Text_Generation
-
AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
2026-04-03 10:29:31Yunhao Feng, Yifan Ding, Yingshui Tan, Xingjun Ma, Yige Li, Yutao Wu, Yifeng Gao, Kun Zhai, Yanming GuoAbstract
Computer-use agents extend language models from text generation to persistent action over tools, files, and execution environments. Unlike chat systems, they maintain state across interactions and translate intermediate outputs into concrete actions. This creates a distinct safety challenge in that harmful behavior may emerge through sequences of individually plausible steps, including intermediate actions that appear locally acceptable but collectively lead to unauthorized actions. We present \textbf{AgentHazard}, a benchmark for evaluating harmful behavior in computer-use agents. AgentHazard contains \textbf{2,653} instances spanning diverse risk categories and attack strategies. Each instance pairs a harmful objective with a sequence of operational steps that are locally legitimate but jointly induce unsafe behavior. The benchmark evaluates whether agents can recognize and interrupt harm arising from accumulated context, repeated tool use, intermediate actions, and dependencies across steps. We evaluate AgentHazard on Claude Code, OpenClaw, and IFlow using mostly open or openly deployable models from the Qwen3, Kimi, GLM, and DeepSeek families. Our experimental results indicate that current systems remain highly vulnerable. In particular, when powered by Qwen3-Coder, Claude Code exhibits an attack success rate of \textbf{73.63\%}, suggesting that model alignment alone does not reliably guarantee the safety of autonomous agents.
Abstract (translated)
计算机使用代理将语言模型从文本生成扩展至对工具、文件和执行环境的持久化操作。与聊天系统不同，它们能在多轮交互中维持状态，并将中间输出转化为具体操作。这带来了独特的安全挑战：有害行为可能通过一系列单独看似合理的步骤逐步显现，包括局部可接受但 collectively 导致越权操作的中间行为。我们提出\textbf{AgentHazard}，一个用于评估计算机使用代理有害行为的基准测试。AgentHazard包含\textbf{2,653}个实例，涵盖多样化的风险类别和攻击策略。每个实例将有害目标与一系列操作上合法但共同引发不安全行为的步骤配对。该基准测试评估代理能否识别并中断由累积上下文、重复工具使用、中间步骤依赖等因素引发的危害。我们在Claude Code、OpenClaw和IFlow上使用Qwen3、Kimi、GLM和DeepSeek系列中多数开源或可公开部署的模型评估了AgentHazard。实验结果表明当前系统仍高度脆弱。尤其当采用Qwen3-Coder驱动时，Claude Code表现出\textbf{73.63\%}的攻击成功率，表明仅靠模型对齐无法可靠保障自主代理的安全性。
URL
https://arxiv.org/abs/2604.02947
PDF
-
Do Audio-Visual Large Language Models Really See and Hear?
2026-04-03 00:48:49Ramaneswaran Selvakumar, Kaousheik Jayakumar, S Sakshi, Sreyan Ghosh, Ruohan Gao, Dinesh ManochaAbstract
Audio-Visual Large Language Models (AVLLMs) are emerging as unified interfaces to multimodal perception. We present the first mechanistic interpretability study of AVLLMs, analyzing how audio and visual features evolve and fuse through different layers of an AVLLM to produce the final text outputs. We find that although AVLLMs encode rich audio semantics at intermediate layers, these capabilities largely fail to surface in the final text generation when audio conflicts with vision. Probing analyses show that useful latent audio information is present, but deeper fusion layers disproportionately privilege visual representations that tend to suppress audio cues. We further trace this imbalance to training: the AVLLM's audio behavior strongly matches its vision-language base model, indicating limited additional alignment to audio supervision. Our findings reveal a fundamental modality bias in AVLLMs and provide new mechanistic insights into how multimodal LLMs integrate audio and vision.
Abstract (translated)
视听大语言模型（AVLLMs）正逐渐成为多模态感知的统一接口。我们首次对AVLLMs进行了机制可解释性研究，分析了音频和视觉特征如何通过AVLLM的不同层演变与融合，以产生最终文本输出。我们发现，尽管AVLLMs在中间层编码了丰富的音频语义信息，但当音频与视觉信息冲突时，这些能力在最终文本生成中 largely 未能显现。探测分析表明，有用的潜在音频信息确实存在，但深层融合层过度偏向视觉表征，从而倾向于压制音频线索。我们进一步将这种不平衡追溯至训练过程：AVLLM的音频行为与其视觉-语言基础模型高度匹配，表明其对音频监督的额外对齐有限。我们的研究揭示了AVLLMs中存在的根本性模态偏差，并为多模态大语言模型如何整合音频与视觉提供了新的机制性见解。
URL
https://arxiv.org/abs/2604.02605
PDF
-
Dependency-Guided Parallel Decoding in Discrete Diffusion Language Models
2026-04-02 22:21:24Liran Ringel, Ameen Ali, Yaniv RomanoAbstract
Discrete diffusion language models (dLLMs) accelerate text generation by unmasking multiple tokens in parallel. However, parallel decoding introduces a distributional mismatch: it approximates the joint conditional using a fully factorized product of per-token marginals, which degrades output quality when selected tokens are strongly dependent. We propose DEMASK (DEpendency-guided unMASKing), a lightweight dependency predictor that attaches to the final hidden states of a dLLM. In a single forward pass, it estimates pairwise conditional influences between masked positions. Using these predictions, a greedy selection algorithm identifies positions with bounded cumulative dependency for simultaneous unmasking. Under a sub-additivity assumption, we prove this bounds the total variation distance between our parallel sampling and the model's joint. Empirically, DEMASK achieves 1.7-2.2$\times$ speedup on Dream-7B while matching or improving accuracy compared to confidence-based and KL-based baselines.
