生成模型的公平定义——等效生成待遇（Equalized Generative Treatment, EGT），它要求所有敏感群体之间具有可比较的生成质量，并通过参考 f-散度来测量这种质量。进一步地，我们分析了由 EGT 引起的权衡关系，表明实施公平性约束将不可避免地使整体模型质量与最难以接近近似的困难群体的质量相挂钩。这表明了一种简单的最小化—最大化微调方法应该能够平衡敏感群体间的 f-散度以满足 EGT 要求。 我们通过一系列针对图像和文本生成任务的实验验证了这一理论见解，并展示了使用最小—最大方法可以持续实现比其他文献中的方法更加公平的结果，同时保持在两项任务上的竞争力。
URL
https://arxiv.org/abs/2602.08660
PDF
-
Geometric Image Editing via Effects-Sensitive In-Context Inpainting with Diffusion Transformers
2026-02-09 08:39:47Shuo Zhang, Wenzhuo Wu, Huayu Zhang, Jiarong Cheng, Xianghao Zang, Chao Ban, Hao Sun, Zhongjiang He, Tianwei Cao, Kongming Liang, Zhanyu MaAbstract
Recent advances in diffusion models have significantly improved image editing. However, challenges persist in handling geometric transformations, such as translation, rotation, and scaling, particularly in complex scenes. Existing approaches suffer from two main limitations: (1) difficulty in achieving accurate geometric editing of object translation, rotation, and scaling; (2) inadequate modeling of intricate lighting and shadow effects, leading to unrealistic results. To address these issues, we propose GeoEdit, a framework that leverages in-context generation through a diffusion transformer module, which integrates geometric transformations for precise object edits. Moreover, we introduce Effects-Sensitive Attention, which enhances the modeling of intricate lighting and shadow effects for improved realism. To further support training, we construct RS-Objects, a large-scale geometric editing dataset containing over 120,000 high-quality image pairs, enabling the model to learn precise geometric editing while generating realistic lighting and shadows. Extensive experiments on public benchmarks demonstrate that GeoEdit consistently outperforms state-of-the-art methods in terms of visual quality, geometric accuracy, and realism.
Abstract (translated)
最近在扩散模型方面取得的进展显著提高了图像编辑的质量。然而，在处理几何变换（如平移、旋转和缩放）时，特别是在复杂场景中，仍然面临挑战。现有方法存在两大主要局限性：(1) 难以实现对象平移、旋转和缩放的精确几何编辑；(2) 不充分地模拟复杂的光照和阴影效果，导致生成的结果不够真实。 为了解决这些问题，我们提出了GeoEdit框架，该框架利用扩散变换器模块通过上下文生成来整合几何变换，从而实现对物体的精准编辑。此外，我们引入了敏感效应注意力机制（Effects-Sensitive Attention），以增强复杂照明和阴影效果的建模，提高现实感。 为了进一步支持训练，我们构建了一个大规模的几何编辑数据集RS-Objects，包含超过120,000对高质量图像对，使模型能够学习精准的几何编辑并生成逼真的光照与阴影。在公共基准测试中的广泛实验表明，GeoEdit在视觉质量、几何精度和现实感方面始终优于当前最先进的方法。
URL
https://arxiv.org/abs/2602.08388
PDF
-
TIDE: Tuning-Integrated Dynamic Evolution for LLM-Based Automated Heuristic Design
2026-02-07 09:56:34Chentong Chen, Mengyuan Zhong, Ye Fan, Jialong Shi, Jianyong SunAbstract
Although Large Language Models have advanced Automated Heuristic Design, treating algorithm evolution as a monolithic text generation task overlooks the coupling between discrete algorithmic structures and continuous numerical parameters. Consequently, existing methods often discard promising algorithms due to uncalibrated constants and suffer from premature convergence resulting from simple similarity metrics. To address these limitations, we propose TIDE, a Tuning-Integrated Dynamic Evolution framework designed to decouple structural reasoning from parameter optimization. TIDE features a nested architecture where an outer parallel island model utilizes Tree Similarity Edit Distance to drive structural diversity, while an inner loop integrates LLM-based logic generation with a differential mutation operator for parameter tuning. Additionally, a UCB-based scheduler dynamically prioritizes high-yield prompt strategies to optimize resource allocation. Extensive experiments across nine combinatorial optimization problems demonstrate that TIDE discovers heuristics that significantly outperform state-of-the-art baselines in solution quality while achieving improved search efficiency and reduced computational costs.
