As code large language models (LLMs) evolve into tool-interactive agents via the Model Context Protocol (MCP), their generalization is increasingly limited by low-quality synthetic data and the diminishing returns of quantity scaling; moreover, quantity-centric scaling exhibits an early bottleneck that underutilizes trajectory data. We propose TDScaling, a Trajectory Diversity Scaling-based data synthesis framework for code agents that scales performance through diversity rather than raw volume. Moreover, TDScaling is more data-efficient: under a fixed training budget, increasing trajectory diversity yields larger gains than adding more trajectories, improving the performance-cost trade-off for agent training. TDScaling integrates four innovations: (1) a Business Cluster mechanism that captures real-service logical dependencies; (2) a Blueprint-driven multi-agent paradigm that enforces trajectory coherence; (3) an adaptive evolution mechanism that steers synthesis toward long-tail scenarios using Domain Entropy, Reasoning Mode Entropy, and Cumulative Action Complexity to prevent mode collapse; and (4) a sandboxed code tool that mitigates catastrophic forgetting of intrinsic coding capabilities. Experiments on general tool-use benchmarks (BFCL, 𝜏²-Bench) and code agent tasks (RebenchT, CodeCI, BIRD) demonstrate a win-win outcome: TDScaling improves both tool-use generalization and inherent coding proficiency. Crucially, we show that trajectory diversity scaling attains a substantially higher performance ceiling than quantity scaling, establishing a resource-efficient paradigm for training robust code agents under data bottlenecks.

Reward models (RMs) play a critical role in aligning large language models (LLMs) with human preferences. Yet in the domain of tool learning, the lack of RMs specifically designed for function-calling tasks has limited progress toward more capable agentic AI. We introduce ToolRM, a family of lightweight reward models tailored for general tool-use scenarios. To build these models, we propose a novel pipeline that constructs high-quality pairwise preference data using rule-based scoring and multidimensional sampling. This yields ToolPref-Pairwise-30K, a diverse, balanced, and challenging preference dataset that supports both generative and discriminative reward modeling. We also introduce TRBench_BFCL, a benchmark built on the agent evaluation suite BFCL to evaluate RMs on tool calling tasks. Trained on our constructed data, models from the Qwen3-4B/8B series achieve up to 17.94% higher accuracy, substantially outperforming frontier LLMs and RMs in pairwise reward judgments. Beyond training objectives, generative ToolRM generalizes to broader critique tasks, including Best-of-N sampling and self-correction. Experiments on ACEBench highlight its effectiveness and efficiency, enabling inference-time scaling while reducing output token usage by over 66%. Its support for downstream RL training further validates its practical utility. We release data to facilitate future research.

Evaluating code large language models (Code LLMs) requires reliable detection of data leakage, where benchmark performance is artificially inflated by exposure to benchmark data during pre-training. Existing approaches either assume access to proprietary training corpora, rely on brittle heuristics such as timestamp filtering, or use external reference sets with manually tuned, non-generalizable thresholds. To address these limitations, we introduce SrDetection, a unified self-referential leakage detection framework for both gray-box (access to model logits) and black-box (access to model outputs) settings. SrDetection generates semantically equivalent variants of a benchmark sample and detects leakage by contrasting the model’s behavior on the original versus its variants, flagging cases where the original is disproportionately easier for the model. We further design a controlled leakage detection testbed and evaluate SrDetection in this environment. Across different models and training stages, SrDetection improves average F1 by 21.52 points in the gray-box setting and 14.46 points in the black-box setting over strong baselines, demonstrating robust, threshold-independent leakage detection. Finally, a gray-box study of 15 widely used Code LLMs on four popular benchmarks reveals benchmark-specific leakage patterns beyond prior overlap-based analyses[Source code and data are available at https://github.com/SMinL/SrDetectionCode].

Large Reasoning Models (LRMs) have demonstrated strong performance by producing extended Chain-of-Thought (CoT) traces before answering. However, this paradigm often induces over-reasoning: redundant calculations and circular self-verification that increase computational cost without improving outcomes. Existing evaluations largely emphasize final accuracy or coarse token counts, and lack automated tools to separate essential logic from structural redundancy. We introduce CoTJudger, a graph-driven framework that quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path (SEP) needed to reach a correct solution. This yields an interpretable efficiency signal – how much of a CoT is necessary versus structurally redundant – that is comparable across models and tasks. Evaluating 21 LRMs, CoTJudger reveals pervasive redundancy and surfaces recurring failure modes, including verification obsession and compensatory redundancy. These results provide a practical metric for disentangling reasoning ability from computational waste, enabling more targeted evaluation and diagnosis of LRM efficiency.