Abstract (translated)
离散扩散语言模型（dLLMs）通过并行解掩多个词元来加速文本生成。然而，并行解码会引入分布失配问题：它使用完全因式分解的逐词元边缘乘积来近似联合条件分布，当所选词元高度相关时，这会降低输出质量。我们提出了DEMASK（基于依赖引导的解掩码），一种轻量级依赖预测器，它附加在dLLM的最终隐藏状态上。在单次前向传播中，它估计掩码位置之间的成对条件影响。利用这些预测，贪心选择算法可识别累积依赖有界的同步解掩位置。在次可加性假设下，我们证明了这限制了并行采样与模型联合分布之间的总变差距离。实验表明，DEMASK在Dream-7B上实现了1.7-2.2倍加速，同时在准确率上匹配或优于基于置信度和KL散度的基线方法。
URL
https://arxiv.org/abs/2604.02560
PDF
-
Pragmatics Meets Culture: Culturally-adapted Artwork Description Generation and Evaluation
2026-04-02 22:16:07Lingjun Zhao, Dayeon Ki, Marine Carpuat, Hal Daum\'e IIIAbstract
Language models are known to exhibit various forms of cultural bias in decision-making tasks, yet much less is known about their degree of cultural familiarity in open-ended text generation tasks. In this paper, we introduce the task of culturally-adapted art description generation, where models describe artworks for audiences from different cultural groups who vary in their familiarity with the cultural symbols and narratives embedded in the artwork. To evaluate cultural competence in this pragmatic generation task, we propose a framework based on culturally grounded question answering. We find that base models are only marginally adequate for this task, but, through a pragmatic speaker model, we can improve simulated listener comprehension by up to 8.2%. A human study further confirms that the model with higher pragmatic competence is rated as more helpful for comprehension by 8.0%.
Abstract (translated)
语言模型在决策任务中已知存在多种形式的文化偏见，但关于其在开放式文本生成任务中的文化熟悉度却鲜有研究。本文提出文化适应艺术描述生成任务，要求模型为具有不同文化背景的受众描述艺术品，这些受众对作品中嵌入的文化符号与叙事熟悉程度各异。为评估该语用生成任务中的文化能力，我们提出一个基于文化 grounded 问答的评估框架。研究发现基础模型仅勉强适用于该任务，但通过语用说话人模型可将模拟听者理解度提升最多8.2%。人类研究进一步证实，具备更高语用能力的模型在帮助理解方面获得评价提升8.0%。
URL
https://arxiv.org/abs/2604.02557
PDF
-
Expert-Choice Routing Enables Adaptive Computation in Diffusion Language Models
2026-04-02 05:01:36Shuibai Zhang, Caspian Zhuang, Chihan Cui, Zhihan Yang, Fred Zhangzhi Peng, Yanxin Zhang, Haoyue Bai, Zack Jia, Yang Zhou, Guanhua Chen, Ming LiuAbstract
Diffusion language models (DLMs) enable parallel, non-autoregressive text generation, yet existing DLM mixture-of-experts (MoE) models inherit token-choice (TC) routing from autoregressive systems, leading to load imbalance and rigid computation allocation. We show that expert-choice (EC) routing is a better fit for DLMs: it provides deterministic load balancing by design, yielding higher throughput and faster convergence than TC. Building on the property that EC capacity is externally controllable, we introduce timestep-dependent expert capacity, which varies expert allocation according to the denoising step. We find that allocating more capacity to low-mask-ratio steps consistently achieves the best performance under matched FLOPs, and provide a mechanistic explanation: tokens in low-mask-ratio contexts exhibit an order-of-magnitude higher learning efficiency, so concentrating compute on these steps yields the largest marginal return. Finally, we show that existing pretrained TC DLMs can be retrofitted to EC by replacing only the router, achieving faster convergence and improved accuracy across diverse downstream tasks. Together, these results establish EC routing as a superior paradigm for DLM MoE models and demonstrate that computation in DLMs can be treated as an adaptive policy rather than a fixed architectural constant. Code is available at this https URL.
