Michael R. Lyu

Jen-tse Huang, Jiaxu Zhou, Tailin Jin et al. · allen-ai, cmu

Large language model-based multi-agent systems have shown great abilities across various tasks due to the collaboration of expert agents, each focusing on a specific domain. However, the impact of clumsy or even malicious agents--those who frequently make errors in their tasks--on the overall performance of the system remains underexplored. This paper investigates: (1) What is the resilience of various system structures (e.g., A$\rightarrow$B$\rightarrow$C, A$\leftrightarrow$B$\leftrightarrow$C) under faulty agents, on different downstream tasks? (2) How can we increase system resilience to defend against these agents? To simulate faulty agents, we propose two approaches--AutoTransform and AutoInject--which introduce mistakes into the agents' responses. Experiments on four downstream tasks using six systems show that the "hierarchical" structure, i.e., A$\rightarrow$(B$\leftrightarrow$C), exhibits superior resilience with the lowest performance drop of 5.5%, compared to 10.5% and 23.7% of other two structures. To further improve resilience, we introduce (1) Challenger, that introduces a mechanism for each agent to challenge others' outputs, and (2) Inspector, an additional agent to review and correct messages, recovering up to 96.4% errors made by faulty agents. Our code and data are available at https://github.com/CUHK-ARISE/MAS-Resilience.

Jen-tse Huang, Man Ho Lam, Eric John Li et al. · pku, tencent-ai

Evaluating Large Language Models' (LLMs) anthropomorphic capabilities has become increasingly important in contemporary discourse. Utilizing the emotion appraisal theory from psychology, we propose to evaluate the empathy ability of LLMs, i.e., how their feelings change when presented with specific situations. After a careful and comprehensive survey, we collect a dataset containing over 400 situations that have proven effective in eliciting the eight emotions central to our study. Categorizing the situations into 36 factors, we conduct a human evaluation involving more than 1,200 subjects worldwide. With the human evaluation results as references, our evaluation includes seven LLMs, covering both commercial and open-source models, including variations in model sizes, featuring the latest iterations, such as GPT-4, Mixtral-8x22B, and LLaMA-3.1. We find that, despite several misalignments, LLMs can generally respond appropriately to certain situations. Nevertheless, they fall short in alignment with the emotional behaviors of human beings and cannot establish connections between similar situations. Our collected dataset of situations, the human evaluation results, and the code of our testing framework, i.e., EmotionBench, are publicly available at https://github.com/CUHK-ARISE/EmotionBench.

Wenxuan Wang, Zhaopeng Tu, Chang Chen et al. · pku, tencent-ai

Safety lies at the core of developing and deploying large language models (LLMs). However, previous safety benchmarks only concern the safety in one language, e.g. the majority language in the pretraining data such as English. In this work, we build the first multilingual safety benchmark for LLMs, XSafety, in response to the global deployment of LLMs in practice. XSafety covers 14 kinds of commonly used safety issues across 10 languages that span several language families. We utilize XSafety to empirically study the multilingual safety for 4 widely-used LLMs, including both close-API and open-source models. Experimental results show that all LLMs produce significantly more unsafe responses for non-English queries than English ones, indicating the necessity of developing safety alignment for non-English languages. In addition, we propose several simple and effective prompting methods to improve the multilingual safety of ChatGPT by evoking safety knowledge and improving cross-lingual generalization of safety alignment. Our prompting method can significantly reduce the ratio of unsafe responses from 19.1% to 9.7% for non-English queries. We release our data at https://github.com/Jarviswang94/Multilingual_safety_benchmark.

Jen-tse Huang, Wenxuan Wang, Eric John Li et al. · pku, tencent-ai

Large Language Models (LLMs) have recently showcased their remarkable capacities, not only in natural language processing tasks but also across diverse domains such as clinical medicine, legal consultation, and education. LLMs become more than mere applications, evolving into assistants capable of addressing diverse user requests. This narrows the distinction between human beings and artificial intelligence agents, raising intriguing questions regarding the potential manifestation of personalities, temperaments, and emotions within LLMs. In this paper, we propose a framework, PsychoBench, for evaluating diverse psychological aspects of LLMs. Comprising thirteen scales commonly used in clinical psychology, PsychoBench further classifies these scales into four distinct categories: personality traits, interpersonal relationships, motivational tests, and emotional abilities. Our study examines five popular models, namely text-davinci-003, gpt-3.5-turbo, gpt-4, LLaMA-2-7b, and LLaMA-2-13b. Additionally, we employ a jailbreak approach to bypass the safety alignment protocols and test the intrinsic natures of LLMs. We have made PsychoBench openly accessible via https://github.com/CUHK-ARISE/PsychoBench.

Jingjing Li, Zichao Li, Tao Ge et al.

Text revision refers to a family of natural language generation tasks, where the source and target sequences share moderate resemblance in surface form but differentiate in attributes, such as text formality and simplicity. Current state-of-the-art methods formulate these tasks as sequence-to-sequence learning problems, which rely on large-scale parallel training corpus. In this paper, we present an iterative in-place editing approach for text revision, which requires no parallel data. In this approach, we simply fine-tune a pre-trained Transformer with masked language modeling and attribute classification. During inference, the editing at each iteration is realized by two-step span replacement. At the first step, the distributed representation of the text optimizes on the fly towards an attribute function. At the second step, a text span is masked and another new one is proposed conditioned on the optimized representation. The empirical experiments on two typical and important text revision tasks, text formalization and text simplification, show the effectiveness of our approach. It achieves competitive and even better performance than state-of-the-art supervised methods on text simplification, and gains better performance than strong unsupervised methods on text formalization \footnote{Code and model are available at \url{https://github.com/jingjingli01/OREO}}.

Yifan Gao, Qingyu Yin, Zheng Li et al. · amazon-science

Keyphrase generation is the task of automatically predicting keyphrases given a piece of long text. Despite its recent flourishing, keyphrase generation on non-English languages haven't been vastly investigated. In this paper, we call attention to a new setting named multilingual keyphrase generation and we contribute two new datasets, EcommerceMKP and AcademicMKP, covering six languages. Technically, we propose a retrieval-augmented method for multilingual keyphrase generation to mitigate the data shortage problem in non-English languages. The retrieval-augmented model leverages keyphrase annotations in English datasets to facilitate generating keyphrases in low-resource languages. Given a non-English passage, a cross-lingual dense passage retrieval module finds relevant English passages. Then the associated English keyphrases serve as external knowledge for keyphrase generation in the current language. Moreover, we develop a retriever-generator iterative training algorithm to mine pseudo parallel passage pairs to strengthen the cross-lingual passage retriever. Comprehensive experiments and ablations show that the proposed approach outperforms all baselines.

Jianping Zhang, Zhuoer Xu, Shiwen Cui et al.

Latent diffusion models achieve state-of-the-art performance on a variety of generative tasks, such as image synthesis and image editing. However, the robustness of latent diffusion models is not well studied. Previous works only focus on the adversarial attacks against the encoder or the output image under white-box settings, regardless of the denoising process. Therefore, in this paper, we aim to analyze the robustness of latent diffusion models more thoroughly. We first study the influence of the components inside latent diffusion models on their white-box robustness. In addition to white-box scenarios, we evaluate the black-box robustness of latent diffusion models via transfer attacks, where we consider both prompt-transfer and model-transfer settings and possible defense mechanisms. However, all these explorations need a comprehensive benchmark dataset, which is missing in the literature. Therefore, to facilitate the research of the robustness of latent diffusion models, we propose two automatic dataset construction pipelines for two kinds of image editing models and release the whole dataset. Our code and dataset are available at \url{https://github.com/jpzhang1810/LDM-Robustness}.