The accurate extraction of scientific measurements from literature is a critical yet challenging task in AI4Science, enabling large-scale analysis and integration of quantitative research findings. However, Large Language Models (LLMs) frequently exhibit severe hallucinations, which significantly undermine the reliability of automated scientific document understanding systems. To address this problem, we propose MeasHalu, a novel framework for mitigating scientific measurement hallucinations through enhanced reasoning and targeted optimization. We first present a fine-grained taxonomy of measurement-specific hallucinations, categorizing errors across quantities, units, modifiers, and relations. Our approach incorporates a two-stage reasoning-aware fine-tuning strategy using augmented scientific data and process-based supervision. Furthermore, we introduce a progressive reward curriculum designed to penalize specific hallucination types, significantly improving extraction faithfulness. Experimental results demonstrate that MeasHalu substantially reduces hallucination rates and improves overall accuracy on the MeasEval benchmark. This work provides a targeted solution to a key bottleneck in automated scientific knowledge extraction, facilitating more trustworthy and scalable machine-assisted scientific literature analysis.

Intellectual Property (IP) is a highly specialized domain that integrates technical and legal knowledge, making it inherently complex and knowledge-intensive. Recent advancements in LLMs have demonstrated their potential to handle IP tasks, enabling more efficient analysis, understanding, and generation of IP-related content. However, existing datasets and benchmarks focus narrowly on patents or cover limited aspects of the IP field, lacking alignment with real-world scenarios. To bridge this gap, we introduce IPBench, the first comprehensive IP task taxonomy and a large-scale bilingual benchmark encompassing 8 IP mechanisms and 20 distinct tasks, designed to evaluate LLMs in real-world IP practice. We benchmark 19 main LLMs, ranging from general purpose to domain-specific, including chat-oriented and reasoning-focused models, under zero-shot, few-shot, and chain-of-thought settings. Our results show that even the top-performing model, DeepSeek-V3, achieves only 75.8% accuracy, indicating significant room for improvement. Notably, open-source IP and law-oriented models lag behind closed-source general-purpose models. To foster future research, we publicly release IPBench, and will expand it with additional tasks to better reflect real-world complexities and support model advancements in the IP domain. We provide the data, code in the supplementary materials.

Currently, most reinforcement learning tasks focus on domains like mathematics and programming, where verification is relatively straightforward. However, in subjective tasks such as role-playing, alignment techniques struggle to make progress, primarily because subjective reward modeling using the Bradley-Terry model faces significant challenges when dealing with ambiguous preferences. To improve reward modeling in subjective tasks, this paper proposes AAM (Act-Adaptive Margin), which enhances reward modeling by dynamically calibrating preference margins using the model’s internal parameter knowledge. We design two versions of AAM that efficiently generate contextually-appropriate preference gaps without additional human annotation. This approach fundamentally improves how reward models handle subjective rewards by better integrating generative understanding with preference scoring. To validate AAM’s effectiveness in subjective reward modeling, we conduct evaluations on RewardBench, JudgeBench, and challenging role-playing tasks. Results show that AAM significantly improves subjective reward modeling performance, enhancing Bradley-Terry reward models by 2.95% in general tasks and 4.85% in subjective role-playing tasks. Furthermore, reward models trained with AAM can help downstream alignment tasks achieve better results. Our test results show that applying rewards generated by AAM-Augmented RM to preference learning techniques (e.g., GRPO) achieves state-of-the-art results on CharacterEval and Charm. The code and dataset will be released upon acceptance.

Tool-Integrated Reasoning (TIR) has emerged as a promising direction by extending Large Language Models’ (LLMs) capabilities with external tools during reasoning. Existing TIR methods typically rely on external tool documentation during reasoning. However, this leads to tool mastery difficulty, tool size constraints, and inference inefficiency. To mitigate these issues, we explore Tool-Internalized Reasoning (TInR), aiming at facilitating reasoning with tool knowledge internalized into LLMs. Achieving this goal presents notable requirements, including tool internalization and tool-reasoning coordination. To address them, we propose TInR-U, a tool-internalized reasoning framework for unified reasoning and tool usage. TInR-U is trained through a three-phase pipeline: 1) tool internalization with a bidirectional knowledge alignment strategy; 2) supervised fine-tuning warm-up using high-quality reasoning annotations, and 3) reinforcement learning with TInR-specific rewards. We comprehensively evaluate our method across in-domain and out-of-domain settings. Experiment results show that TInR-U achieves superior performance in both settings, highlighting its effectiveness and efficiency. The codes are attached in the supplementary file for review.

Reasoning-oriented Large Language Models (LLMs) have achieved remarkable progress with Chain-of-Thought (CoT) prompting, yet they remain fundamentally limited by a blind self-thinking paradigm: performing extensive internal reasoning even when critical information is missing or ambiguous. We propose Proactive Interactive Reasoning (PIR), a new reasoning paradigm that transforms LLMs from passive solvers into proactive inquirers that interleave reasoning with clarification. Unlike existing search- or tool-based frameworks that primarily address knowledge uncertainty by querying external environments, PIR targets premise- and intent-level uncertainty through direct interaction with the user. PIR is implemented via two core components: (1) an uncertainty-aware supervised fine-tuning procedure that equips models with interactive reasoning capability, and (2) a user-simulator-based policy optimization framework driven by a composite reward that aligns model behavior with user intent. Extensive experiments on mathematical reasoning, code generation, and document editing demonstrate that PIR consistently outperforms strong baselines, achieving up to 32.70% higher accuracy, 22.90% higher pass rate, and 41.36 BLEU improvement, while reducing nearly half of the reasoning computation and unnecessary interaction turns. Further reliability evaluations on factual knowledge, question answering, and missing-premise scenarios confirm the strong generalization and robustness of PIR.