Abstract (translated)
扩散语言模型（DLMs）支持并行、非自回归的文本生成，但现有的DLM混合专家（MoE）模型沿用了自回归系统中的令牌选择（TC）路由机制，导致负载不均衡和计算分配僵化。我们证明专家选择（EC）路由更适用于DLMs：其设计上即提供确定性负载均衡，相比TC能实现更高吞吐量和更快的收敛。基于EC容量可外部调控的特性，我们引入了随时间步变化的专家容量，使专家分配能根据去噪步骤动态调整。研究发现，在相同浮点运算量（FLOPs）下，为低掩码比步骤分配更多容量 consistently 能实现最佳性能，并提供机理解释：低掩码比上下文中的令牌学习效率高出一个数量级，因此将计算集中于这些步骤能获得最大边际收益。最后，我们证明现有预训练的TC DLMs仅通过替换路由器即可改造为EC路由，在多种下游任务中实现更快收敛和更高精度。综上，这些结果确立了EC路由作为DLM MoE模型的优越范式，并表明DLM中的计算可被视为自适应策略而非固定架构常量。代码已公开于此https链接。
URL
https://arxiv.org/abs/2604.01622
PDF
-
The Ultimate Tutorial for AI-driven Scale Development in Generative Psychometrics: Releasing AIGENIE from its Bottle
2026-03-30 16:25:37Lara Russell-Lasalandra, Hudson Golino, Luis Eduardo Garrido, Alexander P. ChristensenAbstract
Psychological scale development has traditionally required extensive expert involvement, iterative revision, and large-scale pilot testing before psychometric evaluation can begin. The `AIGENIE` R package implements the AI-GENIE framework (Automatic Item Generation with Network-Integrated Evaluation), which integrates large language model (LLM) text generation with network psychometric methods to automate the early stages of this process. The package generates candidate item pools using LLMs, transforms them into high-dimensional embeddings, and applies a multi-step reduction pipeline -- Exploratory Graph Analysis (EGA), Unique Variable Analysis (UVA), and bootstrap EGA -- to produce structurally validated item pools entirely *in silico*. This tutorial introduces the package across six parts: installation and setup, understanding Application Programming Interfaces (APIs), text generation, item generation, the `AIGENIE` function, and the `GENIE` function. Two running examples illustrate the package's use: the Big Five personality model (a well-established construct) and AI Anxiety (an emerging construct). The package supports multiple LLM providers (OpenAI, Anthropic, Groq, HuggingFace, and local models), offers a fully offline mode with no external API calls, and provides the `GENIE()` function for researchers who wish to apply the psychometric reduction pipeline to existing item pools regardless of their origin. The `AIGENIE` package is freely available on R-universe at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2603.28643
PDF
-
EnsemJudge: Enhancing Reliability in Chinese LLM-Generated Text Detection through Diverse Model Ensembles
2026-03-30 01:50:46Zhuoshang Wang, Yubing Ren, Guoyu Zhao, Xiaowei Zhu, Hao Li, Yanan CaoAbstract
Large Language Models (LLMs) are widely applied across various domains due to their powerful text generation capabilities. While LLM-generated texts often resemble human-written ones, their misuse can lead to significant societal risks. Detecting such texts is an essential technique for mitigating LLM misuse, and many detection methods have shown promising results across different datasets. However, real-world scenarios often involve out-of-domain inputs or adversarial samples, which can affect the performance of detection methods to varying degrees. Furthermore, most existing research has focused on English texts, with limited work addressing Chinese text detection. In this study, we propose EnsemJudge, a robust framework for detecting Chinese LLM-generated text by incorporating tailored strategies and ensemble voting mechanisms. We trained and evaluated our system on a carefully constructed Chinese dataset provided by NLPCC2025 Shared Task 1. Our approach outperformed all baseline methods and achieved first place in the task, demonstrating its effectiveness and reliability in Chinese LLM-generated text detection. Our code is available at this https URL.
Abstract (translated)
大语言模型（LLMs）凭借其强大的文本生成能力，已被广泛应用于各个领域。尽管LLM生成的文本常与人类写作难以区分，但其滥用可能引发重大社会风险。检测此类文本是缓解LLM滥用的关键技术，现有多种检测方法在不同数据集上已展现良好效果。然而，真实场景常涉及域外输入或对抗样本，这会在不同程度上影响检测方法的性能。此外，现有研究多聚焦于英文文本，针对中文文本检测的工作相对有限。本研究提出EnsemJudge——一个通过整合定制策略与集成投票机制来检测中文LLM生成文本的鲁棒框架。我们在NLPCC2025共享任务1提供的精心构建的中文数据集上训练并评估了该系统。我们的方法超越了所有基线方法，并在该任务中获得第一名，证明了其在中文LLM生成文本检测中的有效性与可靠性。代码已公开于此https链接。
URL
https://arxiv.org/abs/2603.27949
PDF
-
TailNLG: A Multilingual Benchmark Addressing Verbalization of Long-Tail Entities
2026-03-29 17:01:54Lia Draetta, Michael Oliverio, Virginia Ram\'on-Ferrer, Pier Felice Balestrucci, Flaviana Corallo, Carlos Badenes-Olmedo, Alessandro Mazzei, Marco Antonio Stranisci, Rossana DamianoAbstract
The automatic verbalization of structured knowledge is a key task for making knowledge graphs accessible to non-expert users and supporting retrieval-augmented generation systems. Although recent advances in Data-to-Text generation have improved multilingual coverage, little attention has been paid to potential biases in the verbalization of rare entities, frequently known as long-tail entities. In this work, we present the first systematic study of long-tail entities in Data-to-Text generation. We introduce TailNLG, a new multilingual benchmark in English, Italian, and Spanish, built from Wikidata and covering entities with varying levels of popularity. We evaluate three different families of large language models in zero-shot settings and compare their performance on rare versus common entities, as well as against the established WebNLG benchmark. Our results reveal a consistent bias against long-tail entities: embedding-based scores are lower, and model uncertainty is higher for rare entities. We further show that the impact of long-tail entities varies across models and languages, and that existing evaluation metrics do not consistently capture these differences, highlighting the need for more reliable evaluation frameworks.