Jingyu Xiao, Ming Wang, Man Ho Lam et al.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in automated front-end engineering, e.g., generating UI code from visual designs. However, existing front-end UI code generation benchmarks have the following limitations: (1) While framework-based development becomes predominant in modern front-end programming, current benchmarks fail to incorporate mainstream development frameworks. (2) Existing evaluations focus solely on the UI code generation task, whereas practical UI development involves several iterations, including refining editing, and repairing issues. (3) Current benchmarks employ unidimensional evaluation, lacking investigation into influencing factors like task difficulty, input context variations, and in-depth code-level analysis. To bridge these gaps, we introduce DesignBench, a multi-framework, multi-task evaluation benchmark for assessing MLLMs' capabilities in automated front-end engineering. DesignBench encompasses three widely-used UI frameworks (React, Vue, and Angular) alongside vanilla HTML/CSS, and evaluates on three essential front-end tasks (generation, edit, and repair) in real-world development workflows. DesignBench contains 900 webpage samples spanning over 11 topics, 9 edit types, and 6 issue categories, enabling detailed analysis of MLLM performance across multiple dimensions. Our systematic evaluation reveals critical insights into MLLMs' framework-specific limitations, task-related bottlenecks, and performance variations under different conditions, providing guidance for future research in automated front-end development. Our code and data are available at https://github.com/WebPAI/DesignBench.

Zi-Yuan Hu, Yanyang Li, Michael R. Lyu et al.

As the model size of pre-trained language models (PLMs) grows rapidly, full fine-tuning becomes prohibitively expensive for model training and storage. In vision-and-language (VL), parameter-efficient tuning (PET) techniques are proposed to integrate modular modifications (e.g., Adapter and LoRA) into encoder-decoder PLMs. By tuning a small set of trainable parameters, these techniques perform on par with full fine-tuning. However, excessive modular modifications and neglecting the functionality gap between the encoders and decoders can lead to performance degradation, while existing PET techniques (e.g., VL-Adapter) overlook these critical issues. In this paper, we propose a Vision-and-Language Parameter-Efficient Tuning (VL-PET) framework to impose effective control over modular modifications via a novel granularity-controlled mechanism. Considering different granularity-controlled matrices generated by this mechanism, a variety of model-agnostic VL-PET modules can be instantiated from our framework for better efficiency and effectiveness trade-offs. We further propose lightweight PET module designs to enhance VL alignment and modeling for the encoders and maintain text generation for the decoders. Extensive experiments conducted on four image-text tasks and four video-text tasks demonstrate the efficiency, effectiveness and transferability of our VL-PET framework. In particular, our VL-PET-large with lightweight PET module designs significantly outperforms VL-Adapter by 2.92% (3.41%) and LoRA by 3.37% (7.03%) with BART-base (T5-base) on image-text tasks. Furthermore, we validate the enhanced effect of employing our VL-PET designs on existing PET techniques, enabling them to achieve significant performance improvements. Our code is available at https://github.com/HenryHZY/VL-PET.

Cheryl Lee, Tianyi Yang, Zhuangbin Chen et al.

Prompt and accurate detection of system anomalies is essential to ensure the reliability of software systems. Unlike manual efforts that exploit all available run-time information, existing approaches usually leverage only a single type of monitoring data (often logs or metrics) or fail to make effective use of the joint information among different types of data. Consequently, many false predictions occur. To better understand the manifestations of system anomalies, we conduct a systematical study on a large amount of heterogeneous data, i.e., logs and metrics. Our study demonstrates that logs and metrics can manifest system anomalies collaboratively and complementarily, and neither of them only is sufficient. Thus, integrating heterogeneous data can help recover the complete picture of a system's health status. In this context, we propose Hades, the first end-to-end semi-supervised approach to effectively identify system anomalies based on heterogeneous data. Our approach employs a hierarchical architecture to learn a global representation of the system status by fusing log semantics and metric patterns. It captures discriminative features and meaningful interactions from heterogeneous data via a cross-modal attention module, trained in a semi-supervised manner. We evaluate Hades extensively on large-scale simulated data and datasets from Huawei Cloud. The experimental results present the effectiveness of our model in detecting system anomalies. We also release the code and the annotated dataset for replication and future research.

Jianping Zhang, Jen-tse Huang, Wenxuan Wang et al. · pku, tencent-ai

Deep neural networks have achieved unprecedented success on diverse vision tasks. However, they are vulnerable to adversarial noise that is imperceptible to humans. This phenomenon negatively affects their deployment in real-world scenarios, especially security-related ones. To evaluate the robustness of a target model in practice, transfer-based attacks craft adversarial samples with a local model and have attracted increasing attention from researchers due to their high efficiency. The state-of-the-art transfer-based attacks are generally based on data augmentation, which typically augments multiple training images from a linear path when learning adversarial samples. However, such methods selected the image augmentation path heuristically and may augment images that are semantics-inconsistent with the target images, which harms the transferability of the generated adversarial samples. To overcome the pitfall, we propose the Path-Augmented Method (PAM). Specifically, PAM first constructs a candidate augmentation path pool. It then settles the employed augmentation paths during adversarial sample generation with greedy search. Furthermore, to avoid augmenting semantics-inconsistent images, we train a Semantics Predictor (SP) to constrain the length of the augmentation path. Extensive experiments confirm that PAM can achieve an improvement of over 4.8% on average compared with the state-of-the-art baselines in terms of the attack success rates.

Junjie Huang, Zhihan Jiang, Jinyang Liu et al.

Logs are imperative in the maintenance of online service systems, which often encompass important information for effective failure mitigation. While existing anomaly detection methodologies facilitate the identification of anomalous logs within extensive runtime data, manual investigation of log messages by engineers remains essential to comprehend faults, which is labor-intensive and error-prone. Upon examining the log-based troubleshooting practices at CloudA, we find that engineers typically prioritize two categories of log information for diagnosis. These include fault-indicating descriptions, which record abnormal system events, and fault-indicating parameters, which specify the associated entities. Motivated by this finding, we propose an approach to automatically extract such faultindicating information from logs for fault diagnosis, named LoFI. LoFI comprises two key stages. In the first stage, LoFI performs coarse-grained filtering to collect logs related to the faults based on semantic similarity. In the second stage, LoFI leverages a pre-trained language model with a novel prompt-based tuning method to extract fine-grained information of interest from the collected logs. We evaluate LoFI on logs collected from Apache Spark and an industrial dataset from CloudA. The experimental results demonstrate that LoFI outperforms all baseline methods by a significant margin, achieving an absolute improvement of 25.8~37.9 in F1 over the best baseline method, ChatGPT. This highlights the effectiveness of LoFI in recognizing fault-indicating information. Furthermore, the successful deployment of LoFI at CloudA and user studies validate the utility of our method. The code and data are available at https://github.com/Jun-jie-Huang/LoFI.