Abstract (translated)
尽管大型语言模型已经推进了自动启发式设计，但将算法演化视为单一的文本生成任务忽视了离散算法结构与连续数值参数之间的耦合。因此，现有方法常常因未调校的常数而舍弃有前景的算法，并且由于简单的相似性度量而导致过早收敛。为了克服这些限制，我们提出了TIDE框架（一种集成调优的动态演化框架），旨在将结构推理与参数优化分离。TIDE具有一个嵌套架构，在该架构中，外部并行岛屿模型利用树相似编辑距离来驱动结构多样性，而内部循环则整合基于大型语言模型的逻辑生成与差分变异算子进行参数调整。此外，一种基于UCB（Upper Confidence Bound）调度器动态优先处理高收益提示策略以优化资源分配。 通过九个组合优化问题上的广泛实验表明，TIDE能够发现显著优于现有最佳基线算法的质量更佳的启发式方法，并且实现了改进的搜索效率和降低的计算成本。
URL
https://arxiv.org/abs/2601.21239
PDF
-
DAWN: Dependency-Aware Fast Inference for Diffusion LLMs
2026-02-06 18:51:29Lizhuo Luo, Zhuoran Shi, Jiajun Luo, Zhi Wang, Shen Ren, Wenya Wang, Tianwei ZhangAbstract
Diffusion large language models (dLLMs) have shown advantages in text generation, particularly due to their inherent ability for parallel decoding. However, constrained by the quality--speed trade-off, existing inference solutions adopt conservative parallel strategies, leaving substantial efficiency potential underexplored. A core challenge is that parallel decoding assumes each position can be filled independently, but tokens are often semantically coupled. Thus, the correct choice at one position constrains valid choices at others. Without modeling these inter-token dependencies, parallel strategies produce deteriorated outputs. Motivated by this insight, we propose DAWN, a training-free, dependency-aware decoding method for fast dLLM inference. DAWN extracts token dependencies and leverages two key motivations: (1) positions dependent on unmasked certain positions become more reliable, (2) simultaneously unmasking strongly coupled uncertain positions induces errors. Given those findings, DAWN leverages a dependency graph to select more reliable unmasking positions at each iteration, achieving high parallelism with negligible loss in generation quality. Extensive experiments across multiple models and datasets demonstrate that DAWN speedups the inference by 1.80-8.06x over baselines while preserving the generation quality. Code is released at this https URL.
Abstract (translated)
扩散大型语言模型（dLLMs）在文本生成方面表现出色，尤其得益于其固有的并行解码能力。然而，受限于质量与速度的权衡问题，现有的推理解决方案采用了保守的并行策略，未能充分利用潜在的效率优势。一个核心挑战在于，并行解码假设每个位置都可以独立填充，但实际上令牌之间往往存在语义上的关联性。因此，在某一位置上做出正确的选择会限制其他位置的有效选择范围。若不考虑这些跨令牌依赖关系，简单的并行策略会产生质量较差的结果。 基于这一洞察，我们提出了DAWN——一种训练无关的、具备依存感知能力的解码方法，以实现快速且高效的dLLM推理过程。DAWN能够提取令牌间的依赖关系，并利用两个关键动机：（1）在未屏蔽特定位置时，那些与此相关的位置变得更加可靠；（2）同时对高度相关的不确定位置进行解锁操作会导致错误。 基于这些发现，DAWN通过使用一个依存图来选择在每次迭代中更加可靠的未屏蔽位置，从而实现了高并行性，并且几乎不会损失生成的质量。广泛的实验表明，在多个模型和数据集上，相比基线方法，DAWN能够将推理速度提升1.80到8.06倍，同时保持生成质量不变。 代码已发布于[此链接](https://this-url.com/)（请替换为实际的URL）。
URL
https://arxiv.org/abs/2602.06953
PDF
-
DSB: Dynamic Sliding Block Scheduling for Diffusion LLMs
2026-02-05 18:41:38Lizhuo Luo, Shenggui Li, Yonggang Wen, Tianwei ZhangAbstract
Diffusion large language models (dLLMs) have emerged as a promising alternative for text generation, distinguished by their native support for parallel decoding. In practice, block inference is crucial for avoiding order misalignment in global bidirectional decoding and improving output quality. However, the widely-used fixed, predefined block (naive) schedule is agnostic to semantic difficulty, making it a suboptimal strategy for both quality and efficiency: it can force premature commitments to uncertain positions while delaying easy positions near block boundaries. In this work, we analyze the limitations of naive block scheduling and disclose the importance of dynamically adapting the schedule to semantic difficulty for reliable and efficient inference. Motivated by this, we propose Dynamic Sliding Block (DSB), a training-free block scheduling method that uses a sliding block with a dynamic size to overcome the rigidity of the naive block. To further improve efficiency, we introduce DSB Cache, a training-free KV-cache mechanism tailored to DSB. Extensive experiments across multiple models and benchmarks demonstrate that DSB, together with DSB Cache, consistently improves both generation quality and inference efficiency for dLLMs. Code is released at this https URL.