Abstract (translated)
结构化知识的自动叙述化是让非专业用户访问知识图谱并支持检索增强生成系统的关键任务。尽管近期数据到文本生成技术的进步提升了多语言覆盖范围，但针对稀有实体（通常称为长尾实体）叙述化中潜在偏差的研究尚未受到足够关注。本研究首次系统性地探讨了数据到文本生成中的长尾实体问题。我们提出了TailNLG——一个基于维基数据构建的新的多语言基准测试集，涵盖英语、意大利语和西班牙语，并包含不同流行度级别的实体。我们在零样本设置下评估了三类大型语言模型，比较了它们在稀有与常见实体上的表现，并与成熟的WebNLG基准测试进行了对比。研究结果揭示了针对长尾实体的持续性偏差：稀有实体的基于嵌入的评分更低，模型不确定性更高。我们还发现长尾实体的影响因模型和语言而异，且现有评估指标未能一致捕捉这些差异，这凸显了建立更可靠评估框架的必要性。
URL
https://arxiv.org/abs/2603.27768
PDF
-
Can AI Models Direct Each Other? Organizational Structure as a Probe into Training Limitations
Abstract
Can an expensive AI model effectively direct a cheap one to solve software engineering tasks? We study this question by introducing ManagerWorker, a two-agent pipeline where an expensive "manager" model (text-only, no code execution) analyzes issues, dispatches exploration tasks, and reviews implementations, while a cheap "worker" model (with full repo access) executes code changes. We evaluate on 200 instances from SWE-bench Lite across five configurations that vary the manager-worker relationship, pipeline complexity, and model pairing. Our findings reveal both the promise and the limits of multi-agent direction: (1) a strong manager directing a weak worker (62%) matches a strong single agent (60%) at a fraction of the strong-model token usage, showing that expensive reasoning can substitute for expensive execution; (2) a weak manager directing a weak worker (42%) performs worse than the weak agent alone (44%), demonstrating that the directing relationship requires a genuine capability gap--structure without substance is pure overhead; (3) the manager's value lies in directing, not merely reviewing--a minimal review-only loop adds just 2pp over the baseline, while structured exploration and planning add 11pp, showing that active direction is what makes the capability gap productive; and (4) these behaviors trace to a single root cause: current models are trained as monolithic agents, and splitting them into director/worker roles fights their training distribution. The pipeline succeeds by designing around this mismatch--keeping each model close to its trained mode (text generation for the manager, tool use for the worker) and externalizing organizational structure to code. This diagnosis points to concrete training gaps: delegation, scoped execution, and mode switching are skills absent from current training data.
Abstract (translated)
URL
https://arxiv.org/abs/2603.26458
PDF
-
Automating Clinical Information Retrieval from Finnish Electronic Health Records Using Large Language Models
2026-03-27 14:03:51Mikko Saukkoriipi, Nicole Hernandez, Jaakko Sahlsten, Kimmo Kaski, Otso ArponenAbstract
Clinicians often need to retrieve patient-specific information from electronic health records (EHRs), a task that is time-consuming and error-prone. We present a locally deployable Clinical Contextual Question Answering (CCQA) framework that answers clinical questions directly from EHRs without external data transfer. Open-source large language models (LLMs) ranging from 4B to 70B parameters were benchmarked under fully offline conditions using 1,664 expert-annotated question-answer pairs derived from records of 183 patients. The dataset consisted predominantly of Finnish clinical text. In free-text generation, Llama-3.1-70B achieved 95.3% accuracy and 97.3% consistency across semantically equivalent question variants, while the smaller Qwen3-30B-A3B-2507 model achieved comparable performance. In a multiple-choice setting, models showed similar accuracy but variable calibration. Low-precision quantization (4-bit and 8-bit) preserved predictive performance while reducing GPU memory requirements and improving deployment feasibility. Clinical evaluation identified clinically significant errors in 2.9% of outputs, and semantically equivalent questions occasionally yielded discordant responses, including instances where one formulation was correct and the other contained a clinically significant error (0.96% of cases). These findings demonstrate that locally hosted open-source LLMs can accurately retrieve patient-specific information from EHRs using natural-language queries, while highlighting the need for validation and human oversight in clinical deployment.
Abstract (translated)
临床医生经常需要从电子健康记录（EHR）中检索患者特定信息，这一任务耗时且容易出错。我们提出了一种可本地部署的临床上下文问答（CCQA）框架，该框架无需外部数据传输即可直接从EHR中回答临床问题。研究在完全离线条件下，使用来自183名患者记录的1664对专家标注问答数据，对参数规模从4B到70B的开源大语言模型（LLM）进行了基准测试。数据集主要由芬兰临床文本构成。在自由文本生成任务中，Llama-3.1-70B在语义等价的问题变体上达到了95.3%的准确率和97.3%的一致性，而较小的Qwen3-30B-A3B-2507模型也实现了 comparable performance。在多项选择题设置中，模型表现出相似的准确率但校准程度各异。低精度量化（4位和8位）在保持预测性能的同时降低了GPU内存需求，提升了部署可行性。临床评估发现2.9%的输出存在临床显著错误，且语义等价问题偶尔会产生矛盾响应——包括一种表述正确而另一种包含临床显著错误的情况（占0.96%）。这些 findings表明，本地托管的开源LLM能够使用自然语言查询从EHR中准确检索患者特定信息，同时也凸显了临床部署中验证和人工监督的必要性。
URL
https://arxiv.org/abs/2603.26434
PDF
-
A Formal Framework for Uncertainty Analysis of Text Generation with Large Language Models
2026-03-27 12:44:37Steffen Herbold, Florian LemmerichAbstract
The generation of texts using Large Language Models (LLMs) is inherently uncertain, with sources of uncertainty being not only the generation of texts, but also the prompt used and the downstream interpretation. Within this work, we provide a formal framework for the measurement of uncertainty that takes these different aspects into account. Our framework models prompting, generation, and interpretation as interconnected autoregressive processes that can be combined into a single sampling tree. We introduce filters and objective functions to describe how different aspects of uncertainty can be expressed over the sampling tree and demonstrate how to express existing approaches towards uncertainty through these functions. With our framework we show not only how different methods are formally related and can be reduced to a common core, but also point out additional aspects of uncertainty that have not yet been studied.