Wenxuan Wang, Wenxiang Jiao, Shuo Wang et al. · tsinghua

Zero-shot translation is a promising direction for building a comprehensive multilingual neural machine translation~(MNMT) system. However, its quality is still not satisfactory due to off-target issues. In this paper, we aim to understand and alleviate the off-target issues from the perspective of uncertainty in zero-shot translation. By carefully examining the translation output and model confidence, we identify two uncertainties that are responsible for the off-target issues, namely, extrinsic data uncertainty and intrinsic model uncertainty. Based on the observations, we propose two lightweight and complementary approaches to denoise the training data for model training and explicitly penalize the off-target translations by unlikelihood training during model training. Extensive experiments on both balanced and imbalanced datasets show that our approaches significantly improve the performance of zero-shot translation over strong MNMT baselines.

Wenchao Gu, Yanlin Wang, Lun Du et al.

Code search is to search reusable code snippets from source code corpus based on natural languages queries. Deep learning-based methods of code search have shown promising results. However, previous methods focus on retrieval accuracy but lacked attention to the efficiency of the retrieval process. We propose a novel method CoSHC to accelerate code search with deep hashing and code classification, aiming to perform an efficient code search without sacrificing too much accuracy. To evaluate the effectiveness of CoSHC, we apply our method to five code search models. Extensive experimental results indicate that compared with previous code search baselines, CoSHC can save more than 90% of retrieval time meanwhile preserving at least 99% of retrieval accuracy.

Wenxuan Wang, Wenxiang Jiao, Jingyuan Huang et al. · pku, tencent-ai

This paper identifies a cultural dominance issue within large language models (LLMs) due to the predominant use of English data in model training (e.g., ChatGPT). LLMs often provide inappropriate English-culture-related answers that are not relevant to the expected culture when users ask in non-English languages. To systematically evaluate the cultural dominance issue, we build a benchmark of concrete (e.g., holidays and songs) and abstract (e.g., values and opinions) cultural objects. Empirical results show that the representative GPT models suffer from the culture dominance problem, where GPT-4 is the most affected while text-davinci-003 suffers the least from this problem. Our study emphasizes the need to critically examine cultural dominance and ethical consideration in their development and deployment. We show that two straightforward methods in model development (i.e., pretraining on more diverse data) and deployment (e.g., culture-aware prompting) can significantly mitigate the cultural dominance issue in LLMs.

Yanyang Li, Shuo Liang, Michael R. Lyu et al.

Recent advancements in long-context modeling have enhanced language models (LMs) for complex tasks across multiple NLP applications. Despite this progress, we find that these models struggle with multi-hop reasoning and exhibit decreased performance in the presence of noisy contexts. In this paper, we introduce Reasoning with Attributions, a novel approach that prompts LMs to supply attributions for each assertion during their reasoning. We validate our approach through experiments on three multi-hop datasets, employing both proprietary and open-source models, and demonstrate its efficacy and resilience. Furthermore, we explore methods to augment reasoning capabilities via fine-tuning and offer an attribution-annotated dataset and a specialized training strategy. Our fine-tuned model achieves competitive performance on multi-hop reasoning benchmarks, closely paralleling proprietary LMs such as ChatGPT and Claude-instant.

Jen-tse Huang, Jianping Zhang, Wenxuan Wang et al. · pku, tencent-ai

Due to the labor-intensive nature of manual test oracle construction, various automated testing techniques have been proposed to enhance the reliability of Natural Language Processing (NLP) software. In theory, these techniques mutate an existing test case (e.g., a sentence with its label) and assume the generated one preserves an equivalent or similar semantic meaning and thus, the same label. However, in practice, many of the generated test cases fail to preserve similar semantic meaning and are unnatural (e.g., grammar errors), which leads to a high false alarm rate and unnatural test cases. Our evaluation study finds that 44% of the test cases generated by the state-of-the-art (SOTA) approaches are false alarms. These test cases require extensive manual checking effort, and instead of improving NLP software, they can even degrade NLP software when utilized in model training. To address this problem, we propose AEON for Automatic Evaluation Of NLP test cases. For each generated test case, it outputs scores based on semantic similarity and language naturalness. We employ AEON to evaluate test cases generated by four popular testing techniques on five datasets across three typical NLP tasks. The results show that AEON aligns the best with human judgment. In particular, AEON achieves the best average precision in detecting semantic inconsistent test cases, outperforming the best baseline metric by 10%. In addition, AEON also has the highest average precision of finding unnatural test cases, surpassing the baselines by more than 15%. Moreover, model training with test cases prioritized by AEON leads to models that are more accurate and robust, demonstrating AEON's potential in improving NLP software.

Wenxuan Wang, Juluan Shi, Zixuan Ling et al. · pku, tencent-ai

Equipped with the capability to call functions, modern large language models (LLMs) can leverage external tools for addressing a range of tasks unattainable through language skills alone. However, the effective execution of these tools relies heavily not just on the advanced capabilities of LLMs but also on precise user instructions, which often cannot be ensured in the real world. To evaluate the performance of LLMs tool-use under imperfect instructions, we meticulously examine the real-world instructions queried from users, analyze the error patterns, and build a challenging tool-use benchmark called Noisy ToolBench (NoisyToolBench). We find that due to the next-token prediction training objective, LLMs tend to arbitrarily generate the missed argument, which may lead to hallucinations and risks. To address this issue, we propose a novel framework, Ask-when-Needed (AwN), which prompts LLMs to ask questions to users whenever they encounter obstacles due to unclear instructions. Moreover, to reduce the manual labor involved in user-LLM interaction and assess LLMs performance in tool utilization from both accuracy and efficiency perspectives, we design an automated evaluation tool named ToolEvaluator. Our experiments demonstrate that the AwN significantly outperforms existing frameworks for tool learning in the NoisyToolBench. We will release all related code and datasets to support future research.

Chaozheng Wang, Yuanhang Yang, Cuiyun Gao et al.

Pre-trained models have been shown effective in many code intelligence tasks. These models are pre-trained on large-scale unlabeled corpus and then fine-tuned in downstream tasks. However, as the inputs to pre-training and downstream tasks are in different forms, it is hard to fully explore the knowledge of pre-trained models. Besides, the performance of fine-tuning strongly relies on the amount of downstream data, while in practice, the scenarios with scarce data are common. Recent studies in the natural language processing (NLP) field show that prompt tuning, a new paradigm for tuning, alleviates the above issues and achieves promising results in various NLP tasks. In prompt tuning, the prompts inserted during tuning provide task-specific knowledge, which is especially beneficial for tasks with relatively scarce data. In this paper, we empirically evaluate the usage and effect of prompt tuning in code intelligence tasks. We conduct prompt tuning on popular pre-trained models CodeBERT and CodeT5 and experiment with three code intelligence tasks including defect prediction, code summarization, and code translation. Our experimental results show that prompt tuning consistently outperforms fine-tuning in all three tasks. In addition, prompt tuning shows great potential in low-resource scenarios, e.g., improving the BLEU scores of fine-tuning by more than 26\% on average for code summarization. Our results suggest that instead of fine-tuning, we could adapt prompt tuning for code intelligence tasks to achieve better performance, especially when lacking task-specific data.

Yanyang Li, Jianqiao Zhao, Duo Zheng et al.