Remarkable progress on large language models (LLMs), particularly in English, has facilitated impressive capabilities in following human instructions. However, there remains a noticeable gap in instruction fine-tuning for Chinese, where the complex linguistic features pose significant challenges. Existing datasets, generally distilled from English-centric LLMs, are not well-aligned with Chinese users’ interaction patterns. To bridge this gap, we introduce COIG-CQIA, a new Chinese instruction tuning dataset derived from various real-world data resources and undergoing comprehensive human verification. We conduct extensive experiments on COIG-CQIA, and compare them with strong baseline models and datasets. The experimental results show that models trained on COIG-CQIA achieve highly competitive performance in diverse benchmarks. Additionally, our findings offer several insights for designing effective Chinese instruction-tuning datasets and data mixing strategies. Our dataset are available at https://huggingface.co/datasets/m-a-p/COIG-CQIA.

The advancement of Large Language Models (LLMs) enables flexible and interpretable automatic evaluations. In the field of machine translation evaluation, utilizing LLMs with translation error annotations based on Multidimensional Quality Metrics (MQM) yields more human-aligned judgments. However, current LLM-based evaluation methods still face challenges in accurately identifying error spans and assessing their severity. In this paper, we propose HiMATE, a Hierarchical Multi-Agent Framework for Machine Translation Evaluation. We argue that existing approaches inadequately exploit the fine-grained structural and semantic information within the MQM hierarchy. To address this, we develop a hierarchical multi-agent system grounded in the MQM error typology, enabling granular evaluation of subtype errors. Two key strategies are incorporated to further mitigate systemic hallucinations within the framework: the utilization of the model’s self-reflective capability and the facilitation of agent discussion involving asymmetric information. Empirically, HiMATE outperforms competitive baselines across different datasets in conducting human-aligned evaluations. Further analyses underscore its significant advantage in error span detection and severity assessment, achieving an average F1-score improvement of 89% over the best-performing baseline. We make our code and data publicly available at https://github.com/nlp2ct-shijie/HiMATE.

Recent studies have highlighted the remarkable knowledge retention capabilities of Large Language Models (LLMs) like GPT-4, while simultaneously revealing critical limitations in maintaining knowledge currency and accuracy. Existing knowledge editing methodologies, designed to update specific factual information without compromising general model performance, often encounter two fundamental challenges: parameter conflict during knowledge overwriting and excessive computational overhead. In this paper, we introduce ForGet (Forget for Get), a novel approach grounded in the principle of “forgetting before learning”. By pinpointing the location within the LLM that corresponds to the target knowledge, we first erase the outdated knowledge and then insert the new knowledge at this precise spot. ForGet is the first work to leverage a two-phase gradient-based process for knowledge editing, offering a lightweight solution that also delivers superior results. Experimental findings show that our method achieves more effective knowledge editing at a lower cost compared to previous techniques across various base models.

As large language models (LLMs) are increasingly applied across various domains, enhancing safety while maintaining the helpfulness of LLMs has become a critical challenge. Recent studies solve this problem through safety-constrained online preference optimization or safety-constrained offline preference optimization. However, the safety-constrained online methods often suffer from excessive safety, which might reduce helpfulness, while the safety-constrained offline methods perform poorly in adaptively balancing safety and helpfulness. To address these limitations, we propose MidPO, a Mixture of Experts (MoE) framework for safety-helpfulness dual Preference Optimization. Firstly, MidPO devises single-preference enhanced direct preference optimization approach to transform the base model into two independent experts, termed safety and helpfulness experts, and fine-tunes the two independent experts for optimal safety or helpfulness performance. Secondly, to achieve an effective balance between safety and helpfulness, MidPO incorporates the two experts into the MoE framework and designs a dynamic routing mechanism to allocate contributions from each expert adaptively. We conduct quantitative and qualitative experiments on three popular datasets to demonstrate the proposed MidPO significantly outperforms state-of-the-art approaches in both safety and helpfulness. Code is available at https://github.com/OutdoorManofML/MidPO.

Multimodal Large Language Models (MLLMs) are measured on numerous benchmarks like image captioning, visual question answer, and reasoning. However, these benchmarks often include overly simple or uninformative samples, making it difficult to effectively distinguish the performance of different MLLMs. Additionally, evaluating models across many benchmarks creates a significant computational burden. To address these issues, we propose LIME (Less Is More for MLLM Evaluation), a refined and efficient benchmark curated using a semi-automated pipeline. This pipeline filters out uninformative samples and eliminates answer leakage by focusing on tasks that require image-based understanding. Our experiments show that LIME reduces the number of samples by 76% and evaluation time by 77%, while it can more effectively distinguish different models’ abilities. Notably, we find that traditional automatic metrics like CIDEr are insufficient for evaluating MLLMs’ captioning performance, and excluding the caption task score yields a more accurate reflection of overall model performance. All code and data are available at https://anonymous.4open.science/r/LIME-49CD