Abstract (translated)
扩散型大型语言模型（dLLM）作为一种有前景的文本生成替代方案已经崭露头角，其特点是原生支持并行解码。实践中，块推理对于避免全局双向解码中的顺序错位以及提高输出质量至关重要。然而，广泛使用的固定、预定义块（朴素）调度方法不考虑语义难度，这使得它在质量和效率方面都不是最优策略：可能会强制过早地对不确定位置做出承诺，同时推迟接近区块边界处容易解决的位置。 在这项工作中，我们分析了朴素块调度的局限性，并揭示了根据语义难度动态调整调度表对于可靠且高效推理的重要性。受此启发，我们提出了动态滑动块（DSB），这是一种无训练的块调度方法，使用具有动态大小的滑动块来克服朴素块的刚性。为了进一步提高效率，我们引入了针对DSB定制的无需训练的KV缓存机制——DSB Cache。 在多个模型和基准上的广泛实验表明，无论是生成质量还是推理效率，DSB（连同DSB Cache）都能持续提升dLLM的表现。相关代码已在此网址发布：[此URL]。
URL
https://arxiv.org/abs/2602.05992
PDF
-
RA-QA: Towards Respiratory Audio-based Health Question Answering
2026-02-04 13:25:47Gaia A. Bertolino, Yuwei Zhang, Tong Xia, Domenico Talia, Cecilia MascoloAbstract
Respiratory diseases are a leading cause of death globally, highlighting the urgent need for early and accessible screening methods. While some lung auscultation analysis has been automated and machine learning audio based models are able to predict respiratory pathologies, there remains a critical gap: the lack of intelligent systems that can interact in real-time consultations using natural language. Unlike other clinical domains, such as electronic health records, radiological images, and biosignals, where numerous question-answering (QA) datasets and models have been established, audio-based modalities remain notably underdeveloped. We curated and harmonized data from 11 diverse respiratory audio datasets to construct the first Respiratory Audio Question Answering (RA-QA) dataset. As the first multimodal QA resource of its kind focused specifically on respiratory health, RA-QA bridges clinical audio and natural language in a structured, scalable format. This new data resource contains about 7.5 million QA pairs spanning more than 60 attributes and three question types: single verification, multiple choice, and open-ended questions. Building upon this dataset, we introduce a novel benchmark that compares audio-text generation models with traditional audio classifiers to evaluate their respective performance.\\Our experiments reveal interesting performance variations across different attributes and question types, establishing a baseline and paving the way for more advanced architectures that could further improve the performance. By bridging machine learning with real-world clinical dialogue, our work opens the door to the development of more interactive, intelligent, and accessible diagnostic tools in respiratory healthcare.