With the continuous emergence of Chinese Large Language Models (LLMs), how to evaluate a model's capabilities has become an increasingly significant issue. The absence of a comprehensive Chinese benchmark that thoroughly assesses a model's performance, the unstandardized and incomparable prompting procedure, and the prevalent risk of contamination pose major challenges in the current evaluation of Chinese LLMs. We present CLEVA, a user-friendly platform crafted to holistically evaluate Chinese LLMs. Our platform employs a standardized workflow to assess LLMs' performance across various dimensions, regularly updating a competitive leaderboard. To alleviate contamination, CLEVA curates a significant proportion of new data and develops a sampling strategy that guarantees a unique subset for each leaderboard round. Empowered by an easy-to-use interface that requires just a few mouse clicks and a model API, users can conduct a thorough evaluation with minimal coding. Large-scale experiments featuring 23 Chinese LLMs have validated CLEVA's efficacy.

Wenxuan Wang, Jingyuan Huang, Jen-tse Huang et al. · pku, tencent-ai

The exponential growth of social media platforms has brought about a revolution in communication and content dissemination in human society. Nevertheless, these platforms are being increasingly misused to spread toxic content, including hate speech, malicious advertising, and pornography, leading to severe negative consequences such as harm to teenagers' mental health. Despite tremendous efforts in developing and deploying textual and image content moderation methods, malicious users can evade moderation by embedding texts into images, such as screenshots of the text, usually with some interference. We find that modern content moderation software's performance against such malicious inputs remains underexplored. In this work, we propose OASIS, a metamorphic testing framework for content moderation software. OASIS employs 21 transform rules summarized from our pilot study on 5,000 real-world toxic contents collected from 4 popular social media applications, including Twitter, Instagram, Sina Weibo, and Baidu Tieba. Given toxic textual contents, OASIS can generate image test cases, which preserve the toxicity yet are likely to bypass moderation. In the evaluation, we employ OASIS to test five commercial textual content moderation software from famous companies (i.e., Google Cloud, Microsoft Azure, Baidu Cloud, Alibaba Cloud and Tencent Cloud), as well as a state-of-the-art moderation research model. The results show that OASIS achieves up to 100% error finding rates. Moreover, through retraining the models with the test cases generated by OASIS, the robustness of the moderation model can be improved without performance degradation.

Yuxuan Wan, Tianqing Fang, Zaitang Li et al.

Recent advances in Deep Research Agents (DRAs) are transforming automated knowledge discovery and problem-solving. While the majority of existing efforts focus on enhancing policy capabilities via post-training, we propose an alternative paradigm: self-evolving the agent's ability by iteratively verifying the policy model's outputs, guided by meticulously crafted rubrics. This approach gives rise to the inference-time scaling of verification, wherein an agent self-improves by evaluating its generated answers to produce iterative feedback and refinements. We derive the rubrics based on an automatically constructed DRA Failure Taxonomy, which systematically classifies agent failures into five major categories and thirteen sub-categories. We present DeepVerifier, a rubrics-based outcome reward verifier that leverages the asymmetry of verification and outperforms vanilla agent-as-judge and LLM judge baselines by 12%-48% in meta-evaluation F1 score. To enable practical self-evolution, DeepVerifier integrates as a plug-and-play module during test-time inference. The verifier produces detailed rubric-based feedback, which is fed back to the agent for iterative bootstrapping, refining responses without additional training. This test-time scaling delivers 8%-11% accuracy gains on challenging subsets of GAIA and XBench-DeepResearch when powered by capable closed-source LLMs. Finally, to support open-source advancement, we release DeepVerifier-4K, a curated supervised fine-tuning dataset of 4,646 high-quality agent steps focused on DRA verification. These examples emphasize reflection and self-critique, enabling open models to develop robust verification capabilities.

Jia Li, Zeyang Zhuang, Zhuangbin Chen et al.

Compilation errors pose pervasive and critical challenges in software development, significantly hindering productivity. Therefore, Automated Compilation Error Repair (ACER) techniques are proposed to mitigate these issues. Despite recent advancements in ACER, its real-world performance remains poorly evaluated. This can be largely attributed to the limitations of existing benchmarks, \ie decontextualized single-file data, lack of authentic source diversity, and biased local task modeling that ignores crucial repository-level complexities. To bridge this critical gap, we propose ComBench, the first repository-level, reproducible real-world benchmark for C/C++ compilation error repair. ComBench is constructed through a novel, automated framework that systematically mines real-world failures from the GitHub CI histories of large-scale open-source projects. Our framework contributes techniques for the high-precision identification of ground-truth repair patches from complex version histories and a high-fidelity mechanism for reproducing the original, ephemeral build environments. To ensure data quality, all samples in ComBench are execution-verified -- guaranteeing reproducible failures and build success with ground-truth patches. Using ComBench, we conduct a comprehensive evaluation of 12 modern LLMs under both direct and agent-based repair settings. Our experiments reveal a significant gap between a model's ability to achieve syntactic correctness (a 73% success rate for GPT-5) and its ability to ensure semantic correctness (only 41% of its patches are valid). We also find that different models exhibit distinct specializations for different error types. ComBench provides a robust and realistic platform to guide the future development of ACER techniques capable of addressing the complexities of modern software development.

Jianping Zhang, Yizhan Huang, Weibin Wu et al.

Vision transformers (ViTs) have been successfully deployed in a variety of computer vision tasks, but they are still vulnerable to adversarial samples. Transfer-based attacks use a local model to generate adversarial samples and directly transfer them to attack a target black-box model. The high efficiency of transfer-based attacks makes it a severe security threat to ViT-based applications. Therefore, it is vital to design effective transfer-based attacks to identify the deficiencies of ViTs beforehand in security-sensitive scenarios. Existing efforts generally focus on regularizing the input gradients to stabilize the updated direction of adversarial samples. However, the variance of the back-propagated gradients in intermediate blocks of ViTs may still be large, which may make the generated adversarial samples focus on some model-specific features and get stuck in poor local optima. To overcome the shortcomings of existing approaches, we propose the Token Gradient Regularization (TGR) method. According to the structural characteristics of ViTs, TGR reduces the variance of the back-propagated gradient in each internal block of ViTs in a token-wise manner and utilizes the regularized gradient to generate adversarial samples. Extensive experiments on attacking both ViTs and CNNs confirm the superiority of our approach. Notably, compared to the state-of-the-art transfer-based attacks, our TGR offers a performance improvement of 8.8% on average.

Guangba Yu, Zirui Wang, Yujie Huang et al.

The democratization of open-source Large Language Models (LLMs) allows users to fine-tune and deploy models on local infrastructure but exposes them to a First Mile deployment landscape. Unlike black-box API consumption, the reliability of user-managed orchestration remains a critical blind spot. To bridge this gap, we conduct the first large-scale empirical study of 705 real-world failures from the open-source DeepSeek, Llama, and Qwen ecosystems. Our analysis reveals a paradigm shift: white-box orchestration relocates the reliability bottleneck from model algorithmic defects to the systemic fragility of the deployment stack. We identify three key phenomena: (1) Diagnostic Divergence: runtime crashes distinctively signal infrastructure friction, whereas incorrect functionality serves as a signature for internal tokenizer defects. (2) Systemic Homogeneity: Root causes converge across divergent series, confirming reliability barriers are inherent to the shared ecosystem rather than specific architectures. (3) Lifecycle Escalation: Barriers escalate from intrinsic configuration struggles during fine-tuning to compounded environmental incompatibilities during inference. Supported by our publicly available dataset, these insights provide actionable guidance for enhancing the reliability of the LLM landscape.