Current research in LLM-based simulation systems lacks comprehensive solutions for modeling real-world court proceedings, while existing legal language models struggle with dynamic courtroom interactions. We present AgentCourt, a comprehensive legal simulation framework that addresses these challenges through adversarial evolution of LLM-based agents. Our AgentCourt introduces a new adversarial evolutionary approach for agents called AdvEvol, which performs dynamic knowledge learning and evolution through structured adversarial interactions in a simulated courtroom program, breaking the limitations of the traditional reliance on static knowledge bases or manual annotations. By simulating 1,000 civil cases, we construct an evolving knowledge base that enhances the agents’ legal reasoning abilities. The evolved lawyer agents demonstrated outstanding performance on our newly introduced CourtBench benchmark, achieving a 12.1% improvement in performance compared to the original lawyer agents. Evaluations by professional lawyers confirm the effectiveness of our approach across three critical dimensions: cognitive agility, professional knowledge, and logical rigor. Beyond outperforming specialized legal models in interactive reasoning tasks, our findings emphasize the importance of adversarial learning in legal AI and suggest promising directions for extending simulation-based legal reasoning to broader judicial and regulatory contexts.

Tool learning has emerged as a promising direction by extending Large Language Models’ (LLMs) capabilities with external tools. Existing tool learning studies primarily focus on the general-purpose tool-use capability, which addresses explicit user requirements in instructions. However, they overlook the importance of personalized tool-use capability, leading to an inability to handle implicit user preferences. To address the limitation, we first formulate the task of personalized tool learning, which integrates user’s interaction history towards personalized tool usage. To fill the gap of missing benchmarks, we construct PEToolBench, featuring diverse user preferences reflected in interaction history under three distinct personalized settings, and encompassing a wide range of tool-use scenarios. Moreover, we propose a framework PEToolLLaMA to adapt LLMs to the personalized tool learning task, which is trained through supervised fine-tuning and direct preference optimization. Extensive experiments on PEToolBench demonstrate the superiority of PEToolLLaMA over existing LLMs. We release code and data at https://github.com/travis-xu/PEToolBench.

Stories are central to human culture, serving to share ideas, preserve traditions, and foster connections. Automatic story generation, a key advancement in artificial intelligence (AI), offers new possibilities for creating personalized content, exploring creative ideas, and enhancing interactive experiences. However, existing methods struggle to maintain narrative coherence and logical consistency. This disconnect compromises the overall storytelling experience, underscoring the need for substantial improvements. Inspired by human cognitive processes, we introduce Storyteller, a novel approach that systemically improves the coherence and consistency of automatically generated stories. Storyteller introduces a plot node structure based on linguistically grounded subject-verb-object (SVO) triplets, which capture essential story events and ensure a consistent logical flow. Unlike previous methods, Storyteller integrates two dynamic modules—the STORYLINE and narrative entity knowledge graph (NEKG)—that continuously interact with the story generation process. This integration produces structurally sound, cohesive and immersive narratives. Extensive experiments demonstrate that Storyteller significantly outperforms existing approaches, achieving an 84.33% average win rate through human preference evaluation. At the same time, it is also far ahead in other aspects including creativity, coherence, engagement, and relevance.

Large Language Model (LLM)-based agents have excelled in various domains but face significant challenges when applied to data science workflows due to their complex, multi-stage nature. Current LLM-based agents struggle with non-linear relationships, recursive dependencies, implicit data- and logic-dependent reasoning, and managing extensive context. In this paper, we introduce Data Interpreter, an LLM-based agent that addresses these challenges through hierarchical graph-based modeling to represent the complexity and a progressive strategy for step-by-step verification, refinement, and consistent context management. Extensive experiments confirm the effectiveness of Data Interpreter. On InfiAgent-DABench, it boosts performance by 25% (from 75.9% to 94.9%), and on machine learning and open-ended tasks, it lifts accuracy from 88% to 95% and from 60% to 97%, respectively. Moreover, our method surpasses state-of-the-art baselines by 26% on the MATH dataset. We will release the code upon publication.

The development of Multimodal Large Language Models (MLLMs) has seen significant progress, driven by increasing demands across various fields (e.g., multimodal agents, embodied intelligence). While model-driven approaches aim to enhance MLLM capabilities through diverse architectures, their performance gains have become increasingly marginal. In contrast, data-driven methods, which scale up image-text instruction datasets, have proven more effective but face challenges related to limited data diversity and complexity. The absence of high-quality instruction data remains a major bottleneck in MLLM development. To address this issue, we propose , a novel multimodal instruction data evolution framework. This framework iteratively enhances data quality through a refined combination of fine-grained perception, cognitive reasoning, and interaction evolution, generating a more complex and diverse image-text instruction dataset that significantly improves MLLM capabilities. Starting with an initial dataset, SEED-163K, we employ to systematically expand instruction diversity, extend visual reasoning steps to improve cognitive abilities, and extract fine-grained visual details to enhance understanding and robustness. To rigorously evaluate our approach, we conduct extensive qualitative analysis and quantitative experiments across 13 vision-language tasks. Compared to baseline models trained on the original seed dataset, our method achieves an average accuracy improvement of 3.1 percentage points. Moreover, our approach attains state-of-the-art (SOTA) performance in nine tasks while using significantly less data than existing state-of-the-art models.