Abstract (translated)
呼吸系统疾病是全球死亡的主要原因之一，这凸显了早期和易于获取的筛查方法的迫切需求。虽然一些肺部听诊分析已经被自动化，并且基于音频的机器学习模型能够预测呼吸道病理性问题，但仍存在一个关键缺口：缺乏能够在实时咨询中使用自然语言进行交互的智能系统。与电子健康记录、放射图像和生物信号等其他临床领域相比，这些领域的问答（QA）数据集和模型已经建立得相当多，而基于音频模态的数据则明显处于欠开发状态。我们从11个多样化的呼吸音数据集中整理并协调数据，构建了首个呼吸音问答（RA-QA）数据集。作为首个专注于呼吸道健康的多媒体问答资源，RA-QA以结构化和可扩展的格式将临床音频与自然语言连接起来。这一新的数据资源包含了大约750万个问答对，涵盖了超过60个属性以及三种问题类型：单一验证、多项选择题及开放式问题。 基于这个数据集，我们引入了一个新颖的基准测试，对比了音频-文本生成模型和传统音频分类器在不同性能指标上的表现。我们的实验揭示了各种属性和问题类型的性能差异，建立了基线，并为未来能够进一步提升性能的更先进架构铺平道路。通过将机器学习与现实世界中的临床对话相结合，我们的工作为开发更具互动性、智能化且易于访问的诊断工具打开了大门，在呼吸系统医疗领域中具有重要意义。
URL
https://arxiv.org/abs/2602.18452
PDF
-
HoliAntiSpoof: Audio LLM for Holistic Speech Anti-Spoofing
2026-02-04 13:25:24Xuenan Xu, Yiming Ren, Liwei Liu, Wen Wu, Baoxiang Li, Chaochao Lu, Shuai Wang, Chao ZhangAbstract
Recent advances in speech synthesis and editing have made speech spoofing increasingly challenging. However, most existing methods treat spoofing as binary classification, overlooking that diverse spoofing techniques manipulate multiple, coupled speech attributes and their semantic effects. In this paper, we introduce HoliAntiSpoof, the first audio large language model (ALLM) framework for holistic speech anti-spoofing analysis. HoliAntiSpoof reformulates spoofing analysis as a unified text generation task, enabling joint reasoning over spoofing methods, affected speech attributes, and their semantic impacts. To support semantic-level analysis, we introduce DailyTalkEdit, a new anti-spoofing benchmark that simulates realistic conversational manipulations and provides annotations of semantic influence. Extensive experiments demonstrate that HoliAntiSpoof outperforms conventional baselines across multiple settings, while preliminary results show that in-context learning further improves out-of-domain generalization. These findings indicate that ALLMs not only enhance speech spoofing detection performance but also enable interpretable analysis of spoofing behaviors and their semantic effects, pointing towards more trustworthy and explainable speech security. Data and code are publicly available.
Abstract (translated)
近期在语音合成和编辑领域的进展使得语音伪造变得更加困难。然而，大多数现有的方法将语音伪造视为二元分类问题，忽略了各种伪造技术会操纵多个相互关联的语音属性及其语义效果这一事实。在这篇论文中，我们介绍了HoliAntiSpoof，这是第一个用于全面语音防伪分析的音频大型语言模型（ALLM）框架。HoliAntiSpoof将伪造分析重构为一个统一的文字生成任务，使对伪造方法、受影响的语音属性及它们语义影响的同时推理成为可能。为了支持语义级别的分析，我们引入了DailyTalkEdit，一个新的用于防伪基准测试的数据集，它模拟现实中的对话操作，并提供了关于语义影响的注释。广泛的实验表明，HoliAntiSpoof在多个设置下超越了传统的基线方法，而初步结果显示，在上下文学习的帮助下，模型在外域泛化方面也有显著提升。这些发现表明，ALLMs不仅能提高语音伪造检测的表现力，还能使对伪造行为及其语义效果的解释性分析成为可能，从而朝着更值得信赖和可解释的语音安全迈进了一步。数据和代码现已公开提供。
URL
https://arxiv.org/abs/2602.04535
PDF
-
Vision-aligned Latent Reasoning for Multi-modal Large Language Model
2026-02-04 12:04:02Byungwoo Jeon, Yoonwoo Jeong, Hyunseok Lee, Minsu Cho, Jinwoo ShinAbstract
Despite recent advancements in Multi-modal Large Language Models (MLLMs) on diverse understanding tasks, these models struggle to solve problems which require extensive multi-step reasoning. This is primarily due to the progressive dilution of visual information during long-context generation, which hinders their ability to fully exploit test-time scaling. To address this issue, we introduce Vision-aligned Latent Reasoning (VaLR), a simple, yet effective reasoning framework that dynamically generates vision-aligned latent tokens before each Chain of Thought reasoning step, guiding the model to reason based on perceptual cues in the latent space. Specifically, VaLR is trained to preserve visual knowledge during reasoning by aligning intermediate embeddings of MLLM with those from vision encoders. Empirical results demonstrate that VaLR consistently outperforms existing approaches across a wide range of benchmarks requiring long-context understanding or precise visual perception, while exhibiting test-time scaling behavior not observed in prior MLLMs. In particular, VaLR improves the performance significantly from 33.0% to 52.9% on VSI-Bench, achieving a 19.9%p gain over Qwen2.5-VL.