Yanyang Li, Jianqiao Zhao, Michael R. Lyu et al.

Recent advances in large-scale pre-training provide large models with the potential to learn knowledge from the raw text. It is thus natural to ask whether it is possible to leverage these large models as knowledge bases for downstream tasks. In this work, we answer the aforementioned question in unsupervised knowledge-grounded conversation. We explore various methods that best elicit knowledge from large models. Our human study indicates that, though hallucinations exist, large models post the unique advantage of being able to output common sense and summarize facts that cannot be directly retrieved from the search engine. To better exploit such generated knowledge in dialogue generation, we treat the generated knowledge as a noisy knowledge source and propose the posterior-based reweighing as well as the noisy training strategy. Empirical results on two benchmarks show advantages over the state-of-the-art methods.

Junjie Huang, Daya Guo, Chenglong Wang et al.

Data wrangling, the process of preparing raw data for further analysis in computational notebooks, is a crucial yet time-consuming step in data science. Code generation has the potential to automate the data wrangling process to reduce analysts' overhead by translating user intents into executable code. Precisely generating data wrangling code necessitates a comprehensive consideration of the rich context present in notebooks, including textual context, code context and data context. However, notebooks often interleave multiple non-linear analysis tasks into linear sequence of code blocks, where the contextual dependencies are not clearly reflected. Directly training models with source code blocks fails to fully exploit the contexts for accurate wrangling code generation. To bridge the gap, we aim to construct a high quality datasets with clear and rich contexts to help training models for data wrangling code generation tasks. In this work, we first propose an automated approach, CoCoMine to mine data-wrangling code generation examples with clear multi-modal contextual dependency. It first adopts data flow analysis to identify the code blocks containing data wrangling codes. Then, CoCoMine extracts the contextualized datawrangling code examples through tracing and replaying notebooks. With CoCoMine, we construct CoCoNote, a dataset containing 58,221 examples for Contextualized Data-wrangling Code generation in Notebooks. To demonstrate the effectiveness of our dataset, we finetune a range of pretrained code models and prompt various large language models on our task. Furthermore, we also propose DataCoder, which encodes data context and code&textual contexts separately to enhance code generation. Experiment results demonstrate the significance of incorporating data context in data-wrangling code generation and the effectiveness of our model. We release code and data at url...

Jinyang Liu, Tianyi Yang, Zhuangbin Chen et al.

As modern software systems continue to grow in terms of complexity and volume, anomaly detection on multivariate monitoring metrics, which profile systems' health status, becomes more and more critical and challenging. In particular, the dependency between different metrics and their historical patterns plays a critical role in pursuing prompt and accurate anomaly detection. Existing approaches fall short of industrial needs for being unable to capture such information efficiently. To fill this significant gap, in this paper, we propose CMAnomaly, an anomaly detection framework on multivariate monitoring metrics based on collaborative machine. The proposed collaborative machine is a mechanism to capture the pairwise interactions along with feature and temporal dimensions with linear time complexity. Cost-effective models can then be employed to leverage both the dependency between monitoring metrics and their historical patterns for anomaly detection. The proposed framework is extensively evaluated with both public data and industrial data collected from a large-scale online service system of Huawei Cloud. The experimental results demonstrate that compared with state-of-the-art baseline models, CMAnomaly achieves an average F1 score of 0.9494, outperforming baselines by 6.77% to 10.68%, and runs 10X to 20X faster. Furthermore, we also share our experience of deploying CMAnomaly in Huawei Cloud.

Jinyang Liu, Junjie Huang, Yintong Huo et al.

System logs play a critical role in maintaining the reliability of software systems. Fruitful studies have explored automatic log-based anomaly detection and achieved notable accuracy on benchmark datasets. However, when applied to large-scale cloud systems, these solutions face limitations due to high resource consumption and lack of adaptability to evolving logs. In this paper, we present an accurate, lightweight, and adaptive log-based anomaly detection framework, referred to as SeaLog. Our method introduces a Trie-based Detection Agent (TDA) that employs a lightweight, dynamically-growing trie structure for real-time anomaly detection. To enhance TDA's accuracy in response to evolving log data, we enable it to receive feedback from experts. Interestingly, our findings suggest that contemporary large language models, such as ChatGPT, can provide feedback with a level of consistency comparable to human experts, which can potentially reduce manual verification efforts. We extensively evaluate SeaLog on two public datasets and an industrial dataset. The results show that SeaLog outperforms all baseline methods in terms of effectiveness, runs 2X to 10X faster and only consumes 5% to 41% of the memory resource.

Shuqing Li, Binchang Li, Yepang Liu et al.

In recent years, spatial computing a.k.a. Extended Reality (XR) has emerged as a transformative technology, offering users immersive and interactive experiences across diversified virtual environments. Users can interact with XR apps through interactable GUI elements (IGEs) on the stereoscopic three-dimensional (3D) graphical user interface (GUI). The accurate recognition of these IGEs is instrumental, serving as the foundation of many software engineering tasks, including automated testing and effective GUI search. The most recent IGE detection approaches for 2D mobile apps typically train a supervised object detection model based on a large-scale manually-labeled GUI dataset, usually with a pre-defined set of clickable GUI element categories like buttons and spinners. Such approaches can hardly be applied to IGE detection in XR apps, due to a multitude of challenges including complexities posed by open-vocabulary and heterogeneous IGE categories, intricacies of context-sensitive interactability, and the necessities of precise spatial perception and visual-semantic alignment for accurate IGE detection results. Thus, it is necessary to embark on the IGE research tailored to XR apps. In this paper, we propose the first zero-shot cOntext-sensitive inteRactable GUI ElemeNT dEtection framework for virtual Reality apps, named Orienter. By imitating human behaviors, Orienter observes and understands the semantic contexts of XR app scenes first, before performing the detection. The detection process is iterated within a feedback-directed validation and reflection loop. Specifically, Orienter contains three components, including (1) Semantic context comprehension, (2) Reflection-directed IGE candidate detection, and (3) Context-sensitive interactability classification. Extensive experiments demonstrate that Orienter is more effective than the state-of-the-art GUI element detection approaches.

Renyi Zhong, Yichen Li, Yulun Wu et al.

Logging statements are central to debugging, failure diagnosis, and production observability, yet writing them requires developers to decide where to place a logging statement, which API and severity level to use, and what runtime information to expose. Automated logging aims to reduce this burden, but existing evidence remains dominated by Java-centric repository-snapshot dataset. It is therefore unclear whether conclusions about model behavior and model selection generalize across programming-language ecosystems or realistic code evolution. This paper presents MultiLogBench, a multilingual benchmark and empirical study spanning six programming language ecosystems. MultiLogBench contains 63,965 production-code repository-snapshot instances, 744 revision-history cases where developers introduce logging statements during maintenance, and a paired transformed revision-history branch for robustness analysis. Using seven contemporary large language models under a unified protocol, we evaluate logging-site localization, framework-anchor matching, severity prediction, message generation, variable recovery, and cascaded overall quality. Results show clear cross-language variation: framework-anchor matching is the most language-sensitive component, loop and nested-callable sites are the hardest structural contexts, and model rankings are stable only at the top tier. These patterns persist at a coarse level on revision-history data, while transformed inputs do not cause a broad same-direction performance collapse. Overall, MultiLogBench shows that robust claims about automated logging require multilingual evaluation and maintenance-oriented validation.