Large Language Models (LLMs) demonstrate remarkable capabilities in text understanding and generation. However, their tendency to produce factually inconsistent outputs—commonly referred to as “hallucinations”—remains a critical challenge. Existing approaches, such as retrieval-based and inference-time correction methods, primarily address this issue at the input or output level, often overlooking the intrinsic information refinement process and the role of premature layers. Meanwhile, alignment- and fine-tuning-based methods are resource-intensive. In this paper, we propose PLI (Premature Layers Interpolation), a novel, training-free, and plug-and-play intervention designed to enhance factuality. PLI mitigates hallucinations by inserting premature layers formed through mathematical interpolation with adjacent layers. Inspired by stable diffusion and sampling steps, PLI extends the depth of information processing and transmission in LLMs, improving factual coherence. Experiments on four publicly available datasets demonstrate that PLI effectively reduces hallucinations while outperforming existing baselines in most cases. Further analysis suggests that the success of layer interpolation is closely linked to LLMs’ internal mechanisms. To promote reproducibility, we will release our code and data upon acceptance.

Deploying large language models (LLMs) for long-context inference remains challenging due to their substantial memory and computational demands. While techniques such as Key-Value (KV) cache compression are designed to reduce memory usage, they often neglect the structured sparsity inherent in the relationship between hidden states and their corresponding KV cache. In this work, we explore the role of uncertainty as a potential indicator of sparsity within LLMs. We propose UNComp, an uncertainty-aware framework that leverages truncated matrix entropy to identify areas of low information content, thereby revealing sparsity patterns that can be used for adaptive compression. Unlike traditional methods that apply uniform compression, UNComp dynamically adjusts its approach to compression, guided by uncertainty measures that reflect the importance of various model components. Our analysis shows that sparsity patterns, when derived from uncertainty estimates, can be exploited to reveal special long-range dependencies, such as retrieval heads and retrieval layers. This perspective not only enhances our understanding of how compression can be optimized but also provides new insights into the inherent sparsity of LLMs during long-context inference. By focusing on uncertainty to analyze the sparsity pattern in detail, UNComp reduces the KV cache size to 4.74% of the original, achieves a 6% prefill speedup, and improves throughput by 6.4× — not only delivering strong lossless compression performance, but also validating the effectiveness of the underlying theoretical tool. Our codes are submitted with the paper.

With the different roles that AI is expected to play in human life, imbuing large language models (LLMs) with different personalities has attracted increasing research interest. While the “personification” enhances human experiences of interactivity and adaptability of LLMs, it gives rise to critical concerns about content safety, particularly regarding bias, sentiment, and toxicity of LLM generation. This study explores how assigning different personality traits to LLMs affects the toxicity and biases of their outputs. Leveraging the widely accepted HEXACO personality framework developed in social psychology, we design experimentally sound prompts to test three LLMs’ performance on three toxic and bias benchmarks. The findings demonstrate the sensitivity of all three models to HEXACO personality traits and, more importantly, a consistent variation in the biases, negative sentiment, and toxicity of their output. In particular, adjusting the levels of several personality traits can effectively reduce bias and toxicity in model performance, similar to humans’ correlations between personality traits and toxic behaviors. The findings highlight the additional need to examine content safety besides the efficiency of training or fine-tuning methods for LLM personification, they also suggest a potential for the adjustment of personalities to be a simple and low-cost method to conduct controlled text generation.

As the capabilities of Multimodal Large Language Models (MLLMs) improve, the need for higher-order evaluation of them is increasing. However, there is a lack of work evaluating MLLM for higher-order perception and understanding of Chinese visual content. To address this, we introduce the CII-Bench, which aims to assess MLLMs’ such capabilities for Chinese images. To ensure the authenticity of the Chinese context, images in CII-Bench are sourced from the Chinese Internet and manually reviewed, with corresponding answers also manually crafted. Additionally, CII-Bench incorporates images that represent Chinese traditional culture, such as famous Chinese traditional paintings, which can deeply reflect the model’s understanding of Chinese traditional culture. Through experiments on multiple MLLMs using CII-Bench, significant findings emerged. There is a large gap between MLLMs and humans in performance. The highest MLLM accuracy is 64.4%, while the human average is 78.2% and the peak is 81.0%. MLLMs perform poorly on traditional culture images, indicating limitations in understanding high-level semantics and lacking a deep knowledge base of Chinese traditional culture. Moreover, most models have higher accuracy when image emotion hints are added to the prompts. We believe CII-Bench will help MLLMs better understand Chinese semantics and specific images, and move forward the development of expert artificial general intelligence (AGI). Our project is publicly available at https://cii-bench.github.io.