Abstract (translated)
尽管在多模态大型语言模型（MLLM）处理多样化理解任务方面取得了近期进展，但这些模型仍然难以解决需要复杂多步推理的问题。这主要是因为在长时间上下文生成过程中视觉信息逐渐稀释，从而限制了它们充分利用测试时放大的能力。为了解决这个问题，我们引入了一种简单而有效的推理框架——视觉对齐潜在推理（VaLR）。该框架在每个Chain of Thought推理步骤之前动态生成与视觉对齐的潜在令牌，引导模型根据感知线索在潜在空间中进行推理。具体来说，VaLR通过将MLLM中间嵌入与视觉编码器中的嵌入对齐来训练以在推理过程中保留视觉知识。 实证结果显示，VaLR在一系列需要长时间上下文理解和精确视觉感知的任务上始终优于现有方法，并且在测试时放大行为方面表现出不同于先前MLLM的特性。特别是在VSI-Bench上，VaLR将性能显著提高至从33.0%到52.9%，比Qwen2.5-VL高出19.9个百分点。 通过这些改进，VaLR证明了它不仅在广泛的基准测试中具有强大的通用性，而且还能有效地解决当前MLLM面临的关键挑战。
URL
https://arxiv.org/abs/2602.04476
PDF
-
SimpleTool: Parallel Decoding for Real-Time LLM Function Calling
2026-02-04 08:58:27Xiaoxin Shi, Jiaxin Wan, Linkang Dong, Wei Jiang, Yue Liu, Zengfeng HuangAbstract
LLM-based function calling enables intelligent agents to interact with external tools and environments, yet autoregressive decoding imposes a fundamental latency bottleneck that limits real-time applications such as embodied intelligence, game AI, and interactive avatars (e.g., 10 Hz control frequency). We observe that function calling differs fundamentally from free-form text generation: structured outputs exhibit substantial token redundancy (delimiters, parameter names), and arguments exhibit weak causal dependencies. Crucially, these two properties must be exploited jointly to achieve real-time performance. We present SimpleTool, which introduces special tokens that serve a dual role: compressing low-entropy tokens (4-6x reduction) while acting as mode selectors that enable independent parallel generation of function name and arguments. This synergistic design achieves 3-6x end-to-end speedup (up to 9.6x) with only +8.2% parallelization overhead. Experiments on five benchmarks across Qwen-series models (0.5B-14B) demonstrate substantial speedup while maintaining competitive or improved accuracy. On Mobile Actions, ST-Qwen-0.5B outperforms Google's FunctionGemma in both accuracy and latency consistency. With quantization on consumer-grade GPU, SimpleTool achieves 61.2ms P50 latency, enabling 16 Hz real-time control at 4B model scale, bridging the gap between LLM function calling and latency-critical real-world deployment.