Abstract (translated)
大语言模型（LLMs）生成文本具有固有不确定性，其不确定性来源不仅在于文本生成本身，还涉及所使用的提示词及下游解释。本研究提供了一个形式化框架用于衡量此类不确定性，该框架充分考虑了上述不同维度。我们将提示、生成与解释建模为相互关联的自回归过程，这些过程可整合为单一的采样树。通过引入过滤器和目标函数，我们描述了如何在采样树上表达不确定性的不同方面，并演示了如何通过这些函数呈现现有不确定性处理方法。本框架不仅揭示了不同方法在形式上的关联性及其可归结为共同核心，还指出了尚未被研究的额外不确定性维度。
URL
https://arxiv.org/abs/2603.26363
PDF
-
Towards GUI Agents: Vision-Language Diffusion Models for GUI Grounding
2026-03-27 09:34:07Shrinidhi Kumbhar, Haofu Liao, Srikar Appalaraju, Kunwar Yashraj SinghAbstract
Autoregressive (AR) vision-language models (VLMs) have long dominated multimodal understanding, reasoning, and graphical user interface (GUI) grounding. Recently, discrete diffusion vision-language models (DVLMs) have shown strong performance in multimodal reasoning, offering bidirectional attention, parallel token generation, and iterative refinement. However, their potential for GUI grounding remains unexplored. In this work, we evaluate whether discrete DVLMs can serve as a viable alternative to AR models for GUI grounding. We adapt LLaDA-V for single-turn action and bounding-box prediction, framing the task as text generation from multimodal input. To better capture the hierarchical structure of bounding-box geometry, we propose a hybrid masking schedule that combines linear and deterministic masking, improving grounding accuracy by up to 6.1 points in Step Success Rate (SSR) over the GUI-adapted LLaDA-V trained with linear masking. Evaluations on four datasets spanning web, desktop, and mobile interfaces show that the adapted diffusion model with hybrid masking consistently outperforms the linear-masked variant and performs competitively with autoregressive counterparts despite limited pretraining. Systematic ablations reveal that increasing diffusion steps, generation length, and block length improves accuracy but also increases latency, with accuracy plateauing beyond a certain number of diffusion steps. Expanding the training data with diverse GUI domains further reduces latency by about 1.3 seconds and improves grounding accuracy by an average of 20 points across benchmarks. These results demonstrate that discrete DVLMs are a promising modeling framework for GUI grounding and represent an important step toward diffusion-based GUI agents.
Abstract (translated)
自回归（AR）视觉语言模型（VLMs）长期占据多模态理解、推理及图形用户界面（GUI）定位的主导地位。近期，离散扩散视觉语言模型（DVLMs）在多模态推理中展现出强劲性能，具备双向注意力、并行令牌生成与迭代优化等优势。然而，其在GUI定位领域的潜力尚未得到探索。本研究评估离散DVLMs能否作为AR模型在GUI定位任务中的可行替代方案。我们针对单轮动作与边界框预测任务适配LLaDA-V模型，将任务框架化为多模态输入的文本生成。为更好捕捉边界框几何的层次化结构，我们提出一种结合线性与确定性掩码的混合掩码调度策略，相较于采用线性掩码的GUI适配LLaDA-V，其定位精度在步骤成功率（SSR）指标上最高提升6.1个百分点。在涵盖网页、桌面及移动界面的四个数据集上的评估显示，采用混合掩码的扩散模型持续优于线性掩码变体，并在有限预训练条件下达到与自回归模型相当的性能。系统化消融实验表明，增加扩散步数、生成长度与块长度可提升精度，但也会增加延迟，且精度在扩散步数超过一定阈值后趋于饱和。通过扩展涵盖多样化GUI领域的训练数据，可将延迟进一步降低约1.3秒，并在各基准测试中平均提升20个定位精度点。这些结果证明离散DVLMs是GUI定位任务中极具潜力的建模框架，并为基于扩散模型的GUI智能体迈出了重要一步。
URL
https://arxiv.org/abs/2603.26211
PDF
-
Supercharging Federated Intelligence Retrieval
2026-03-26 12:23:53Dimitris Stripelis, Patrick Foley, Mohammad Naseri, William Lindskog-M\"unzing, Chong Shen Ng, Daniel Janes Beutel, Nicholas D. LaneAbstract
RAG typically assumes centralized access to documents, which breaks down when knowledge is distributed across private data silos. We propose a secure Federated RAG system built using Flower that performs local silo retrieval, while server-side aggregation and text generation run inside an attested, confidential compute environment, enabling confidential remote LLM inference even in the presence of honest-but-curious or compromised servers. We also propose a cascading inference approach that incorporates a non-confidential third-party model (e.g., Amazon Nova) as auxiliary context without weakening confidentiality.
Abstract (translated)
RAG通常假设文档集中访问，但当知识分布于私有数据孤岛时，该假设便会失效。我们提出一种基于Flower构建的安全联邦RAG系统，该系统在本地数据孤岛执行检索，而服务器端的聚合与文本生成则在经认证的机密计算环境中运行，从而在存在诚实但好奇或被入侵的服务器时，仍能实现机密远程大语言模型推理。我们还提出一种级联推理方法，该方法将非机密第三方模型（如Amazon Nova）作为辅助上下文融入系统，且不会削弱机密性。
URL
https://arxiv.org/abs/2603.25374
PDF
-
ReCUBE: Evaluating Repository-Level Context Utilization in Code Generation
2026-03-26 08:04:15Jiseung Hong, Benjamin G. Ascoli, Jinho D. ChoiAbstract
Large Language Models (LLMs) have recently emerged as capable coding assistants that operate over large codebases through either agentic exploration or full-context generation. Existing benchmarks capture a broad range of coding capabilities, such as resolving GitHub issues, but none of them directly isolate and measure how effectively LLMs leverage repository-level context during code generation. To address this, we introduce ReCUBE, a benchmark in which LLMs reconstruct a masked file within a real-world repository, using all remaining source files, dependency specifications, and documentation as their only source of context. ReCUBE evaluates reconstructed code with usage-aware test cases that simulate both internal module logic and external cross-file integration, reflecting real-world software usage patterns. We further propose the Caller-Centric Exploration (CCE) toolkit, a set of dependency graph-based tools that can be integrated into agentic frameworks to guide agents toward the most relevant caller files during repository exploration. Experiments across eight models in four settings show that repository-level context utilization remains highly challenging even for state-of-the-art models, with GPT-5 achieving only 37.57% strict pass rate in the full-context setting. Agents augmented with our CCE toolkit consistently outperform all baselines across all evaluated models, with improvements of up to 7.56% in strict pass rate. We release our benchmark, code, and evaluation framework as open source for the NLP research community.