Jen-tse Huang, Eric John Li, Man Ho Lam et al. · pku, tencent-ai

Decision-making is a complex process requiring diverse abilities, making it an excellent framework for evaluating Large Language Models (LLMs). Researchers have examined LLMs' decision-making through the lens of Game Theory. However, existing evaluation mainly focus on two-player scenarios where an LLM competes against another. Additionally, previous benchmarks suffer from test set leakage due to their static design. We introduce GAMA($γ$)-Bench, a new framework for evaluating LLMs' Gaming Ability in Multi-Agent environments. It includes eight classical game theory scenarios and a dynamic scoring scheme specially designed to quantitatively assess LLMs' performance. $γ$-Bench allows flexible game settings and adapts the scoring system to different game parameters, enabling comprehensive evaluation of robustness, generalizability, and strategies for improvement. Our results indicate that GPT-3.5 demonstrates strong robustness but limited generalizability, which can be enhanced using methods like Chain-of-Thought. We also evaluate 13 LLMs from 6 model families, including GPT-3.5, GPT-4, Gemini, LLaMA-3.1, Mixtral, and Qwen-2. Gemini-1.5-Pro outperforms others, scoring of $69.8$ out of $100$, followed by LLaMA-3.1-70B ($65.9$) and Mixtral-8x22B ($62.4$). Our code and experimental results are publicly available at https://github.com/CUHK-ARISE/GAMABench.

Jia Li, Yuxin Su, Michael R. Lyu

As large language models (LLMs) evolve into autonomous agents, evaluating repository-level reasoning, the ability to maintain logical consistency across massive, real-world, interdependent file systems, has become critical. Current benchmarks typically fluctuate between isolated code snippets and black-box evaluations. We present RepoReason, a white-box diagnostic benchmark centered on abductive assertion verification. To eliminate memorization while preserving authentic logical depth, we implement an execution-driven mutation framework that utilizes the environment as a semantic oracle to regenerate ground-truth states. Furthermore, we establish a fine-grained diagnostic system using dynamic program slicing, quantifying reasoning via three orthogonal metrics: $ESV$ (reading load), $MCL$ (simulation depth), and $DFI$ (integration width). Comprehensive evaluations of frontier models (e.g., Claude-4.5-Sonnet, DeepSeek-v3.1-Terminus) reveal a prevalent aggregation deficit, where integration width serves as the primary cognitive bottleneck. Our findings provide granular white-box insights for optimizing the next generation of agentic software engineering.

Man Tik Ng, Hui Tung Tse, Jen-tse Huang et al. · pku, tencent-ai

The role-play ability of Large Language Models (LLMs) has emerged as a popular research direction. However, existing studies focus on imitating well-known public figures or fictional characters, overlooking the potential for simulating ordinary individuals. Such an oversight limits the potential for advancements in digital human clones and non-player characters in video games. To bridge this gap, we introduce ECHO, an evaluative framework inspired by the Turing test. This framework engages the acquaintances of the target individuals to distinguish between human and machine-generated responses. Notably, our framework focuses on emulating average individuals rather than historical or fictional figures, presenting a unique advantage to apply the Turing Test. We evaluated three role-playing LLMs using ECHO, with GPT-3.5 and GPT-4 serving as foundational models, alongside the online application GPTs from OpenAI. Our results demonstrate that GPT-4 more effectively deceives human evaluators, and GPTs achieves a leading success rate of 48.3%. Furthermore, we investigated whether LLMs could discern between human-generated and machine-generated texts. While GPT-4 can identify differences, it could not determine which texts were human-produced. Our code and results of reproducing the role-playing LLMs are made publicly available via https://github.com/CUHK-ARISE/ECHO.

Yuxuan Wan, Wenxuan Wang, Yiliu Yang et al. · pku, tencent-ai

We introduce LogicAsker, a novel approach for evaluating and enhancing the logical reasoning capabilities of large language models (LLMs) such as ChatGPT and GPT-4. Despite LLMs' prowess in tasks like writing assistance, code generation, and machine translation, assessing their ability to reason has been challenging. Traditional evaluations often prioritize accuracy on downstream tasks over direct assessments of reasoning processes. LogicAsker addresses this gap by employing a set of atomic reasoning skills grounded in propositional and predicate logic to systematically examine and improve the reasoning prowess of LLMs. Our methodology reveals significant gaps in LLMs' learning of logical rules, with identified reasoning failures ranging from 29\% to 90\% across different models. Moreover, we leverage these findings to construct targeted demonstration examples and fine-tune data, notably enhancing logical reasoning in models like GPT-4o by up to 5\%. To our knowledge, this is the first effort to utilize test case outcomes to effectively refine LLMs' formal reasoning capabilities. We make our code, data, and results publicly available (https://github.com/yxwan123/LogicAsker) to facilitate further research and replication of our findings.

Meifang Chen, Zhe Yang, Huang Nianchen et al.

Code secrets are sensitive assets for software developers, and their leakage poses significant cybersecurity risks. While the rapid development of AI code assistants powered by Code Large Language Models (CLLMs), CLLMs are shown to inadvertently leak such secrets due to a notorious memorization phenomenon. This study first reveals that Byte-Pair Encoding (BPE) tokenization leads to unexpected behavior of secret memorization, which we term as \textit{gibberish bias}. Specifically, we identified that some secrets are among the easiest for CLLMs to memorize. These secrets yield high character-level entropy, but low token-level entropy. Then, this paper supports the biased claim with numerical data. We identified that the roots of the bias are the token distribution shift between the CLLM training data and the secret data. We further discuss how gibberish bias manifests under the ``larger vocabulary'' trend. To conclude the paper, we discuss potential mitigation strategies and the broader implications on current tokenizer design.

Jialun Cao, Yuk-Kit Chan, Zixuan Ling et al.

Various benchmarks have been proposed to assess the performance of large language models (LLMs) in different coding scenarios. We refer to them as code-related benchmarks. However, there are no systematic guidelines by which such a benchmark should be developed to ensure its quality, reliability, and reproducibility. We propose How2Bench, which is comprised of a 55-criteria checklist as a set of guidelines to govern the development of code-related benchmarks comprehensively. Using HOW2BENCH, we profiled 274 benchmarks released within the past decade and found concerning issues. Nearly 70% of the benchmarks did not take measures for data quality assurance; over 10% did not even open source or only partially open source. Many highly cited benchmarks have loopholes, including duplicated samples, incorrect reference codes/tests/prompts, and unremoved sensitive/confidential information. Finally, we conducted a human study involving 49 participants, which revealed significant gaps in awareness of the importance of data quality, reproducibility, and transparency.

Yilei Jiang, Yaozhi Zheng, Yuxuan Wan et al.