Abstract (translated)
基于大型语言模型（LLM）的功能调用使智能代理能够与外部工具和环境交互，但自回归解码的基本延迟瓶颈限制了实时应用，例如具身智能、游戏AI和互动化身（如10 Hz控制频率）。我们观察到，功能调用在本质上不同于自由形式的文本生成：结构化的输出表现出大量的令牌冗余（分隔符、参数名），并且参数之间表现出弱因果依赖关系。至关重要的是，必须同时利用这两个特性以实现实时性能。 我们介绍了SimpleTool，它引入了特殊令牌，这些令牌具有双重作用：压缩低熵令牌（减少4-6倍）的同时作为模式选择器使功能名称和参数的独立并行生成成为可能。这种协同设计实现了3-6倍的整体加速（最高可达9.6倍），仅增加了8.2%的并行化开销。我们在五个基准测试中对Qwen系列模型（从0.5B到14B）进行了实验，证明了在保持竞争力甚至提高准确性的同时获得了显著的速度提升。 在Mobile Actions上，ST-Qwen-0.5B在准确性和延迟一致性方面均优于Google的FunctionGemma。通过在消费级GPU上的量化处理，SimpleTool实现了61.2ms的P50延迟，在4B规模模型下达到了每秒16次实时控制的速度，从而弥合了LLM功能调用与延迟敏感的真实世界部署之间的差距。
URL
https://arxiv.org/abs/2603.00030
PDF
-
ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution
2026-02-03 07:16:51Zican Dong, Peiyu Liu, Junyi Li, Zhipeng Chen, Han Peng, Shuo Wang, Wayne Xin ZhaoAbstract
Recently, large language models (LLMs) have shown remarkable reasoning abilities by producing long reasoning traces. However, as the sequence length grows, the key-value (KV) cache expands linearly, incurring significant memory and computation costs. Existing KV cache eviction methods mitigate this issue by discarding less important KV pairs, but often fail to capture complex KV dependencies, resulting in performance degradation. To better balance efficiency and performance, we introduce ForesightKV, a training-based KV cache eviction framework that learns to predict which KV pairs to evict during long-text generations. We first design the Golden Eviction algorithm, which identifies the optimal eviction KV pairs at each step using future attention scores. These traces and the scores at each step are then distilled via supervised training with a Pairwise Ranking Loss. Furthermore, we formulate cache eviction as a Markov Decision Process and apply the GRPO algorithm to mitigate the significant language modeling loss increase on low-entropy tokens. Experiments on AIME2024 and AIME2025 benchmarks of three reasoning models demonstrate that ForesightKV consistently outperforms prior methods under only half the cache budget, while benefiting synergistically from both supervised and reinforcement learning approaches.
Abstract (translated)
近期，大型语言模型（LLM）通过生成长推理轨迹展示了卓越的推理能力。然而，随着序列长度的增长，关键值（KV）缓存会线性扩展，导致显著增加内存和计算成本。现有的KV缓存淘汰方法通过丢弃不重要的KV对来缓解这一问题，但通常无法捕捉复杂的KV依赖关系，从而导致性能下降。为了更好地平衡效率与性能，我们引入了ForesightKV，这是一种基于训练的KV缓存淘汰框架，能够学习在长文本生成过程中预测应被淘汰的KV对。 首先，我们设计了Golden Eviction算法，在每个步骤使用未来注意力分数来识别最佳的淘汰KV对。然后，通过带有成对排名损失（Pairwise Ranking Loss）的监督训练，将这些轨迹和每个步骤中的分数进行提炼。此外，我们将缓存淘汰形式化为马尔可夫决策过程，并应用GRPO算法以减轻低熵令牌上显著的语言建模损耗增加。 在三个推理模型上的AIME2024和AIME2025基准测试中，实验结果表明，ForesightKV在只有半的缓存预算下始终优于先前的方法，并且能够从监督学习和强化学习方法中协同受益。
URL
https://arxiv.org/abs/2602.03203
PDF
-
Which course? Discourse! Teaching Discourse and Generation in the Era of LLMs
2026-02-02 22:35:29Junyi Jessy Li, Yang Janet Liu, Kanishka Misra, Valentina Pyatkin, William SheffieldAbstract
The field of NLP has undergone vast, continuous transformations over the past few years, sparking debates going beyond discipline boundaries. This begs important questions in education: how do we design courses that bridge sub-disciplines in this shifting landscape? This paper explores this question from the angle of discourse processing, an area with rich linguistic insights and computational models for the intentional, attentional, and coherence structure of language. Discourse is highly relevant for open-ended or long-form text generation, yet this connection is under-explored in existing undergraduate curricula. We present a new course, "Computational Discourse and Natural Language Generation". The course is collaboratively designed by a team with complementary expertise and was offered for the first time in Fall 2025 as an upper-level undergraduate course, cross-listed between Linguistics and Computer Science. Our philosophy is to deeply integrate the theoretical and empirical aspects, and create an exploratory mindset inside the classroom and in the assignments. This paper describes the course in detail and concludes with takeaways from an independent survey as well as our vision for future directions.