Abstract (translated)
大语言模型（LLMs）已作为能操作大型代码库的编码助手出现，其通过智能体探索或全上下文生成的方式工作。现有基准测试虽能评估广泛的编码能力（如解决GitHub问题），但均未直接分离和衡量LLMs在代码生成中利用仓库级上下文的效果。为填补此空白，我们提出ReCUBE——一个让LLMs利用剩余源代码文件、依赖规范及文档作为唯一上下文，在真实仓库中重建被掩码文件的基准测试。ReCUBE通过感知代码使用场景的测试用例评估重建结果，这些用例既模拟内部模块逻辑也模拟跨文件集成，反映真实软件使用模式。我们还提出调用者中心探索（CCE）工具包，这是一套基于依赖图的工具，可集成至智能体框架以引导其在仓库探索时聚焦最相关的调用文件。在四种设置下对八个模型的实验表明，即使是最先进的模型，仓库级上下文利用仍极具挑战性——GPT-5在全上下文设置下仅达到37.57%的严格通过率。集成CCE工具包的智能体在所有评估模型中持续超越所有基线，严格通过率最高提升7.56%。我们已将基准测试、代码及评估框架开源，供自然语言处理研究社区使用。
URL
https://arxiv.org/abs/2603.25770
PDF
-
Argument Mining as a Text-to-Text Generation Task
2026-03-25 05:10:39Masayuki Kawarada, Tsutomu Hirao, Wataru Uchida, Masaaki NagataAbstract
Argument Mining(AM) aims to uncover the argumentative structures within a text. Previous methods require several subtasks, such as span identification, component classification, and relation classification. Consequently, these methods need rule-based postprocessing to derive argumentative structures from the output of each subtask. This approach adds to the complexity of the model and expands the search space of the hyperparameters. To address this difficulty, we propose a simple yet strong method based on a text-to-text generation approach using a pretrained encoder-decoder language model. Our method simultaneously generates argumentatively annotated text for spans, components, and relations, eliminating the need for task-specific postprocessing and hyperparameter tuning. Furthermore, because it is a straightforward text-to-text generation method, we can easily adapt our approach to various types of argumentative structures. Experimental results demonstrate the effectiveness of our method, as it achieves state-of-the-art performance on three different types of benchmark datasets: the Argument-annotated Essays Corpus(AAEC), AbstRCT, and the Cornell eRulemaking Corpus(CDCP)
Abstract (translated)
论辩挖掘（Argument Mining, AM）旨在揭示文本中的论辩结构。以往方法需执行多个子任务（如跨度识别、组件分类与关系分类），因此依赖基于规则的后处理步骤，从各子任务输出中推导出论辩结构。该方法增加了模型复杂性并扩大了超参数搜索空间。为应对此挑战，我们提出一种基于文本到文本生成框架的简洁而有效的方法，利用预训练编码器-解码器语言模型实现。该方法可同时生成标注跨度、组件及关系的论文化文本，从而无需任务特定的后处理与超参数调优。此外，由于采用直接的文本到文本生成方式，该方法能便捷适配多种类型的论辩结构。实验结果表明，本方法在三个不同基准数据集上均取得最先进性能：论文化作文语料库（Argument-annotated Essays Corpus, AAEC）、AbstRCT及康奈尔电子立法语料库（Cornell eRulemaking Corpus, CDCP）。
URL
https://arxiv.org/abs/2603.23949
PDF
-
Parametric Knowledge and Retrieval Behavior in RAG Fine-Tuning for Electronic Design Automation
2026-03-24 10:33:57Julian Oestreich, Maximilian Bley, Frank Binder, Lydia M\"uller, Maksym Sydorenko, Andr\'e AlcaldeAbstract
Retrieval-Augmented Generation (RAG) fine-tuning has shown substantial improvements over vanilla RAG, yet most studies target document question answering and often rely on standard NLP metrics that can obscure factual differences. We evaluate RAG fine-tuning for long-form text generation in electronic design automation, adapting a 7B model under five context augmentation strategies with varying retrieval conditions. We introduce TriFEX, a human-validated, triple-based evaluation pipeline that attributes generated claims to their origin-user query, context and reference-and propose Parametric Knowledge Precision (PKP), which isolates internalized knowledge by filtering out claims leaked in the prompt. We show that ROUGE and BERTScore fail to detect factual differences that our triple-based evaluation reveals. Additionally, we demonstrate that an existing metric for knowledge internalization is retrieva-sensitive, with about 75% of its cross-condition variance driven by changes in the rate at which internal knowledge is expressed (PR), rather than by changes in its actual correctness (PKP). The fine-tuned 7B variants outperform a 72B baseline on most metrics, further showing generalization across conditions and on a related benchmark. These results underscore the limitations of available metrics in RAG evaluation and show that smaller models could be reasonably well adapted to specialized tasks for cost-efficient, on-premises deployment.