Argumentation Mining (AM) aims to extract argumentative structures from texts by identifying argumentation components (ACs) and their argumentative relations (ARs). While previous works focus on representation learning to encode ACs and AC pairs, they fail to explicitly model the underlying reasoning patterns of AM, resulting in limited interpretability. This paper proposes a novel  ̲First- ̲Order  ̲Logic reasoning framework for  ̲AM (FOL-AM), designed to explicitly capture logical reasoning paths within argumentative texts. By interpreting multiple AM subtasks as a unified relation query task modeled using FOL rules, FOL-AM facilitates multi-hop relational reasoning and enhances interpretability. The framework supports two flexible implementations: a fine-tuned approach to leverage task-specific learning, and a prompt-based method utilizing large language models to harness their generalization capabilities. Extensive experiments on two AM benchmarks demonstrate that FOL-AM outperforms strong baselines while significantly improving explainability.

Model distillation is a fundamental technique in building large language models (LLMs), transferring knowledge from a teacher model to a student model. However, distillation can lead to model homogenization, reducing diversity among models and impairing their ability to robustly handle complex or novel tasks. These limitations underscore the need to systematically quantify the distillation process and its impact. In this work, we propose a framework to evaluate and quantify model distillation. Our method addresses two key aspects: (1) Identifying identity cognition contradictions to assess discrepancies in how models perceive and represent identity-related information, and (2) Analyzing multi-granularity response similarities across models to measure the extent of homogenization. Experimental results demonstrate two key insights: (1) Well-known closed-source and open-source LLMs usually exhibit high distillation degrees, except for Claude, Doubao, and Gemini. (2) Base LLMs show higher distillation degrees compared to aligned LLMs. By offering a systematic approach to improve the transparency of LLM data distillation, we call for LLMs with more independent development and more transparent technical reports to improve LLMs’ robustness and safety. The code and data are available at https://github.com/Aegis1863/LLMs-Distillation-Quantification.

Speculative decoding (SD) is a promising method for accelerating the decoding process of Large Language Models (LLMs). The efficiency of SD primarily hinges on the consistency between the draft model and the verify model. However, existing drafting approaches typically require additional modules to be trained, which can be challenging to implement and ensure compatibility across various LLMs. In this paper, we propose CLaSp, an in-context layer-skipping strategy for self-speculative decoding. Unlike prior methods, CLaSp does not require additional drafting modules or extra training. Instead, it employs a plug-and-play mechanism by skipping intermediate layers of the verify model to construct a compressed draft model. Specifically, we develop a dynamic programming algorithm that optimizes the layer-skipping process by leveraging the complete hidden states from the last verification stage as an objective. This enables CLaSp to dynamically adjust its layer-skipping strategy after each verification stage, without relying on pre-optimized sets of skipped layers. Experimental results across diverse downstream tasks demonstrate that CLaSp achieves a speedup of 1.3× ∼ 1.7× on LLaMA3 series models without altering the original distribution of the generated text.

Role-Playing Agents (RPAs), benefiting from large language models, is an emerging interactive AI system that simulates roles or characters with diverse personalities. However, existing methods primarily focus on mimicking dialogues among roles in textual form, neglecting the role’s voice traits (e.g., voice style and emotions) as playing a crucial effect in interaction, which tends to be more immersive experiences in realistic scenarios. Towards this goal, we propose OmniCharacter, a first seamless speech-language personality interaction model to achieve immersive RPAs with low latency. Specifically, OmniCharacter enables agents to consistently exhibit role-specific personality traits and vocal traits throughout the interaction, enabling a mixture of speech and language responses. To align the model with speech-language scenarios, we construct a dataset named OmniCharacter-10K, which involves more distinctive characters (20), richly contextualized multi-round dialogue (10K), and dynamic speech response (135K). Experimental results showcase that our method yields better responses in terms of both content and style compared to existing RPAs and mainstream speech-language models, with a response latency as low as 289ms.

Multimodal Emotion Recognition in Conversations (ERC) aims to identify emotions in conversational videos. Current efforts focus on modeling both context-sensitive and speaker-sensitive dependencies and multimodal fusion. Despite the progress, models in Multimodal ERC (MERC) still struggle due to a lack of CommonSense Knowledge (CSK). In contrast, models in textual ERC typically employ CSK to enhance emotion inference. However, in multimodal scenarios, relying solely on textual CSK while neglecting visual CSK may hinder the understanding of visual emotional cues. To address this, we introduce a novel approach called Multiple Knowledge Enhanced Interactive Graph Network (MKE-IGN) to integrate multiple knowledge, such as textual and visual CSK, into the edge representations, thereby facilitating the modeling of relations between utterances and different types of CSK. Furthermore, considering that irrelevant CSK might be retained as noise, MKE-IGN adaptively selects this CSK guided by the mood-congruent effect and refines it based on contexts. Experimental results show that MKE-IGN outperforms state-of-the-art methods on two popular datasets.