Abstract (translated)
近年来，自然语言处理（NLP）领域经历了巨大的持续变革，引发了超越学科界限的辩论。这引发了一些关于教育的重要问题：我们如何设计课程来连接这一不断变化景观中的各个子学科？本文从话语处理的角度探讨了这个问题，该领域具有丰富的语言学见解和计算模型，涵盖语言的目的性、注意力结构和连贯性结构。话语在开放式的或长篇文本生成中极为相关，但这种关联在现有的本科课程体系中却鲜有探索。我们提出了一门新的课程：“计算话语与自然语言生成”。这门课程由一个拥有互补专业知识的团队共同设计，并于2025年秋季首次作为高级本科生课程推出，在语言学和计算机科学之间交叉开设。我们的理念是将理论与实证方面深度融合，并在课堂内以及作业中培养一种探索性的思维模式。本文详细描述了该课程的内容，并通过独立调查的结果总结出一些观点，同时展望未来的发展方向。
URL
https://arxiv.org/abs/2602.02878
PDF
-
The Verification Crisis: Expert Perceptions of GenAI Disinformation and the Case for Reproducible Provenance
2026-02-02 13:45:12Alexander Loth, Martin Kappes, Marc-Oliver PahlAbstract
The growth of Generative Artificial Intelligence (GenAI) has shifted disinformation production from manual fabrication to automated, large-scale manipulation. This article presents findings from the first wave of a longitudinal expert perception survey (N=21) involving AI researchers, policymakers, and disinformation specialists. It examines the perceived severity of multimodal threats -- text, image, audio, and video -- and evaluates current mitigation strategies. Results indicate that while deepfake video presents immediate "shock" value, large-scale text generation poses a systemic risk of "epistemic fragmentation" and "synthetic consensus," particularly in the political domain. The survey reveals skepticism about technical detection tools, with experts favoring provenance standards and regulatory frameworks despite implementation barriers. GenAI disinformation research requires reproducible methods. The current challenge is measurement: without standardized benchmarks and reproducibility checklists, tracking or countering synthetic media remains difficult. We propose treating information integrity as an infrastructure with rigor in data provenance and methodological reproducibility.
Abstract (translated)
URL
https://arxiv.org/abs/2602.02100
PDF
-
GuideWeb: A Benchmark for Automatic In-App Guide Generation on Real-World Web UIs
2026-02-02 10:21:03Chengguang Gan, Yoshihiro Tsujii, Yunhao Liang, Tatsunori Mori, Shiwen Ni, Hiroki ItohAbstract
Digital Adoption Platform (DAP) provide web-based overlays that deliver operation guidance and contextual hints to help users navigate complex websites. Although modern DAP tools enable non-experts to author such guidance, maintaining these guides remains labor-intensive because website layouts and functionalities evolve continuously, which requires repeated manual updates and re-annotation. In this work, we introduce \textbf{GuideWeb}, a new benchmark for automatic in-app guide generation on real-world web UIs. GuideWeb formulates the task as producing page-level guidance by selecting \textbf{guide target elements} grounded in the webpage and generating concise guide text aligned with user intent. We also propose a comprehensive evaluation suite that jointly measures the accuracy of guide target element selection and the quality of generated intents and guide texts. Experiments show that our proposed \textbf{GuideWeb Agent} achieves \textbf{30.79\%} accuracy in guide target element prediction, while obtaining BLEU scores of \textbf{44.94} for intent generation and \textbf{21.34} for guide-text generation. Existing baselines perform substantially worse, which highlights that automatic guide generation remains challenging and that further advances are necessary before such systems can be reliably deployed in real-world settings.
Abstract (translated)
URL
https://arxiv.org/abs/2602.01917
PDF