Abstract (translated)
检索增强生成（RAG）微调相较于原始RAG已展现出显著提升，但现有研究多聚焦于文档问答任务，且常依赖可能掩盖事实差异的标准自然语言处理指标。本研究评估电子设计自动化领域的长文本生成场景下的RAG微调效果，采用五种不同检索条件的上下文增强策略对70亿参数模型进行适配。提出经人工验证的三元组评估流程TriFEX，可将生成声明溯源至原始用户查询、上下文及参考依据；并设计参数化知识精确度（PKP）指标，通过过滤提示中泄露的内容来隔离模型内部知识。研究发现ROUGE与BERTScore无法检测出三元组评估揭示的事实差异，且现有知识内化指标对检索敏感——其跨条件约75%的方差源于内部知识表达率（PR）变化，而非实际正确性（PKP）变化。经微调的70亿参数模型在多数指标上超越720亿参数基线，并在跨条件及关联基准测试中展现泛化能力。这些结果凸显了当前RAG评估指标的局限性，并表明较小模型可通过合理适配在专业任务中实现成本效益更高的本地部署。
URL
https://arxiv.org/abs/2603.23047
PDF
-
LLM-guided headline rewriting for clickability enhancement without clickbait
2026-03-23 18:25:48Yehudit Aperstein, Linoy Halifa, Sagiv Bar, Alexander ApartsinAbstract
Enhancing reader engagement while preserving informational fidelity is a central challenge in controllable text generation for news media. Optimizing news headlines for reader engagement is often conflated with clickbait, resulting in exaggerated or misleading phrasing that undermines editorial trust. We frame clickbait not as a separate stylistic category, but as an extreme outcome of disproportionate amplification of otherwise legitimate engagement cues. Based on this view, we formulate headline rewriting as a controllable generation problem, where specific engagement-oriented linguistic attributes are selectively strengthened under explicit constraints on semantic faithfulness and proportional emphasis. We present a guided headline rewriting framework built on a large language model (LLM) that uses the Future Discriminators for Generation (FUDGE) paradigm for inference-time control. The LLM is steered by two auxiliary guide models: (1) a clickbait scoring model that provides negative guidance to suppress excessive stylistic amplification, and (2) an engagement-attribute model that provides positive guidance aligned with target clickability objectives. Both guides are trained on neutral headlines drawn from a curated real-world news corpus. At the same time, clickbait variants are generated synthetically by rewriting these original headlines using an LLM under controlled activation of predefined engagement tactics. By adjusting guidance weights at inference time, the system generates headlines along a continuum from neutral paraphrases to more engaging yet editorially acceptable formulations. The proposed framework provides a principled approach for studying the trade-off between attractiveness, semantic preservation, and clickbait avoidance, and supports responsible LLM-based headline optimization in journalistic settings.
Abstract (translated)
在保持信息保真度的同时提升读者参与度，是新闻媒体可控文本生成的核心挑战。优化新闻标题以吸引读者常与点击诱饵混为一谈，导致夸张或误导性表述，从而损害编辑公信力。本研究将点击诱饵视为一种独立文体类别，而是 legitimate engagement cues 被不成比例放大的极端结果。基于此视角，我们将标题重写构建为可控生成问题：在语义忠实度与比例性强调的明确约束下，选择性地强化特定面向参与度的语言特征。我们提出一种基于大语言模型（LLM）的引导式标题重写框架，采用 Future Discriminators for Generation（FUDGE）范式实现推理时控制。该LLM由两个辅助引导模型驱动：（1）点击诱饵评分模型提供负向引导，抑制过度的文体放大；（2）参与度属性模型提供与目标点击率对齐的正向引导。两模型均基于精选的真实新闻语料库中的中性标题训练。同时，通过LLM在受控激活预定义参与策略条件下重写原始标题，合成生成点击诱饵变体。通过调整推理时的引导权重，系统可在中性释义与更具吸引力且编辑可接受的表述之间生成连续谱系的标题。该框架为研究吸引力、语义保持与点击诱饵规避之间的权衡提供了原则性方法，并支持新闻场景中基于LLM的负责任标题优化。
URL
https://arxiv.org/abs/2603.22459
PDF
-
Gumbel Distillation for Parallel Text Generation
2026-03-23 17:13:22Chi Zhang, Xixi Hu, Bo Liu, Qiang LiuAbstract
The slow, sequential nature of autoregressive (AR) language models has driven the adoption of parallel decoding methods. However, these non-AR models often sacrifice generation quality as they struggle to model the complex joint distribution of token sequences. To narrow this performance gap, we introduce Gumbel Distillation, a novel distillation technique that enables parallel decoders to learn this distribution effectively. Our method leverages the Gumbel-Max trick to create a deterministic mapping from a latent Gumbel noise space to the output tokens of a high-performing AR teacher. As a model-agnostic technique, Gumbel Distillation seamlessly integrates with diverse parallel decoding architectures, including MDLM and BD3-LM. Experiments on LM1B and OpenWebText show that Gumbel Distillation substantially improves the generation quality of parallel language models, achieving a 30.0% improvement in MAUVE score and 10.5% in generative perplexity over MDLM trained on OpenWebText dataset. Code available at this https URL.