Automating the transformation of user interface (UI) designs into front-end code holds significant promise for accelerating software development and democratizing design workflows. While multimodal large language models (MLLMs) can translate images to code, they often fail on complex UIs, struggling to unify visual perception, layout planning, and code synthesis within a single monolithic model, which leads to frequent perception and planning errors. To address this, we propose ScreenCoder, a modular multi-agent framework that decomposes the task into three interpretable stages: grounding, planning, and generation. By assigning these distinct responsibilities to specialized agents, our framework achieves significantly higher robustness and fidelity than end-to-end approaches. Furthermore, ScreenCoder serves as a scalable data engine, enabling us to generate high-quality image-code pairs. We use this data to fine-tune open-source MLLM via a dual-stage pipeline of supervised fine-tuning and reinforcement learning, demonstrating substantial gains in its UI generation capabilities. Extensive experiments demonstrate that our approach achieves state-of-the-art performance in layout accuracy, structural coherence, and code correctness. Our code is made publicly available at https://github.com/leigest519/ScreenCoder.

Wenxin Tang, Jingyu Xiao, Wenxuan Jiang et al.

Manual slide creation is labor-intensive and requires expert prior knowledge. Existing natural language-based LLM generation methods struggle to capture the visual and structural nuances of slide designs. To address this, we formalize the Reference Image to Slide Generation task and propose Slide2Code, the first benchmark with difficulty-tiered samples based on a novel Slide Complexity Metric. We introduce SlideCoder, a layout-aware, retrieval-augmented framework for generating editable slides from reference images. SlideCoder integrates a Color Gradient-based Segmentation algorithm and a Hierarchical Retrieval-Augmented Generation method to decompose complex tasks and enhance code generation. We also release SlideMaster, a 7B open-source model fine-tuned with improved reverse-engineered data. Experiments show that SlideCoder outperforms state-of-the-art baselines by up to 40.5 points, demonstrating strong performance across layout fidelity, execution accuracy, and visual consistency. Our code is available at https://github.com/vinsontang1/SlideCoder.

Yiwu Zhong, Zi-Yuan Hu, Michael R. Lyu et al.

Visual representation learning has been a cornerstone in computer vision, involving typical forms such as visual embeddings, structural symbols, and text-based representations. Despite the success of CLIP-type visual embeddings, they often lack access to world knowledge critical for visual reasoning. In this work, we propose Visual Table, a novel form of visual representation tailored for visual reasoning. Visual tables are constructed as hierarchical descriptions of visual scenes, featuring a scene description and multiple object-centric descriptions covering categories, attributes, and knowledge. Thanks to the structural and textual formats, visual tables offer unique advantages over mere visual embeddings, such as interpretability and controllable editing. Furthermore, they deliver instance-level world knowledge and detailed attributes that are essential for visual reasoning. To create visual tables, we develop a generator trained on the dataset with collected, small-scale annotations. Extensive results on 11 visual reasoning benchmarks demonstrate that the generated visual tables significantly outperform previous structural and text-based representations. Moreover, they consistently enhance state-of-the-art multimodal large language models across diverse benchmarks, showcasing their potential for advancing visual reasoning tasks. Our code is available at https://github.com/LaVi-Lab/Visual-Table.

Chaozheng Wang, Zezhou Yang, Shuzheng Gao et al.

Code editing constitutes a fundamental practice in software development, wherein developers modify existing codebases according to natural language requirements. Accurate code editing necessitates a comprehensive understanding of both the existing codebase and the modification requirements. Although large language models (LLMs) have demonstrated promising performance in code editing tasks, they suffer from substantial inefficiency by generating entire modified files that largely consist of unchanged code. While smaller models could potentially address this inefficiency, they typically lack the capacity to effectively comprehend long code contexts required for accurate editing. To ensure both effectiveness and efficiency, we propose to decompose code editing into a two-stage cascade: \textbf{edit sketch generation}, wherein a large model first produces concise sketches representing the requisite modifications (the more challenging phase), and \textbf{edit sketch application}, wherein a smaller model integrates these sketches into the original code to produce the final output edited code (the simpler phase). This cascaded design reduces the number of tokens generated by the large model, as the majority of the output is handled by the smaller, more efficient model, thereby enhancing overall efficiency. However, the effectiveness of this approach is constrained by current small models' limited capabilities in handling long-context scenarios and cross-file dependencies, which are essential for accurate sketch application in real-world codebases. To address these limitations and enhance smaller models' sketch application capabilities, ...

Jingyu Xiao, Zhongyi Zhang, Yuxuan Wan et al.

Multimodal Large Language Models have demonstrated exceptional performance in UI2Code tasks, significantly enhancing website development efficiency. However, these tasks incur substantially higher computational overhead than traditional code generation due to the large number of input image tokens and extensive output code tokens required. Our comprehensive study identifies significant redundancies in both image and code tokens that exacerbate computational complexity and hinder focus on key UI elements, resulting in excessively lengthy and often invalid HTML files. We propose EfficientUICoder, a compression framework for efficient UI code generation with three key components. First, Element and Layout-aware Token Compression preserves essential UI information by detecting element regions and constructing UI element trees. Second, Region-aware Token Refinement leverages attention scores to discard low-attention tokens from selected regions while integrating high-attention tokens from unselected regions. Third, Adaptive Duplicate Token Suppression dynamically reduces repetitive generation by tracking HTML/CSS structure frequencies and applying exponential penalties. Extensive experiments show EfficientUICoderachieves a 55%-60% compression ratio without compromising webpage quality and delivers superior efficiency improvements: reducing computational cost by 44.9%, generated tokens by 41.4%, prefill time by 46.6%, and inference time by 48.8% on 34B-level MLLMs. Code is available at https://github.com/WebPAI/EfficientUICoder.

Yanyang Li, Tin Long Wong, Cheung To Hung et al.

Recent advances in large language models (LLMs) have shown significant promise, yet their evaluation raises concerns, particularly regarding data contamination due to the lack of access to proprietary training data. To address this issue, we present C$^2$LEVA, a comprehensive bilingual benchmark featuring systematic contamination prevention. C$^2$LEVA firstly offers a holistic evaluation encompassing 22 tasks, each targeting a specific application or ability of LLMs, and secondly a trustworthy assessment due to our contamination-free tasks, ensured by a systematic contamination prevention strategy that fully automates test data renewal and enforces data protection during benchmark data release. Our large-scale evaluation of 15 open-source and proprietary models demonstrates the effectiveness of C$^2$LEVA.

Jen-tse Huang, Yuhang Yan, Linqi Liu et al. · pku, tencent-ai

Recent failures such as Google Gemini generating people of color in Nazi-era uniforms illustrate how AI outputs can be factually plausible yet socially harmful. AI models are increasingly evaluated for "fairness," yet existing benchmarks often conflate two fundamentally different dimensions: factual correctness and normative fairness. A model may generate responses that are factually accurate but socially unfair, or conversely, appear fair while distorting factual reality. We argue that identifying the boundary between fact and fair is essential for meaningful fairness evaluation. We introduce Fact-or-Fair, a benchmark with (i) objective queries aligned with descriptive, fact-based judgments, and (ii) subjective queries aligned with normative, fairness-based judgments. Our queries are constructed from 19 statistics and are grounded in cognitive psychology, drawing on representativeness bias, attribution bias, and ingroup-outgroup bias to explain why models often misalign fact and fairness. Experiments across ten frontier models reveal different levels of fact-fair trade-offs. By reframing fairness evaluation, we provide both a new theoretical lens and a practical benchmark to advance the responsible model assessments. Our test suite is publicly available at https://github.com/uclanlp/Fact-or-Fair.