The instruction-following ability of large language models enables humans to interact with AI agents in a natural way. However, when required to generate responses of a specific length, large language models often struggle to meet users’ needs due to their inherent difficulty in accurately perceiving numerical constraints. To explore the ability of large language models to control the length of generated responses, we propose the Target Length Generation Task (TLG) and design two metrics, Precise Match (PM) and Flexible Match (FM) to evaluate the model’s performance in adhering to specified response lengths. Furthermore, we introduce a novel, model-agnostic approach called Ruler, which employs Meta Length Tokens (MLTs) to enhance the instruction-following ability of large language models under length-constrained instructions. Specifically, Ruler equips LLMs with the ability to generate responses of a specified length based on length constraints within the instructions. Moreover, Ruler can automatically generate appropriate MLT when length constraints are not explicitly provided, demonstrating excellent versatility and generalization. Comprehensive experiments show the effectiveness of Ruler across different LLMs on Target Length Generation Task, e.g., at All Level 27.97 average gain on PM, 29.57 average gain on FM. In addition, we conduct extensive ablation experiments to further substantiate the efficacy and generalization of Ruler. Our code and data is available on the internet.

Numeral systems and units of measurement are two conjoined topics in activities of human beings and have mutual effects with the languages expressing them. Currently, the evaluation of Large Language Models (LLMs) often involves mathematical reasoning, yet little attention is given to how minor changes in numbers or units can drastically alter the complexity of problems and the performance of LLMs. In this paper, we scrutinize existing LLMs on processing of numerals and units of measurement by constructing datasets with perturbations. We first anatomize the reasoning of math word problems to different sub-procedures like numeral conversions from language to numbers and measurement conversions based on units. Then we further annotate math word problems from ancient Chinese arithmetic works which are challenging in numerals and units of measurement. Experiments on perturbed datasets demonstrate that LLMs still encounter difficulties in handling numeral and measurement conversions.

Using large language models (LLMs) to assist psychological counseling is a significant but challenging task at present. Attempts have been made on improving empathetic conversations or acting as effective assistants in the treatment with LLMs. However, the existing datasets lack consulting knowledge, resulting in LLMs lacking professional consulting competence. Moreover, how to automatically evaluate multi-turn dialogues within the counseling process remains an understudied area. To bridge the gap, we propose CPsyCoun, a report-based multi-turn dialogue reconstruction and evaluation framework for Chinese psychological counseling. To fully exploit psychological counseling reports, a two-phase approach is devised to construct high-quality dialogues while a comprehensive evaluation benchmark is developed for the effective automatic evaluation of multi-turn psychological consultations. Competitive experimental results demonstrate the effectiveness of our proposed framework in psychological counseling. We open-source the datasets and model for future research.

Stance detection is a challenging task that aims to identify public opinion from social media platforms with respect to specific targets. Previous work on stance detection largely focused on pure texts. In this paper, we study multi-modal stance detection for tweets consisting of texts and images, which are prevalent in today’s fast-growing social media platforms where people often post multi-modal messages. To this end, we create five new multi-modal stance detection datasets of different domains based on Twitter, in which each example consists of a text and an image. In addition, we propose a simple yet effective Targeted Multi-modal Prompt Tuning framework (TMPT), where target information is leveraged to learn multi-modal stance features from textual and visual modalities. Experimental results on our five benchmark datasets show that the proposed TMPT achieves state-of-the-art performance in multi-modal stance detection.

Argumentative Essay Generation (AEG) is a challenging task in computational argumentation, where detailed logical reasoning and effective rhetorical skills are essential.Previous methods on argument generation typically involve planning prior to generation.However, the planning strategies in these methods overlook the exploration of the logical reasoning process.Inspired by argument structure-related theories, we propose an argumentative planning strategy for prompting large language models (LLMs) to generate high-quality essays.This strategy comprises two stages: (1) Sketch planning, which creates a rough outline of the essay, and (2) Dialectical planning, which refines the outline through critical self-reflection.Such a planning strategy enables LLMs to write argumentative essays that are more logical, diverse, and persuasive.Furthermore, due to the scarcity of existing AEG datasets, we construct three new datasets.These datasets are from two domains: exam essays and news editorials, covering both Chinese and English.Automatic and manual evaluation on four datasets show that our method can generate more dialectical and persuasive essays with higher diversity compared to several strong baselines.

End-to-end argumentation mining (AM) aims to extract the argumentation structure including argumentation components and their argumentation relations from text. Recent developments in end-to-end AM models have demonstrated significant progress by redefining the AM task as a sequence generation task, exhibiting simplicity and competitive performance. Nevertheless, these models overlook the integration of supplementary discourse structure information, a crucial factor for comprehending argumentation structures, resulting in suboptimal outcomes. In this study, we propose the DENIM framework, which generates discourse structure-aware prefixes for each layer of the generation model. These prefixes imbue the generation-based AM model with discourse structures, thereby augmenting the overall generation process. Moreover, we introduce a multi-task prompt coupled with a three-step decoding strategy, aiming to optimize the efficiency and effectiveness of argumentation structure decoding. Extensive experiments and analyses on two benchmark datasets show that DENIM achieves state-of-the-art performances on two AM benchmarks.