Abstract (translated)
自回归（AR）语言模型的缓慢、顺序特性推动了并行解码方法的采用。然而，这些非自回归模型往往在生成质量上做出妥协，因为它们难以对标记序列的复杂联合分布进行建模。为缩小这一性能差距，我们提出了冈贝尔蒸馏（Gumbel Distillation），一种新颖的蒸馏技术，使并行解码器能够有效学习该分布。我们的方法利用冈贝尔-最大值技巧（Gumbel-Max trick），建立了从潜在冈贝尔噪声空间到高性能自回归教师模型输出标记之间的确定性映射。作为一种与模型无关的技术，冈贝尔蒸馏可无缝集成到多种并行解码架构中，包括MDLM和BD3-LM。在LM1B和OpenWebText数据集上的实验表明，冈贝尔蒸馏显著提升了并行语言模型的生成质量，在OpenWebText数据集上相较于MDLM，MAUVE分数提高30.0%，生成困惑度降低10.5%。代码已公开于此链接。
URL
https://arxiv.org/abs/2603.22216
PDF
-
Dual-Space Knowledge Distillation with Key-Query Matching for Large Language Models with Vocabulary Mismatch
2026-03-23 14:53:55Stella Eva Tsiapali, Cong-Thanh Do, Kate KnillAbstract
Large language models (LLMs) achieve state-of-the-art (SOTA) performance across language tasks, but are costly to deploy due to their size and resource demands. Knowledge Distillation (KD) addresses this by training smaller Student models to mimic larger Teacher models, improving efficiency without significant performance loss. Dual-Space Knowledge Distillation with Cross-Model Attention (DSKD-CMA) has emerged as a SOTA method for KD between LLMs with distinct tokenizers, yet its internal workings remain largely opaque. In this work, we systematically analyse the attention mechanism of DSKD-CMA through manual token alignment probing and heatmap visualisations, revealing both strengths and limitations. Building on this, we introduce a novel method, DSKD-CMA-GA, based on Generative Adversarial (GA) learning, to address the mismatched distributions between the keys and queries computed from distinct models. Experiments show modest but consistent ROUGE-L gains in text generation quality, particularly on out-of-distribution data (+0.37 on average), narrowing the gap between cross- and same-tokenizer KD.
Abstract (translated)
大型语言模型（LLMs）在各类语言任务上达到最先进（SOTA）性能，但其规模与资源需求导致部署成本高昂。知识蒸馏（KD）通过训练更小的学生模型来模仿大型教师模型，从而在不显著损失性能的前提下提升效率。针对使用不同分词器的LLM间蒸馏，基于跨模型注意力的双空间知识蒸馏（DSKD-CMA）已成为SOTA方法，但其内部机制仍 largely opaque。本研究通过人工分词对齐探测与热力图可视化，系统分析了DSKD-CMA的注意力机制，揭示了其优势与局限。在此基础上，我们提出一种基于生成对抗（GA）学习的新方法DSKD-CMA-GA，以解决由不同模型计算出的键值对分布不匹配问题。实验表明，该方法在文本生成质量上取得了 modest but consistent 的ROUGE-L提升，尤其在分布外数据上效果显著（平均+0.37），缩小了跨分词器与同分词器蒸馏之间的性能差距。
URL
https://arxiv.org/abs/2603.22056
PDF
-
FontCrafter: High-Fidelity Element-Driven Artistic Font Creation with Visual In-Context Generation
2026-03-23 14:53:12Wuyang Luo, Chengkai Tan, Chang Ge, Binye Hong, Su Yang, Yongjiu MaAbstract
Artistic font generation aims to synthesize stylized glyphs based on a reference style. However, existing approaches suffer from limited style diversity and coarse control. In this work, we explore the potential of element-driven artistic font generation. Elements are the fundamental visual units of a font, serving as reference images for the desired style. Conceptually, we categorize elements into object elements (e.g., flowers or stones) with distinct structures and amorphous elements (e.g., flames or clouds) with unstructured textures. We introduce FontCrafter, an element-driven framework for font creation, and construct a large-scale dataset, ElementFont, which contains diverse element types and high-quality glyph images. However, achieving high-fidelity reconstruction of both texture and structure of reference elements remains challenging. To address this, we propose an in-context generation strategy that treats element images as visual context and uses an inpainting model to transfer element styles into glyph regions at the pixel level. To further control glyph shapes, we design a lightweight Context-aware Mask Adapter (CMA) that injects shape information. Moreover, a training-free attention redirection mechanism enables region-aware style control and suppresses stroke hallucination. In addition, edge repainting is applied to make boundaries more natural. Extensive experiments demonstrate that FontCrafter achieves strong zero-shot generation performance, particularly in preserving structural and textural fidelity, while also supporting flexible controls such as style mixture.
Abstract (translated)
艺术字体生成旨在基于参考风格合成风格化的字形。然而，现有方法存在风格多样性有限且控制粗糙的问题。本研究探索了元素驱动的艺术字体生成潜力。元素是字体的基础视觉单元，可作为期望风格的参考图像。概念上，我们将元素分为具有明确结构的实体元素（如花朵或石头）与具有非结构化纹理的无形元素（如火焰或云朵）。我们引入了FontCrafter——一个用于字体创作的元素驱动框架，并构建了包含多样化元素类型与高质量字形图像的大规模数据集ElementFont。然而，实现参考元素纹理与结构的高保真重建仍具挑战性。为此，我们提出一种上下文生成策略，将元素图像作为视觉上下文，并利用修复模型在像素层面将元素风格迁移至字形区域。为进一步控制字形形状，我们设计了轻量级的上下文感知掩模适配器（CMA）以注入形状信息。此外，一种无需训练的注意力重定向机制可实现区域感知的风格控制并抑制笔画幻觉。同时应用边缘重绘以使边界更自然。大量实验表明，FontCrafter实现了强大的零样本生成性能，尤其在保持结构与纹理保真度方面表现突出，并支持风格混合等灵活控制。
URL
https://arxiv.org/abs/2603.22054
PDF