Man Ho Lam, Chaozheng Wang, Hange Liu et al.

Coding agents powered by large language models are increasingly expected to perform realistic software maintenance tasks beyond isolated issue resolution. Existing benchmarks have shifted toward realistic software evolution, but they rarely capture continuous maintenance at the granularity of package releases, where changes are bundled, shipped, and inherited by subsequent versions. We present SWE-Chain, a benchmark for evaluating agents on chained release-level package upgrades, where each transition builds on the agent's prior codebase. To produce upgrade specifications, we design a divide-and-conquer synthesis pipeline that aligns release notes with code diffs for each version transition, ensuring the requirements are grounded in actual code changes, informative to agents, and feasible to implement. SWE-Chain contains 12 upgrade chains across 9 real Python packages, with 155 version transitions and 1,660 grounded upgrade requirements. Across nine frontier agent-model configurations, agents achieve an average of 44.8% resolving, 65.4% precision, and 50.2% F1 under the Build+Fix regime, with Claude-Opus-4.7 (Claude Code) leading at 60.8% resolving, 80.6% precision, and 68.5% F1. These results show that SWE-Chain is both feasible and discriminative, and reveal that current agents still struggle to make correct upgrades across chained package releases without breaking existing functionality.

Ruofan Lu, Yintong Huo, Meng Zhang et al.

The rapid advancement of large language models (LLMs) has led to the widespread adoption of AI-powered coding assistants integrated into a development environment. On one hand, low-latency code completion offers completion suggestions but is fundamentally constrained to the cursor's current position. On the other hand, chat-based editing can perform complex modifications, yet forces developers to stop their work, describe the intent in natural language, which causes a context-switch away from the code. This creates a suboptimal user experience, as neither paradigm proactively predicts the developer's next edit in a sequence of related edits. To bridge this gap and provide the seamless code edit suggestion, we introduce the task of Next Edit Prediction, a novel task designed to infer developer intent from recent interaction history to predict both the location and content of the subsequent edit. Specifically, we curate a high-quality supervised fine-tuning dataset and an evaluation benchmark for the Next Edit Prediction task. Then, we conduct supervised fine-tuning on a series of models and performed a comprehensive evaluation of both the fine-tuned models and other baseline models, yielding several novel findings. This work lays the foundation for a new interaction paradigm that proactively collaborate with developers by anticipating their following action, rather than merely reacting to explicit instructions. The code is available at https://github.com/lurf21/NextEditPrediction.

Runxin Yang, Yuxuan Wan, Shuqing Li et al.

Developing 3D games requires specialized expertise across multiple domains, including programming, 3D modeling, and engine configuration, which limits access to millions of potential creators. Recently, researchers have begun to explore automated game development. However, existing approaches face three primary challenges: (1) limited scope to 2D content generation or isolated code snippets; (2) requirement for manual integration of generated components into game engines; and (3) poor performance on handling interactive game logic and state management. While Multimodal Large Language Models (MLLMs) demonstrate potential capabilities to ease the game generation task, a critical gap still remains in translating these outputs into production-ready, executable game projects based on game engines such as Unity and Unreal Engine. To bridge the gap, this paper introduces UniGen, the first end-to-end coordinated multi-agent framework that automates zero-coding development of runnable 3D games from natural language requirements. Specifically, UniGen uses a Planning Agent that interprets user requirements into structured blueprints and engineered logic descriptions; after which a Generation Agent produces executable C# scripts; then an Automation Agent handles engine-specific component binding and scene construction; and lastly a Debugging Agent provides real-time error correction through conversational interaction. We evaluated UniGen on three distinct game prototypes. Results demonstrate that UniGen not only democratizes game creation by requiring no coding from the user, but also reduces development time by 91.4%. We release UniGen at https://github.com/yxwan123/UniGen. A video demonstration is available at https://www.youtube.com/watch?v=xyJjFfnxUx0.

Tianyi Yang, Jiacheng Shen, Yuxin Su et al.

Service reliability is one of the key challenges that cloud providers have to deal with. In cloud systems, unplanned service failures may cause severe cascading impacts on their dependent services, deteriorating customer satisfaction. Predicting the cascading impacts accurately and efficiently is critical to the operation and maintenance of cloud systems. Existing approaches identify whether one service depends on another via distributed tracing but no prior work focused on discriminating to what extent the dependency between cloud services is. In this paper, we survey the outages and the procedure for failure diagnosis in two cloud providers to motivate the definition of the intensity of dependency. We define the intensity of dependency between two services as how much the status of the callee service influences the caller service. Then we propose AID, the first approach to predict the intensity of dependencies between cloud services. AID first generates a set of candidate dependency pairs from the spans. AID then represents the status of each cloud service with a multivariate time series aggregated from the spans. With the representation of services, AID calculates the similarities between the statuses of the caller and the callee of each candidate pair. Finally, AID aggregates the similarities to produce a unified value as the intensity of the dependency. We evaluate AID on the data collected from an open-source microservice benchmark and a cloud system in production. The experimental results show that AID can efficiently and accurately predict the intensity of dependencies. We further demonstrate the usefulness of our method in a large-scale commercial cloud system.

Wenchao Gu, Zongjie Li, Cuiyun Gao et al.

Code retrieval is a common practice for programmers to reuse existing code snippets in open-source repositories. Given a user query (i.e., a natural language description), code retrieval aims at searching for the most relevant ones from a set of code snippets. The main challenge of effective code retrieval lies in mitigating the semantic gap between natural language descriptions and code snippets. With the ever-increasing amount of available open-source code, recent studies resort to neural networks to learn the semantic matching relationships between the two sources. The statement-level dependency information, which highlights the dependency relations among the program statements during the execution, reflects the structural importance of one statement in the code, which is favorable for accurately capturing the code semantics but has never been explored for the code retrieval task. In this paper, we propose CRaDLe, a novel approach for Code Retrieval based on statement-level semantic Dependency Learning. Specifically, CRaDLe distills code representations through fusing both the dependency and semantic information at the statement level and then learns a unified vector representation for each code and description pair for modeling the matching relationship. Comprehensive experiments and analysis on real-world datasets show that the proposed approach can accurately retrieve code snippets for a given query and significantly outperform the state-of-the-art approaches to the task.

Yifan Gao, Chien-Sheng Wu, Jingjing Li et al.

Document interpretation and dialog understanding are the two major challenges for conversational machine reading. In this work, we propose Discern, a discourse-aware entailment reasoning network to strengthen the connection and enhance the understanding for both document and dialog. Specifically, we split the document into clause-like elementary discourse units (EDU) using a pre-trained discourse segmentation model, and we train our model in a weakly-supervised manner to predict whether each EDU is entailed by the user feedback in a conversation. Based on the learned EDU and entailment representations, we either reply to the user our final decision "yes/no/irrelevant" of the initial question, or generate a follow-up question to inquiry more information. Our experiments on the ShARC benchmark (blind, held-out test set) show that Discern achieves state-of-the-art results of 78.3% macro-averaged accuracy on decision making and 64.0 BLEU1 on follow-up question generation. Code and models are released at https://github.com/Yifan-Gao/Discern.