The rapid development of Large Language Models (LLMs) has led to their increasing utilization in Chinese K-12 education. Despite the growing integration of LLMs and education, the absence of a dedicated benchmark for evaluating LLMs within this domain presents a pressing concern. Consequently, there is an urgent need for a comprehensive natural language processing benchmark to precisely assess the capabilities of various LLMs in Chinese K-12 education. In response, we introduce E-EVAL, the first comprehensive evaluation benchmark specifically tailored for Chinese K-12 education. E-EVAL comprises 4,351 multiple-choice questions spanning primary, middle, and high school levels, covering a diverse array of subjects. Through meticulous evaluation, we find that Chinese-dominant models often outperform English-dominant ones, with many exceeding GPT 4.0. However, most struggle with complex subjects like mathematics. Additionally, our analysis indicates that most Chinese-dominant LLMs do not achieve higher scores at the primary school level compared to the middle school level, highlighting the nuanced relationship between proficiency in higher-order and lower-order knowledge domains. Furthermore, experimental results highlight the effectiveness of the Chain of Thought (CoT) technique in scientific subjects and Few-shot prompting in liberal arts. Through E-EVAL, we aim to conduct a rigorous analysis delineating the strengths and limitations of LLMs in educational applications, thereby contributing significantly to the advancement of Chinese K-12 education and LLMs.

Large language models (LLMs) have achieved promising results in sentiment analysis through the in-context learning (ICL) paradigm. However, their ability to distinguish subtle sentiments still remains a challenge. Inspired by the human ability to adjust understanding via feedback, this paper enhances ICL by incorporating prior predictions and feedback, aiming to rectify sentiment misinterpretation of LLMs. Specifically, the proposed framework consists of three steps: (1) acquiring prior predictions of LLMs, (2) devising predictive feedback based on correctness, and (3) leveraging a feedback-driven prompt to refine sentiment understanding. Experimental results across nine sentiment analysis datasets demonstrate the superiority of our framework over conventional ICL methods, with an average F1 improvement of 5.95%.

Long-context modeling capabilities of Large Language Models (LLMs) have garnered widespread attention, leading to the emergence of LLMs with ultra-context windows. Meanwhile, benchmarks for evaluating long-context language models are gradually catching up. However, existing benchmarks employ irrelevant noise texts to artificially extend the length of test cases, diverging from the real-world scenarios of long-context applications. To bridge this gap, we propose a novel long-context benchmark, Loong, aligning with realistic scenarios through extended multi-document question answering (QA). Unlike typical document QA, in Loong’s test cases, each document is relevant to the final answer, ignoring any document will lead to the failure of the answer. Furthermore, Loong introduces four types of tasks with a range of context lengths: Spotlight Locating, Comparison, Clustering, and Chain of Reasoning, to facilitate a more realistic and comprehensive evaluation of long-context understanding. Extensive experiments indicate that existing long-context language models still exhibit considerable potential for enhancement. Retrieval augmented generation (RAG) achieves poor performance, demonstrating that Loong can reliably assess the model’s long-context modeling capabilities.

In recent years, instruction fine-tuning (IFT) on large language models (LLMs) has garnered considerable attention to enhance model performance on unseen tasks. Attempts have been made on automatic construction and effective selection for IFT data. However, we posit that previous methods have not fully harnessed the potential of LLMs for enhancing data quality. The responses within IFT data could be further enhanced by leveraging the capabilities of LLMs themselves.In this paper, we propose CoEvol, an LLM-based multi-agent cooperation framework for the improvement of responses for instructions. To effectively refine the responses, we develop an iterative framework following a debate-advise-edit-judge paradigm. A two-stage multi-agent debate strategy is further devised to ensure the diversity and reliability of editing suggestions within the framework. Empirically, models equipped with CoEvol outperform competitive baselines evaluated by MT-Bench and AlpacaEval, demonstrating its effectiveness in enhancing instruction-following capabilities for LLMs.

Zero-shot stance detection (ZSSD) aims to detect the stance for an unseen target during the inference stage. In this paper, we propose a joint contrastive learning (JointCL) framework, which consists of stance contrastive learning and target-aware prototypical graph contrastive learning. Specifically, a stance contrastive learning strategy is employed to better generalize stance features for unseen targets. Further, we build a prototypical graph for each instance to learn the target-based representation, in which the prototypes are deployed as a bridge to share the graph structures between the known targets and the unseen ones. Then a novel target-aware prototypical graph contrastive learning strategy is devised to generalize the reasoning ability of target-based stance representations to the unseen targets. Extensive experiments on three benchmark datasets show that the proposed approach achieves state-of-the-art performance in the ZSSD task.

With the increasing popularity of posting multimodal messages online, many recent studies have been carried out utilizing both textual and visual information for multi-modal sarcasm detection. In this paper, we investigate multi-modal sarcasm detection from a novel perspective by constructing a cross-modal graph for each instance to explicitly draw the ironic relations between textual and visual modalities. Specifically, we first detect the objects paired with descriptions of the image modality, enabling the learning of important visual information. Then, the descriptions of the objects are served as a bridge to determine the importance of the association between the objects of image modality and the contextual words of text modality, so as to build a cross-modal graph for each multi-modal instance. Furthermore, we devise a cross-modal graph convolutional network to make sense of the incongruity relations between modalities for multi-modal sarcasm detection. Extensive experimental results and in-depth analysis show that our model achieves state-of-the-art performance in multi-modal sarcasm detection.